Selling a home with AI as a system: from paperwork to handing over the keys

Selling a flat is, for most people, the biggest transaction of their life — and almost everyone does it for the first time, untrained and with their emotions in the way. A real estate agency solves that gap for a commission; anyone who won't or can't pay it faces a mountain of tasks they've never done: gathering paperwork, writing a listing, photographing the flat, answering the same questions from thirty people, telling a serious buyer apart from a tourist or a scammer, and finally not making a mistake in a contract that can cost far more than the whole commission would have.

This guide shows that it's doable — not through courage, but through a system: all the facts about the flat live in one structured document, and everything else — the info sheet, the website, the ads, the replies to buyers — is simply derived from it. When the price or the closing date changes, you edit one place and let the change flow outward; you never have to go hunting for which of five listing versions still has the old number. It's the same mental move as the brand-as-a-system guide applies to company materials: inventory the source material, build one source of truth, derive the outputs, keep it editable, keep it secure. Here we apply it to one flat and one sale.

Two rules sit above the whole piece. First: AI drafts, the owner vouches — every sentence in the listing, every fact in the info sheet, you sign with your name and your liability for defects. Second: AI never replaces a lawyer where money and ownership are on the line — the reservation contract, the purchase contract, escrow, and registering the transfer at the land registry are a lawyer's job; AI helps you understand that work and get the most out of the lawyer's time. Exactly where that line runs gets its own table at the end of the guide. You can read it phase by phase — each one stands on its own and comes with copy-paste prompts, just fill in the brackets.

A typical scenario

Jana and Petr inherited a three-bedroom flat (in the Czech system, "3+1" — three rooms plus a kitchen) in a regional city from their grandmother. It's largely original from the 1970s: new windows and risers (the building went through a renovation), but the original bathroom core, a kitchen that remembers the Communist era, and a damp stain in the corner of the bedroom that their grandmother had fixed last year. Both of them live elsewhere, they don't need the flat, and the mortgage on their own homes is squeezing them — they want to sell, reasonably fast and without an unnecessary loss.

Their first move was the obvious one: they approached a real estate agency. The offer sounded professional, but the commission worked out to what Petr would earn in half a year — and the contract wanted exclusivity for months on end. A second agency offered an "instant buyout" at a price that made their eyes roll. They decided to try selling it themselves, with a lawyer for the contracts — with one problem: they'd never sold anything, let alone real estate. They don't know what documents exist, what's mandatory, how to write a listing, or what to say when a buyer asks about the repair fund.

The classic self-seller scenario looks like this: a hastily written listing with phone photos shot at dusk, three different versions of the text on three portals (one still has the old price), buyers scattered across phone notes, viewings booked at random, chaos after three weeks, and surrender into the arms of the first agency that calls. Jana and Petr do something different instead: they spend two evenings on a paperwork inventory and building a property document — one file that knows everything about the flat. From it they derive an info sheet, a simple website with a gallery and viewing slots, listing variants for the portals, and answers to buyer questions. They track buyers in a spreadsheet with statuses from inquiry to offer. They go into viewings with a rehearsed script and answers to the hard questions prepared ahead of time. A lawyer with escrow handles the contracts — and AI helps them ask the lawyer the right questions. The rest of the guide breaks down those two evenings and everything that follows, step by step.

Phase 1: paperwork inventory — what you're actually selling

Before you write the first sentence of a listing, you need to know exactly what you're selling — legally and technically. That's not a formality: every document from this phase gets reused at least twice later (in the info sheet, at a viewing, with the lawyer), and every gap in the paperwork comes back as an awkward moment in front of a buyer or a delay before signing. The good news: it's an evening's work, and AI turns it into a guided process instead of groping around blind.

What to gather and where to get it

The basic set of paperwork for selling a privately owned flat looks like this:

• Title deed (list vlastnictví, LV) — an extract from the land registry (katastr nemovitostí) stating who owns the flat, how it's defined, and whether it carries a mortgage lien, an easement, or a foreclosure. You can look it up online for free for a rough check; a certified extract costs a small fee at a CzechPoint office. For an inherited flat, confirm the inheritance is already recorded in the registry — as long as the deceased is still listed there, you have nothing to sell.
• Declaration of the owner and floor plan — the document that originally split the building into units, usually including a schematic of the flat with measurements. It's typically kept in the registry's collection of deeds or with the owners' association (SVJ). If a usable floor plan doesn't exist, measure the flat yourself and sketch it — even a hand-drawn sketch with dimensions has value; it can be redrawn later.
• Energy performance certificate (PENB) — covered in detail shortly, since it's where most of the myths live.
• Documents from the SVJ: the repair-fund contribution and service-charge amounts, the latest utility statement, the bylaws, and minutes from recent meetings. The minutes are a goldmine: they'll tell you whether insulation work, an elevator replacement, or a loan is coming — exactly what any informed buyer will ask about.
• Repair and inspection history: what's been done in the flat and building, when, and by whom — new windows, risers, wiring, that repaired damp stain. Even an incomplete list beats "I don't know" — "I don't know" at a viewing sounds like "I'm hiding something."
• Utility bills for the last period — buyers ask about real living costs, not table values. And for a flat without a PENB, you'll need three years of bills formally anyway.
• Title of acquisition — for Jana and Petr, a probate court decision; elsewhere, a purchase or gift contract. Both the lawyer and the buyer's bank will want to see it.

Scan any paper-only documents into PDF with your phone — the approach from the scan it with your phone tip works well here. Claude can read scans and photos of documents, handwriting included, so even a stack of your grandmother's old papers isn't a blocker.

Save everything into one folder (say, sell-flat) with subfolders documents, photos, bills — this folder is where the info sheet and website will later come from. And here's the first security rule, right away: the paperwork contains personal data, so work with it in a paid account with contractual data protection, not an anonymous free chat. The title deed and the title of acquisition contain national ID numbers — black them out for AI work, or upload those documents only when you genuinely need to.

I'm selling a flat myself: [three-bedroom, privately owned, panel-
building construction, regional city], which I [inherited / bought
in year...]. I've never sold real estate before.

Build me a checklist of paperwork to gather before the sale. For
each item, cover:
1. What I'll need it for (listing, viewing, lawyer, buyer's bank)
2. Where to get it (land registry, owners' association, my own
   files, utility providers)
3. How to tell it's in order — what to check on the document
4. What happens if I don't have it

Split the checklist into "needed before the first listing," "needed
before viewings," and "needed before signing contracts." Ask me
step by step what I already have, and keep a running status: done /
missing / ordered.

You'll get a checklist tailored to your type of property and — the main benefit — phased by when each item is actually needed: you don't have to have everything on day one, but you do need to know what catches up with you when. Watch item 2: if AI names a specific office or registry, verify its name and existence — models occasionally invent institutions that sound plausible. And treat the checklist as a draft to fill in, not a complete list: a housing cooperative flat, co-ownership, or a lien on the deed each add items of their own that you should describe in the prompt.

PENB, honestly: what's mandatory and what's a myth

The energy performance certificate is where self-sellers most often either panic for no reason or take an unnecessary risk. The facts: the energy rating belongs in the listing itself — anyone without a certificate has to state the worst rating, G, which needlessly tanks the listing (and a fine for omitting it is also on the table). For a unit in a building, the certificate covers the whole building: the building's owner, typically the SVJ, has it prepared, and it's valid for ten years — there's a fair chance one already exists, sitting in a binder with the chairperson. So the move is: request the certificate from the SVJ in writing; if the SVJ doesn't deliver it, the law allows a unit to substitute three years of energy utility bills instead. That written request isn't red tape — it's the condition without which the bill-based substitute doesn't count.

I'm selling a flat in a building managed by an SVJ and dealing with
the energy performance certificate (PENB). Prepare:
1. A short, polite written request to the SVJ board for the
   building's energy performance certificate for the purpose of
   selling the unit — factual, referencing the unit owner's
   obligation to present the certificate
2. A checklist of what to do under each possible response: the
   certificate exists / doesn't exist and the SVJ will have one
   prepared / the SVJ doesn't respond
3. A list of the bills I should gather in the meantime (energy
   utility bills for the last 3 years), so I have a fallback ready
4. How to correctly state the energy rating in the listing under
   each scenario

Wherever you're not sure of the current wording of the regulations,
say so explicitly — I'll verify at the source rather than rely on
your memory.

You'll get a ready-to-send request and a decision tree. That last paragraph of the prompt matters more broadly: treat AI as navigation, not authority, on anything with the force of a legal obligation — regulations change, and a model can be working from an outdated version. Verifying is quick (ministry websites, the SVJ, a lawyer), and the gap between "AI told me so" and "I checked it myself" is exactly the gap you're liable for when selling.

If the inventory turns up a problem

The inventory sometimes surfaces something unpleasant: a mortgage lien left over from the deceased, a life-estate easement, a mismatch between the registry's floor area and reality, an inheritance not yet recorded. That's not a reason to panic — it's the reason the inventory happens first, not a week before signing. Every one of those findings has a fix (releasing the lien, negotiating with whoever holds the easement, correcting the registry record), but all of them take time, and most need a lawyer. Finding them in phase 1 means handling them alongside the rest of the prep; finding them in phase 10 means your buyer walks away in the meantime. Log these things in the property document (next phase) too — an honest source of truth includes what doesn't suit you as well.

A housing cooperative flat: same system, different paperwork

A short detour, since every other reader asks about it: if the flat is a housing cooperative unit, you're not selling real estate but transferring a cooperative share — which changes the paperwork, not the system. Instead of a title deed, the key document is the cooperative's confirmation of membership and that the share carries no debts; instead of the owner's declaration, it's the cooperative's bylaws (a transfer usually can't be blocked by the cooperative, but the tenancy tied to the flat is governed by the bylaws — read them before promising anyone a timeline); and there's an extra item, the annuity — the building loan's unpaid balance attached to the share, which passes to the buyer and must be disclosed honestly, both in the listing and in the price. Buyers also can't get a standard mortgage secured against a cooperative share, which narrows the pool of buyers and belongs in the property document's timeline-and-process section. Everything else — the document as source of truth, the info sheet, the website, buyer tracking, viewings, a lawyer for the contracts (here, for the share transfer) — applies unchanged; you just write "cooperative share" instead of "private ownership" in the prompts from this guide, and AI adapts the checklists. All the more reason to apply the verification rule with a cooperative: confirm transfer conditions directly with the cooperative, not with a model.

Phase 2: the property document — one source of truth

Now the most important decision of the whole sale — and it looks unassuming: everything you know about the flat gets written into one structured document, and every output gets derived from it. The listing is an excerpt of the document. The info sheet is the document, nicely laid out. The website is the document with a gallery. A reply to a buyer's question is a quote from the document. When anything changes — the price, the handover date, a repair you make along the way — you change the document and let the change flow into the outputs; you never fix five places by hand while praying you didn't miss one.

This principle is the quiet hero of the whole guide, so it's worth explaining why it works. Without a source of truth, every piece of text gets created separately: a listing on Monday from memory, an info sheet on Wednesday copied from the listing, a reply to a buyer on Friday from recollection. Every re-write is a chance for an error, and every change means manually working through every copy. With a source of truth, the flow is one-directional: reality → document → outputs. And there's one more effect you'll only appreciate once you're in the thick of it: when the thirtieth buyer asks a question the document doesn't cover, you add the answer to the document — and from that moment on, the info sheet, the website, and every future conversation knows it too. The document learns as the sale goes on.

Document structure

Write the document as a plain text file (flat.md) — no special format, just clear sections. A structure that works well:

1. Identification and legal status: address, unit number, ownership, what's on the title deed (liens, easements — and their resolution status), title of acquisition.
2. Parameters: layout, floor area per the owner's declaration (with a note if it differs from reality), floor, elevator, cellar, balcony, orientation, energy rating and its source.
3. Condition and history: what's original, what's been renovated and when, known defects — honestly, including that repaired damp stain, with a date and the method of repair.
4. Fixtures and fittings: what stays in the flat, what you're taking, what's negotiable. The number one source of later disputes — write it down now.
5. Costs: repair-fund contribution, service-charge advances, real energy usage from the bills, planned SVJ projects from the minutes.
6. The building and surroundings: the building's condition (renovation, risers), neighbors in general, amenities, transport, parking — facts, not marketing adjectives.
7. Timeline and process: when the flat becomes vacant, when viewings happen, how fast you can sign, what you prefer for financing.
8. Price and its rationale: the current asking price, what it's based on (comparable listings, an appraisal), and your real bottom line — that one is known only to you and the document, and it never goes into any output.
9. Questions and answers: a living section — every question from a buyer the document couldn't answer gets added here, along with the answer.

Assembly: from a pile of paperwork to a structured document

Don't build the document from memory — have it distilled from the paperwork you gathered in phase 1, then fill in what the papers don't cover.

I'm uploading the paperwork for a flat I'm selling myself: [floor
plan, land registry extract with national ID numbers blacked out,
utility and energy bills, SVJ meeting minutes, list of repairs,
notes].

Build a structured property document in markdown from them, with
sections: identification and legal status, parameters, condition
and history, fixtures and fittings, costs, building and
surroundings, timeline and process, price and rationale, questions
and answers.

Rules:
- rely only on the uploaded paperwork; don't invent anything
- where the paperwork is missing a fact, write TO FILL IN with a
  question about exactly what you need from me
- where the paperwork disagrees (e.g. floor area in the registry vs.
  in the building manager's listing), flag the discrepancy and don't
  decide for me
- no evaluative adjectives ("beautiful," "sunny") — facts only;
  the selling language comes later, in the outputs

You'll get a skeleton document riddled with TO FILL IN markers — and that's correct: they're your evening's work list. Check the numbers especially (floor areas, service charges, usage): copying numbers off tables and scans is where AI most easily slips up, and a wrong floor area in the listing is the kind of thing a buyer brings up at the worst possible moment. Every number in the document should be traceable back to a source in the paperwork.

Probing for gaps: AI as the curious buyer

Paperwork answers the questions authorities ask. Buyers ask differently. The fastest way to find the gaps is to flip the roles — the same trick the brand-voice guide uses for probing.

Here's the property document [insert]. Act as a careful buyer
looking for a flat to live in: a family with one child, taking out a
mortgage, comparing this flat against three other listings.

Ask me every question you can't answer from the document that this
buyer would ask before or during a viewing. Focus on the
uncomfortable ones too: age of the wiring and the bathroom core,
sound insulation, neighbors, dampness, exactly what that repaired
stain was, SVJ debts, planned increases in service charges, why the
flat is being sold.

Ask one question at a time, and work my answers straight into the
questions-and-answers section. If I answer evasively, tell me so and
ask again more specifically.

This is the most valuable half hour of prep: you're answering questions calmly at your desk, not for the first time live in front of strangers in your living room. That last sentence in the prompt isn't decoration — an evasive answer AI catches now is an evasive answer you won't catch in yourself at a viewing. And if you don't actually know the answer to something (age of the wiring, SVJ debts), you've got a new item for the inventory while there's still time.

Stress test: does the document hold up on its own?

Before declaring the document finished, test whether it can handle the role you're giving it: being the sole source of answers.

Here's the property document [insert]. Answer the following ten
buyer questions using ONLY the document — no inference. Where the
document doesn't contain the answer, write CANNOT ANSWER and what's
missing from the document:

1. What will I realistically pay per month for housing, utilities
   included?
2. What exactly happened with that dampness, and should I worry
   about it?
3. Does the kitchen stay, appliances included?
4. What's the earliest I could move in?
5. Is the SVJ planning anything that would raise the repair fund?
6. Why are you selling?
7. What's the energy rating, and how do you know?
8. Is parking available near the building?
9. Is a cellar included, and how big is it?
10. Are you taking a mortgage, or cash only?

If nine out of ten pass, the document is ready. If half come back CANNOT ANSWER, it just saved you several awkward phone calls — fill in the gaps and run the test again. Add the finished document to a Project on claude.ai as persistent context: from then on every conversation has it on hand, and you don't have to attach it to every prompt. Just remember the rule from the editability point: when the document changes, update it in the Project too — there's only one source of truth.

Upkeep: the document lives alongside the sale

A source of truth only works if it's kept current — a stale document is worse than none, because you trust it. Build three habits. First, a date and version in the header: "version 4, Sep 12" — every output then carries a note of which version it came from (you'll see this with the listing variants in phase 7), so staleness is visible instead of sneaky. Second, a changelog at the end of the document: one line per change — what, when, why ("dropped the price after 3 weeks with no offers," "added an answer about parking"). It's not bureaucracy — a month from now, negotiating, that log is your memory and your bargaining record. Third, a list of live outputs right in the document: info sheet (version), website (address), listings (portal, date, source master version) — a checklist that every change flows through, so an old price never gets left standing anywhere. These three habits are the whole difference between "I have some notes written down somewhere" and a system you can trust in the week fifteen people are calling.

Phase 3: photos on your phone — good light, no deception

Photos decide whether anyone clicks the listing; the text decides whether they call. A professional photographer is a good investment, but today's phones are enough for a solid listing — if you know what to shoot and when. And since photos are also "claims about the flat," the same ethics apply to them as to the text: adjusting light and perspective, yes; hiding reality, no.

Prep and shot plan

The most important photo edit happens before you press the shutter: a tidy, depersonalized flat. Clear the fridge, towels, cables, family photos (privacy matters here too — the whole internet will see these photos). For a flat inherited from a grandmother, consider clearing out furniture that shrinks the rooms — an empty flat photographs better than a crowded one. Shoot in daylight, ideally when the sun is in the flat, at eye height, from the corners of rooms (a corner gives the space depth), phone held level — leaning verticals are the telltale sign of an amateur listing.

I'm getting ready to photograph a flat [three-bedroom, panel
building, east-west orientation] for a listing; shooting on a
phone. You have the property document in the Project.

Build me a shot plan:
1. A room-by-room prep checklist — what to tidy away, what to hide,
   what to leave as is (the flat is furnished from my grandmother's
   time; some furniture is going)
2. Shooting order and time of day based on which way each room faces
   the sun
3. A shot list: for each room, the main shot (angle, what should be
   in frame) plus detail shots worth taking (new windows, repaired
   risers) — and shots to skip
4. Shots outside the flat: the building, entrance, cellar, view,
   surroundings — what buyers actually care about here
5. A post-shoot check: what to review on the photos afterward
   (verticals, reflections in mirrors, forgotten personal items)

You'll get a concrete shot list that turns the shoot into a two-hour job instead of a whole day of repeated attempts. Underestimate item 5 at your own risk: a photographer's reflection in the bathroom mirror and documents with your address on the fridge are the internet's favorite catches, not yours.

Selecting and editing: what's fair

Shoot plenty, keep few — a listing wants one strong photo per room, not three similar ones. Claude can help with the selection: it can look at the photos and compare them.

I'm uploading [40] photos of the flat. Pick the best set for the
listing: the hero photo (the one that decides the click) and 10-14
more, in the order they should appear in the gallery — from the
strongest rooms down to details.

For each photo you pick:
- why this one over similar alternatives
- what to fix, if anything (exposure, straightening verticals,
  cropping)
- a short factual caption for the gallery (no superlatives)

List separately the photos to leave out of the listing, and why
(too dark, a confusing crop, personal items visible). And flag it
if any edit you're suggesting would change the substance — e.g.
hiding a defect the property document discloses.

One clear rule for editing that pays off legally too: you may fix what a viewing visitor would also see. Brightening a dark photo, straightening verticals, cropping — yes, that's genuinely what the flat looks like, the phone just didn't capture it well. Retouching out a crack, mold, or a stain — no, that's deception, and it surfaces at the very first viewing, undermining trust in everything else you say. A special case is virtual home staging — AI-rendered furniture in empty rooms. It's fine to use, but only clearly labeled ("visualization; the flat is handed over empty") and placed next to a real photo of the same room, never in place of it. An unlabeled visualization is exactly the kind of trick that leaves a buyer feeling cheated at the viewing — and an angry buyer doesn't buy.

Phase 4: listing copy that doesn't lie

Now the first derived output: the listing text. Writing it with AI takes minutes — and that's exactly why this phase is mostly about what shouldn't be in the text. Real estate listings have a well-worn vocabulary of euphemisms that buyers have long since learned to decode: "investment opportunity" means a flat you can't live in, "renovate to your own taste" means a wreck, "close to downtown" means far from downtown. AI knows those phrases from its training data better than anyone — and it'll happily decorate your listing with them if you don't stop it. But you're not an anonymous agency with a legal department: you're personally on the hook for the truth of the listing, and an undisclosed defect comes back after the sale as a demand for a discount, or worse, a request to void the contract. For real estate, liability for hidden defects stretches on for years after handover. Honesty here isn't a moral bonus — it's risk management.

The good news: an honest listing sells better than most self-sellers expect. A buyer who learns about the damp stain from the listing ("fixed at the root cause in [year], documentation available") reads it as resolved history. A buyer who discovers it at a viewing reads it as the tip of an iceberg — and mentally deducts far more from the price than the repair actually cost.

A first draft from the document

Write a listing for the flat from the property document in the
Project.

Structure: headline (factual, no exclamation points), opening
paragraph (3-4 sentences: what, where, who it suits), parameters as
bullets, condition and repair history (including the fixed
dampness — phrase it factually, with the repair date), living
costs, building and surroundings, timeline and how the sale works
(legal service by a lawyer, escrow), energy rating.

Rules:
- every claim must be backed by the document; don't add anything
- no real estate euphemisms ("investment opportunity," "renovate to
  your own taste") and no superlatives you can't back up
- phrase positives concretely: not "great location" but "bus stop 3
  minutes on foot, school around the corner" — only if the document
  supports it
- don't hide or bury defects at the end; state them as facts with
  their fix
- keep it under [2,000] characters so it fits on the portals

You'll get a usable draft. Read it, though, as a proposal from a capable but not entirely accountable colleague: AI is trained to sound persuasive and will occasionally embellish — "renovated" instead of "windows replaced," a "quiet location" it has no way of knowing from the document. Test every sentence against one question: "Would I sign this into a contract?" If not, cut it or back it up.

Red team: the listing against the document

Don't rely on your own read — after an evening of writing, you can't see the text clearly anymore. Have it checked against the source of truth, sentence by sentence.

Here's the final listing text [insert]. Compare it against the
property document in the Project like a strict auditor:

1. Match every claim in the listing to the place in the document
   that supports it. List unsupported claims separately — we either
   back them up or cut them.
2. List what the document contains that the listing omits, and for
   each item judge: is this omitting a material defect (must go in),
   or just a detail (can wait for the viewing)?
3. Flag phrasing that's technically true but steers toward a wrong
   conclusion — e.g. "new windows" in a flat where everything else is
   original, with no mention of the rest's condition.
4. Check the numbers against the document: floor area, floor number,
   service charges, energy rating.

Don't judge the writing style. I only care about accuracy and
completeness.

Item 3 is the most valuable one — the gap between lying and misleading is exactly where most bad listings operate. The output is a listing you know will survive a viewing and any later dispute. Save it into the property document as an "approved listing text" section — it'll be the basis for the portal variants in phase 7.

Phase 5: the info sheet — an editable brochure, not a one-off PDF

The info sheet is the property's sales brochure: a few pages with photos, the floor plan, parameters, costs, and the sale process. Self-sellers usually don't have one — which is exactly why it works. A serious buyer, after a viewing, doesn't just get a verbal impression but a document they can open at home with a partner; the buyer's bank and appraiser can pull parameters from it without phone calls; and you come across as someone with the sale under control, which counts at the negotiating table.

The key decision is the same as for the brochure in the brand-as-a-system guide: the info sheet has to be an editable source file, not a one-off generated image. A sale is a live process — the price gets adjusted after three weeks, the handover date shifts, an answer to a common question gets added. With a generated image, every such change means regenerating the whole thing (and getting something different); with a source file, it's one edit and a new PDF export. Changing the price or date = one edit — that's the whole point of this phase.

What to lay it out in

At the time of writing, there's more than one path from Claude into layout, and it's worth knowing which is which. In spring 2026 Anthropic introduced Claude connectors for professional creative apps, among them Affinity (now the Canva app family, with a free tier) — Claude Cowork, a desktop mode that works over a folder of files, can create documents in Affinity through it, lay out text and images, and export; the source file is an .af format on your own disk. The second path is the Canva connector: the design is created and stays editable in Canva's cloud, which is plenty for an info sheet and is the more approachable option for non-technical sellers. Both connections are fairly new and still evolving, so check the documentation for current capabilities — the principle this phase relies on (you brief and approve, the agent lays out, the result is an editable design) doesn't depend on the specific tool. The detailed craft of laying out via Cowork — the "draft, approve, produce" rhythm, proofing, exporting to PDF — is walked through step by step in the brand guide, phase 2; we won't repeat it here and will focus on what's different for real estate.

Drafting a structure from the folder

Cowork works over a folder — and you already have one: sell-flat from phase 1, with the flat.md document, the selected photos, and the floor plan.

Work over the sell-flat folder. It contains flat.md (the single
source of truth about the property), a subfolder photos/selected
with the final photo picks, and floorplan.pdf.

Propose a structure for a 6-page A4 sales info sheet: for each page,
its purpose, content (which sections from flat.md), which photos
(filenames). I'm thinking: a cover with the hero photo, the flat at
a glance (parameters + floor plan), photo gallery, condition and
repair history, costs and the building, the buying process (steps,
legal service, escrow, contact).

Two rules: all facts strictly from flat.md — if content is missing
for a page, say so and don't invent anything. And propose which
fields on the pages should be marked as VARIABLE (price, dates, sale
status) — those will change during the sale, and I want them in one
defined place in the document so a change means one edit.
Don't produce anything yet, I'm waiting to approve the plan.

That note about variable fields is a small thing that turns the brochure into a system: the price and dates live in one defined spot in the layout (typically the cover and the process page), so a change is fixing two text boxes, not archaeology across six pages.

Building, iterating, exporting

Commission the build spread by spread, approving previews as you go — with the rule that text comes from flat.md, never from memory. Once approved, export two versions: a lightweight PDF for emailing to buyers, and a full-quality one for printing at viewings. And crucially: the source file stays in the folder. When the price changes three weeks later, edit flat.md, open a conversation, and say:

The asking price and handover date changed in flat.md (timeline and
process section). Carry the change into the info sheet: find every
place in the layout where the price or date appears, list them with
the old and new value, and once I approve, fix them and export a new
PDF (infosheet-v2). Don't change anything else in the layout.
Finally, remind me where else outside the info sheet the change
still needs to be carried, based on the outputs list in flat.md.

That last sentence is a preview of phase 7: the property document also keeps a list of every derived output (info sheet, website, portal listings), so no change gets lost along the way. We'll get to that shortly — but first, the biggest derived output of all: your own website.

Phase 6: a one-page property website

A website for a single flat sounds like overkill — and it's one of the most practical steps of the whole sale. The reasons are mundane. Real estate portals compress photos, truncate text into their own templates, and your listing is one of thousands; your own website shows the flat the way you want: a full-resolution gallery, a downloadable floor plan, complete FAQ, viewing slots. Then every listing, email, and message carries one link — and you have one place where the current state (price, open slots, "reserved") is always accurate. And an FAQ section, generated straight from the property document, provably cuts down phone calls: anyone who still has questions after reading the site is either a serious buyer or an unusually thorough browser — both useful to know.

How to build and deploy a website with Claude Code on Vercel — from the first prompt through git as a safety net to a custom domain — is covered step by step in a dedicated guide; it applies here without changes, and we won't repeat it. A one-page property website is actually an ideal first project: no database, no login, just one page built from your own data. The domain can be simple and descriptive; the free address Vercel assigns works fine too — buyers aren't buying a domain, they're buying a flat.

Briefing the website from the document

Build a one-page website to sell a flat. The single source of
content is the file flat.md in this directory — every fact comes
from it; don't invent anything, you can shorten text, not change it.

Sections top to bottom: hero photo with headline and price, the flat
at a glance (parameters in a clean grid), photo gallery (photos from
the photos/selected folder, in the order given in flat.md), floor
plan (image + link to the PDF), condition and repair history, living
costs, frequently asked questions (the full questions-and-answers
section from flat.md), viewing times and contact.

Requirements:
- works well on mobile — most buyers arrive from a phone
- no contact forms that store data; contact is email and phone as
  links
- price, sale status, and viewing times live in code in one place
  (a config file), so I only change them once
- no third-party analytics, no extra cookie banners
- a footer note that legal service is handled by a lawyer and the
  purchase price goes through escrow

Go section by section, show me a preview after each, and wait for my
approval.

The request for a config file is worth noting: it's the same variable-fields principle as the info sheet. Changing the price is then one edit to one file plus a deploy — Claude Code can do both from an instruction like "update the price from flat.md and deploy." And the "no forms that store data" requirement isn't laziness: a form means storing visitors' personal data, and that's a liability you don't need for a single sale. Email and phone are enough; you'll track buyers on your own side (phase 8).

What doesn't belong on the website

The property website is public — which changes the rules compared to the info sheet, which you send to specific people. Don't publish the exact street number until it's necessary; the street and neighborhood are enough for buyers to decide about a viewing, and there's no need to invite thieves into an empty flat with a fresh address attached. No scanned documents: the title deed carries national ID numbers and doesn't belong online even with them blacked out — anyone who wants to see it gets it at a viewing. No photos with personal items, mail, or a name on the doorbell visible. And phrase when the flat sits empty as "viewings by appointment within these windows," not as a public schedule of when you're away.

The website is finished; run a check before deploying:
1. Compare every fact on the site against flat.md — list every
   discrepancy (a number, a phrasing, a missing fact); the website
   must not diverge from the source of truth
2. Read the text and photos as someone trying to find out where I
   live and when I'm not home: list everything that reveals more
   than it needs to (exact address, viewing schedule, security
   details, personal items in photos)
3. Confirm that no document with a national ID number, a signature,
   or a third party's full name exists anywhere in the repo or on
   the site
4. Verify mobile functionality: gallery, floor-plan download,
   clickable phone and email
List findings by severity, and don't fix anything without my
approval.

Run this prompt before the first deploy, and again after every major change. It's a small routine with a big payoff: the website is the one output truly anyone can see, and a mistake in it spreads fastest.

One more status detail worth having: the website's config file should also carry a sale status — "on the market," "reserved," "sold." The moment you sign a reservation, flip the website to "reserved" and leave it up: buyers calling about a dead listing are annoyed buyers, while a visible "reserved" with an option to leave contact info in case the reservation falls through is a backup door that's come in handy more than once — reservations really do fall through on financing sometimes. The website is the one channel that can convey that in-between state gracefully; the portals only know "up" and "gone."

Phase 7: portal listing variants — from one source

You'll list the flat in more than one place: major real estate portals, classifieds sites, social media, a flyer in the building's entryway (don't underestimate it — the buyer is sometimes a neighbor wanting the flat for their daughter). Each place has different length limits, different field structures, different audiences, different tone. The classic self-seller mistake: write one text and manually bend it three different ways — and from that moment on, three versions exist that drift apart the instant the price changes.

The systematic fix: the approved listing text from phase 4 is the one "master"; variants are generated from it, and none get edited by hand. When the content changes, the master changes (and the property document), and the variants get regenerated. Manual variant edits are off-limits — not out of pedantry, but because that's the only way to keep the "everywhere says the same thing" rule intact.

A matrix of variants

From the property document in the Project (approved listing text
section), prepare variants for these places: [major real estate
portal — full length, second portal — 1,500 character limit,
classifieds site — 600 character limit, a post for a local Facebook
group, a flyer for the building entryway, A4 format].

Rules for every variant:
- facts must not differ from the master; shorten by cutting, not by
  rephrasing facts
- every variant keeps: layout, floor area, floor, energy rating,
  price, mention of the dampness with its fix (for the shortest
  variant, at least the sentence "repair history in the listing on
  the website"), a link to the property website
- write the Facebook group post plainly, first person ("selling my
  grandmother's flat..."), but without emotional pressure
- flyer: large type, the essential facts, a link to the website
  including as a QR code (generate me the source for it)

For each variant, add a METADATA line: where it belongs, length
limit, actual length, and the date of the master version it was
built from.

The METADATA line looks like bureaucracy and is the smartest part of the prompt: it records which master version each variant came from — so three weeks from now, you can safely tell which variant has gone stale. Save the variants into an ads folder next to the document; filling out the portal forms then becomes a copy-paste job.

Changing the price without a forgotten listing

The most common self-seller embarrassment: an old price still sitting on one portal. Buyers compare listings against each other, and a mismatch reads as either confusion or a trick. The fix is simple — the property document keeps a list of every live output, and every change is worked through as a checklist:

The asking price changed (flat.md is already current, price and
rationale section + timeline). Generate a checklist for propagating
the change based on the outputs list in flat.md:
- which listing variants need to be regenerated (do it now, and bump
  the master version on their METADATA line)
- which places I have to change by hand myself (portals where the
  listing is entered through a form) — list them as a checkbox list
- website: remind me to update the config file and deploy
- info sheet: remind me of the procedure from the info-sheet phase
  (fix the variable fields, new export)
Add a DATE COMPLETED field to each item — I'll fill it in once
handled, so it's visible that nothing sat with the old price.

Five minutes gets you the certainty that real estate agencies rely on an internal system for. Use the same checklist for every material change: handover date, "reserved," a new answer added to the FAQ. And notice what just happened to the promise from the intro: a price change is one edit in flat.md — and then just controlled, checkable propagation into the outputs, not panicked guesswork about where the flat is still listed.

Phase 8: tracking buyers and viewings — a tiny CRM for one sale

Within a week of publishing, something happens that self-sellers usually aren't ready for: a lot of people get in touch at once. Calls, texts, emails, messages from portals and the Facebook group. Some are serious buyers, some are curious browsers, some are agents fishing for an exclusive listing ("I have a buyer for your flat" — they don't), and some are scammers (covered in the security phase). Without tracking, this turns into a mess within three days: you don't know who's already had a viewing, who you promised to call back, who wanted the floor plan, who's gone quiet since Tuesday. And speed and reliability of response are exactly where a self-seller can outdo an agent — you're only selling one flat.

The fix is a miniature CRM: a spreadsheet where every buyer gets a row, a status, and a next step. No extra software needed — a spreadsheet in Excel or Google Sheets is enough; anyone who wants a full database with logins can find the approach in the guide to an internal CRM with Supabase, but for a single sale that's overkill.

Tracking structure

Choose statuses as a funnel a buyer moves through: inquiry → paperwork sent → viewing scheduled → post-viewing → offer → reservation (plus the end states "dropped out" and "fraud/spam"). Columns: name, phone, email, source (which portal — after two weeks you'll see which channel brings serious buyers), status, date of last contact, next step and its date, notes (what they asked about, financing, what matters to them), and for offers, the amount and terms.

Set up a tracking sheet for flat buyers as a table (CSV, I'll import
it into a spreadsheet app). Columns: name, phone, email, source,
status, last contact, next step, next step date, notes, offer and
terms.

Along with it, write short operating rules on one page:
1. Definitions of each status (inquiry / paperwork sent / viewing
   scheduled / post-viewing / offer / reservation / dropped out /
   fraud) and exactly what has to be true for a buyer to be in it
2. First-contact rule: what to send everyone (link to the property
   website, available viewing times) and what to ask them (who
   they're buying for, financing, timeline)
3. Follow-up rules: after how many days without a reply to follow
   up, how many times at most, and when to move a buyer to "dropped
   out"
4. A data cleanup rule: delete the sheet after the sale, keeping
   only what relates to signed contracts

You'll get a table and a set of rules that keep you steady in the rush. Item 2's rule is worth following to the letter: asking about financing and timeline in the very first reply isn't rude — it's a decent filter. A serious buyer answers gladly (it saves them time too); a browser evaporates. And take item 4 seriously: the tracking sheet is full of strangers' personal data; you're collecting it for one purpose, and you delete it once that purpose is served.

Weekly review and follow-ups: the agent reads, the human sends

Tracking only works if it's checked regularly — and that's exactly the kind of mechanical work AI is good for. Once a week (or daily during a busy stretch), give it the current state of the sheet and have it prepare an operational summary. Anyone using scheduled tasks in Claude can turn this into a routine; the substance stays the same:

Here's the current buyer tracking sheet [insert table]. Prepare an
operational summary:
1. Who's waiting on my reply (status vs. next-step date) — sort by
   urgency
2. Whose follow-up window has passed: for each, a draft of a short
   follow-up message continuing from the last contact in the notes
   (SMS or email, factual, no pressure), with a question that moves
   the status forward — e.g. offering a specific viewing time
3. Who should move to "dropped out" per the rules — suggest it, I'll
   confirm
4. Funnel summary: how many people at each status, which source
   brings the most viewings, and whether the pace matches the flat
   having been listed for [X] weeks

Don't send any messages — hand them back to me for editing and
sending. Only read offers and their amounts, never comment on them
to buyers.

That last paragraph is the "AI drafts, human approves" rule in practice: you review the draft messages, edit them in your own words, and send them yourself. Partly because it's your relationship with actual people, partly because a single auto-sent message with the wrong name or an old price costs more trust than the automation ever saves. Item 4's summary has one more role: it's your early-warning system. If inquiries keep climbing for three weeks but viewings don't, something's off with the price or the photos; if viewings climb but offers don't, something's off at the flat itself — and either is better to know in week three than month three.

Scheduling viewings without the back-and-forth

Scheduling is the single biggest time sink of this whole phase. Simplify it: write two or three viewing windows per week into the property document (say, Tuesday early evening, Saturday morning) and offer specific slots from them — "Tuesday 5:00, 5:45, or Saturday 10:00?" is a message answered in one word, while "when works for you?" kicks off a week of back-and-forth. Blocks also mean viewings run back to back and you're at the flat twice a week, not five times. Anyone who wants to can build a simple booking page for the slots — but a spreadsheet and offering slots from it is plenty.

Phase 9: viewings — a script instead of improvisation

For a self-seller, a viewing is the hardest part, because it's live: strangers in your home, unrehearsed questions, emotions on both sides. Good news: ninety percent of a viewing can be prepared in advance — and you've already done most of that prep, because you worked through buyers' questions with AI in phase 2, and the answers live in the property document.

Script and rehearsal

A viewing should run on a path you control: welcome, a walk through the flat in a deliberate order (end in the strongest room, not the cellar), facts about the building and costs, room for questions, and a clear close — what happens next, by when you'll be in touch, how to submit an offer. Keep the printed info sheet and a folder of documents on hand for reference (the PENB or bills, floor plan, a sample reservation contract from the lawyer). Rehearsal handles the nerves around questions:

From the property document in the Project, prepare viewing
materials for me:

1. A 30-minute script for viewing a three-bedroom flat: room order
   with what to say in each (a fact, not a slogan — the year the
   windows were replaced, honestly noting the original bathroom
   core), where to mention the dampness repair before anyone asks,
   and how to close the viewing with a next step
2. A list of the 15 most likely questions with answers from the
   document — one-sentence answers each, plus where the supporting
   document is in the folder
3. Three uncomfortable situations to rehearse: a buyer claiming
   "dampness always comes back," a buyer pushing on price right at
   the viewing, an agent posing as a buyer fishing for an exclusive
   listing. Role-play them with me one at a time — you play the
   other side, and give me feedback after each exchange on what to
   say better
4. A pre-viewing checklist: air the flat out, turn on lights,
   valuables and personal documents put away, info sheet and folder
   ready

Item 3, done in voice mode on your phone, is almost a dress rehearsal — treat it that way: saying "the core is original, we've priced that in" out loud twice beforehand means you'll say it calmly at the viewing instead of squirming. Buyers read squirming as concealment, even when nothing's being concealed.

Three sober safety rules for viewings: go in pairs (the second person doesn't have to say anything, just be present), keep valuables and documents out of the flat, and get buyers' surnames beforehand — anyone who won't give a surname before a viewing has no reason to be at one. It's not paranoia; viewings are a standard recon channel for thieves, and a "buyer" photographing locks and asking about the alarm is information for your tracking sheet, not a joke.

Video, second viewings, and company

Three situations that round out the script. A buyer from far away may want a video viewing: run it just as deliberately as a live one (script, daylight, honest shots including the bathroom core), and treat it as a preliminary round, not a substitute — a serious buyer comes in person, and anyone who wants to buy from video alone is either an experienced investor or a risk (see the security section). A second viewing is a great signal: the buyer brings a partner, a parent, or a tradesperson. Don't resist a tradesperson or a technical inspector either — a flat honestly described has nothing to fear from an inspection, and a buyer who paid for one is taking the purchase seriously; just note the date and who's coming in your tracking sheet, and be there for it. Parents in an advisory role tend to be the toughest audience — exactly where the folder of documents pays off: the objection "that dampness will come back" is answered better by a repair invoice than by any sentence. In general: the second viewing is where offers happen, so treat it as a meeting to prepare for, not a tour.

After the viewing: notes and follow-up

Right after a viewing (even by voice, on the way home), dictate notes into the tracking sheet: who came, what they asked, how they reacted to the price, how interested they seemed, what was promised. The next day, everyone who attended should get a short message — a thank-you, an answer to whatever was left open ("attaching the scanned wiring inspection you asked about"), and a question about the next step. The weekly review from phase 8 will draft these for you; there's just one extra rule here: send a follow-up even to people who showed no interest — one sentence, "if you change your mind, reach out," costs nothing, and a second round of interest is common in real estate once a favored buyer falls through elsewhere.

Phase 10: a serious buyer, a reservation — and the hard line where the lawyer takes over

The moment arrives when someone says "we'll take it." Paradoxically, this is where the sale gets most risky — and it's also where this guide has its clearest rule: from the reservation contract onward, a lawyer takes over. AI helps you understand the lawyer and use their time well; it never replaces them. This isn't a cop-out — it's arithmetic: a mistake in the listing text costs you an awkward moment; a mistake in the purchase contract or the payment mechanism can cost you the flat and the money both. A lawyer's full transfer service (reservation contract, purchase contract, escrow, filing the registration of title) is a fraction of what a commission would cost — and it's the one line item in the whole sale you don't cut corners on.

Vetting how serious a buyer is

Before you pull the flat from the market for someone, you need to know two things: that they want to buy, and that they can. The first shows in behavior (they show up to scheduled times, ask relevant questions, work through concrete steps); the second you have to actually check — and that's a normal business question, not rudeness. For a buyer with a mortgage, how far along the process is matters most: "the bank will approve it" said offhand means months of uncertainty; a loan already in motion with an appraisal and a preliminary approval is a different league. For a buyer "paying cash," ask for proof of ability to pay before signing a reservation — typically confirmation of a reservation deposit placed into escrow. And basic due diligence: check the insolvency register (public, online, free); if a company is buying, check it in the business registers — the approach is covered in the tip on vetting a business partner.

I have a serious buyer for the flat: [description — a couple buying
with a mortgage, want to sign as soon as possible; a second buyer
claims they're paying cash]. Prepare:
1. A set of financing questions for each of them — worded politely,
   but so the answers are verifiable: what stage the loan is at,
   which bank, has an appraisal happened, how long the preliminary
   approval is valid, how much is their own funds; for the cash
   buyer, how and by when they'll prove the funds
2. A list of documents to request or verify before signing the
   reservation (and which of those a lawyer verifies, not me)
3. Warning signs in their answers: what evasiveness on which
   question means, and when it makes sense to favor the other buyer
   with a lower offer but more solid financing
4. A way to tell buyers transparently that the reservation is signed
   at a lawyer's office and the deposit goes into escrow — and that
   this protects both sides

None of this is legal judgment — just meeting prep; I'll go through
the legal steps with a lawyer.

Item 3 tends to prove the most valuable: a higher offer with shaky financing is worse than a lower, solid one, because every month spent waiting on an unapproved loan is a month other buyers cool off. This is a call neither AI nor the lawyer will make for you — but with the right questions asked, you make it on facts, not sympathy.

Reservation contract, escrow, purchase contract: why a lawyer, no exceptions

Worth explaining why the line sits where it does, so it's a decision you understand, not dogma. Contracts transferring real estate aren't prose — they're mechanisms: who signs what and when, when and where money gets deposited, what happens if the buyer's bank doesn't send the loan, who bears the risk between signing and the registration of title, how withdrawal is handled. A template pulled off the internet or generated by AI can look flawless and still not fit your situation exactly — the buyer's mortgage lien, land-registry timelines, your damp stain in the condition disclosures. And above all: the purchase price never changes hands directly — it moves through attorney or notary escrow, or a bank letter of credit — the seller gets the money after the transfer of ownership is registered, and the buyer has certainty they're paying for a flat they'll actually receive. This mechanism is the core of a safe transfer, and it's set up by a professional who's liable for it — an insured lawyer or notary, not a template and good faith.

AI's role sits elsewhere, and it's a big one: turning you into an informed client. An hour with a lawyer is best spent on your questions, not on explaining terms a machine could have covered. How to read a contract with AI in depth is covered in a dedicated tip on contracts before signing; here's the version for this case:

I'm uploading a draft reservation contract from a lawyer [insert;
or the other side's draft]. I'm the seller, with no legal training.

1. Walk me through the contract clause by clause in plain language:
   what each one means, who it obligates to what, and what happens
   on breach
2. List every deadline and amount in one table: what, who, by when,
   what happens if it's missed — I'll check it against what was
   agreed verbally
3. Flag anything risky or unusual for me as the seller, and anywhere
   the contract is silent (what if the buyer's loan falls through?
   what if I withdraw?)
4. Prepare a list of questions for the lawyer, ranked by importance —
   specific, referencing the relevant clause

Don't advise me on whether to sign — that's a decision for me and
the lawyer. Your job is making sure I understand every sentence
going into the meeting.

Repeat the same approach for the purchase contract and the escrow agreement. And if the other side (say, a buyer with their own lawyer) sends their own draft, add one extra step: have AI compare the two versions and list the differences — "what was added, what was dropped, in whose favor" is half an hour of work that cuts a lawyer meeting in half. Just watch for one failure mode that comes up with long contracts: a model can skim past a clause or summarize it too loosely. So always use the output as a map for reading the contract, never as a substitute — the lawyer has the final word.

Between signing and the keys: what happens, and how long it takes

Self-sellers are often caught off guard that signing the purchase contract doesn't end things — a period follows that has its own order and takes weeks. Roughly: the buyer (or their bank) deposits the purchase price into escrow; a filing for registration of title goes to the land registry; the registry marks a plomba (a registration flag) and a statutory protection period runs before the transfer can be approved — plan on this taking weeks, not days; once registered, the escrow agent releases the funds per the escrow agreement, and the flat is handed over. With a mortgaged buyer, the bank's own lien filing gets woven in too, typically registered ahead of the purchase contract. The lawyer sets the exact order and deadlines in the contracts — your job is understanding it well enough to know when silence is normal and when it's worth asking about. Here's a good prompt for that, once contracts are signed: "list a timeline from the signed contracts — what, who, by when, and where I should actively follow up if nothing happens." Put the timeline in your calendar; waiting with deadlines in hand is calm, waiting blind is stress.

Phase 11: price and negotiation — AI as a sparring partner

We've sidestepped price so far because it deserves its own phase — it's the single most emotional decision in the whole sale. Especially for an inherited flat: the price "grandma's flat ought to fetch" tends to be a mix of memories and neighborhood lore, not market data. The systematic approach is more sober: the price is a hypothesis you set from data, which the market then confirms or disproves — and you know in advance how you'll tell which, and what you'll do about it.

Price research

Data sources: current listings for comparable flats (layout, condition, location — watch out, asking prices are wishes, not reality), data on actually completed sales (registry prices, price maps), and, when uncertain, a professional appraisal — an appraiser puts a stamp on the number and costs a fraction of a commission; the buyer's bank will order one anyway. Deep research with citations handles the comparable-listings side well — the general approach is covered in the tip on deep research. Here's the prompt for our case:

Run deep research on listings comparable to mine: [three-bedroom,
approx. X sqm, panel building post-renovation, original bathroom
core, regional city — neighborhood, floor with elevator].

1. Find current comparable listings nearby; for each, cite the
   source, its parameters, how it differs from my flat (condition,
   floor, location), and a link
2. Split them into genuinely comparable vs. only superficially
   similar, and explain why
3. From the differences, infer what the market in this location
   values (a renovated core? a balcony? the floor?) and how that
   maps onto the price range
4. State what CAN'T be determined from public data (actual sale
   prices vs. asking prices, how long listings sat) and how to
   verify that another way

Cite a source for every claim. Don't give me a single "correct
price" number — I want to understand the range and the factors; the
decision is mine.

That last sentence matters: AI confidently handing you one number is more dangerous than none — it has no access to actual sale prices, and asking-price listings are systematically inflated. The output of the research is a range and a set of arguments; you pick the number yourself and write it into the property document — both the asking price and your real bottom line, which nobody but you should know. That bottom line, set calmly ahead of time, is the best protection against a decision made in the heat of a viewing.

Sparring: negotiate before it counts

Negotiating on price is a skill buyers often have (they've bought before, or bring along an experienced friend) and a first-time seller often doesn't. Preparation closes much of that gap: walk through the other side's likely moves ahead of time, prepare your responses, and say them out loud.

The flat is listed at [price from flat.md], my bottom line is [X],
it's been listed for [3] weeks, I have [2] serious buyers and [5]
completed viewings.

Be my sparring partner for negotiation:
1. List the moves buyers are likely to try: an aggressive first
   offer well below asking, listing defects as leverage (the core,
   "the dampness will come back"), "I have another flat, deciding
   today," a salami tactic of extra requests piled on afterward
   (furniture thrown in, a later handover), emotional pressure
   around the inheritance
2. For each move: what it actually means, how to respond factually,
   and when it's legitimate instead (a defect I disclose in the
   listing IS grounds for negotiation — prepare a fair concession
   range for me justified by facts from flat.md)
3. Role-play two rounds with me: in the first, play a polite but
   tough buyer with a mortgage; in the second, a fast "cash" buyer
   wanting a discount for speed. After each round, give me feedback:
   where I gave in unnecessarily, where I was needlessly hard, what
   to say next time

Two principles tend to emerge from a session like this almost every time. First, tie concessions to something in return: a discount for a fast handover or for taking the flat "as is" is a trade; a discount "because they asked" is just a loss. Second, don't decide on the spot: "thanks for the offer, I'll get back to you by tomorrow" is a legitimate answer to any proposal — and the only reliable defense against pressure to decide immediately, which is itself a warning sign. Log offers in the tracking sheet along with their terms; comparing offers from memory is how you pick wrong.

When the market goes quiet: a plan B set in advance

The last piece of price strategy is written on the day the listing goes live, not after a month of silence: write down in the property document, ahead of time, what you'll do if things aren't moving. For example: after two weeks with no viewing booked, revise the photos and headline; after four weeks with no offer, adjust the price by a pre-decided step; after two adjustments with no result, consult an appraiser or an agent. The reason is psychological: the decision "I'll lower the price" is easy made calmly and nearly impossible made under anxiety — a seller a month into silence either stubbornly holds the price (and the listing "goes stale," since the market reads a listing's history, and a long-sitting one raises suspicion) or panics and underprices. A plan B written in advance turns that into execution instead of drama — and the weekly funnel summary from phase 8 tells you exactly when a trigger point has been hit. One more thing: one large price adjustment reads more clearly to the market than a string of small cuts; have AI work that out from your own funnel data too, when the time comes.

Security and scams: who shows up when you're selling a flat

A public listing with a phone number is an invitation not just for buyers. A self-seller is an attractive target for scammers: valuable assets, no experience, acting under time pressure — and unlike an agency, no filter that spots a fraudulent pattern on sight. That filter can be learned, though, because scams repeat across a small handful of patterns.

Patterns you'll run into

• A "buyer" from abroad who's never seen the flat. Writes in broken language, wants to buy sight unseen, often "for a son relocating for work." Followed by an overpayment variant (they send "too much" and ask for the difference back — their payment turns out fake later, and your refund is gone), or an attempt to extract documents "for the bank." A genuine buyer from abroad exists, but behaves like anyone else: a viewing (in person or through a proxy), a lawyer, escrow.
• A fee up front. Anyone asking you for money in exchange for buying — a "verification fee," a "buyer's reservation fee," a fee to some obscure platform that "safely" processes the payment. The direction of money in a sale is unambiguous: toward you, and only through escrow. Every exception is a scam.
• Extracting documents and data. Requests for a scanned ID, a national ID number, a complete title deed, an account number with extra details — usually "to prepare the contract" or "for the bank," before anything has actually happened. Personal data goes to a lawyer for an actual contract, not to a buyer by email. Be especially careful with national ID numbers: don't email documents that carry one (the title deed, the title of acquisition) to anyone you haven't verified — that's exactly the data identity theft is built from.
• Fake agents and "I have a buyer for you." They call within an hour of publishing. The goal is an exclusive contract, not your buyer; the "vetted buyer" mysteriously vanishes after signing. A polite "I'm not working with agencies" and move on.
• Phishing through the portals. Messages like "your listing has been suspended, verify here," or fake portal payment pages. Log into listing accounts only directly, never from a link in a message — and turn on two-factor authentication; a hijacked listing account with your flat on it is a tool for scamming other people under your name.
• Pressure to move fast, anywhere in the process. "Sign today, the offer's off tomorrow" — from buyout firms, "investors," and some buyers alike. A legitimate counterpart can handle "I'm sending the draft to my lawyer, I'll be in touch in three days." Anyone who can't handle that sentence just told you the important thing about themselves.

AI as a second pair of eyes

The advantage: scam patterns are heavily represented in models' training data, and AI spots them more reliably than a rattled person who just heard "I'll buy it right now at full price." Make it a habit — run every message that feels off, unusually good, or time-pressured through AI before you reply:

I'm selling a flat myself. I got this message from a buyer [insert
the full exchange including headers/numbers, black out any third
party's personal data]. Break it down:
1. Which elements match known real estate scam patterns (buying
   sight unseen, overpayment, upfront fees, extracting documents, a
   fake agent, phishing) — and which are actually normal buyer
   behavior
2. What verification I can do before replying (calling the number
   given, checking the sender's domain, looking up the name)
3. Suggest a safe reply: gives away nothing extra, asks for a
   legitimate next step (a viewing, a lawyer, escrow), and lets a
   scammer back out quietly without insulting a genuine buyer
If you're not sure, say so — better to err on caution than offer
false reassurance.

And three rules that hold regardless of what AI says, because they're not situational, they're absolute: money never gets accepted before contracts are signed except into escrow (and none ever gets sent out), documents carrying national ID numbers never go by email to anyone unverified, and no decision gets made under time pressure created by the other side.

Protecting data in your own system

Security has an internal side too — the data you collect and send out yourself during the sale. A summary of the rule that's run through the whole guide: paperwork about the flat goes into AI only inside a paid account with contractual data protection; national ID numbers and scanned documents don't belong in prompts at all (black them out, leave them out) — a model doesn't need them to explain a contract; the buyer tracking sheet holds strangers' personal data — keep it lean (only what you need), never forward it, and delete it after the sale; and the exact address, your absence schedule, and any document don't belong on the public website. And for AI agents, the rule from across the whole site applies: they draft text and read data, but they don't send messages, pay, or delete anything — every output headed toward a live human leaves your own hand. The broader framework for what belongs in AI and what doesn't is in the chapter on ethics and safety.

After the sale: handover, re-registrations, and cleaning up the system

The sale doesn't end with the signed purchase contract and the filed registration — an administrative aftermath follows, right when your energy is lowest, and it's exactly where the last disputes are born: unpaid utility balances, "where are the cellar keys," a forgotten registered address. It's exactly the kind of work checklists generated from the source of truth are good for.

Handover protocol

The flat is handed over once ownership is registered (the exact moment is set by the purchase contract), and the handover needs paperwork: a handover protocol with readings from every meter (electricity, gas, water, heat — with photos of the meters), counts of keys and fobs handed over (flat, building, mailbox, cellar), a list of fixtures staying in the flat (matched against the fixtures-and-fittings section of the property document — this is why you wrote it in phase 2), and a statement of condition. Both sides sign, each keeps a copy; the meter readings from the protocol then feed into the utility transfers, so old-period balances stay with you and new-period ones go to the buyer — exactly as it should be, and none of it is provable without the protocol.

The sale is signed, we're waiting on registration and handover. From
the property document in the Project, prepare:
1. A handover protocol to fill in: parties, date, meter readings
   (table: meter, meter number, reading, photo taken yes/no), keys
   and fobs by count, fixtures staying (pull the list from the
   fixtures-and-fittings section), a condition statement, signatures
   — have the draft checked by the lawyer so it lines up with the
   purchase contract
2. A checklist of re-registrations and cancellations with order and
   dependencies: electricity, gas, water/heat through the SVJ, the
   building manager/SVJ (change of owner, service charges), home and
   contents insurance, registered address, mail, government data
   boxes, anywhere I'm listed at this address
3. A checklist of taxes and obligations, flagging that I should
   verify exact rules with current sources or a tax advisor: income
   tax on the sale and exemption conditions, who files what for next
   year's property tax
4. Wrapping up the sale system: end the listings everywhere on the
   outputs list in flat.md, take down the website, delete the buyer
   tracking sheet, archive the contracts and the protocol

Item 3 doubles down on the phase-1 rule: tax rules (exemptions based on ownership duration or reinvesting the proceeds into your own housing, reporting obligations) change and carry exceptions — AI can map out the questions, but verify the answers with the tax authority or an advisor. It's one consultation; a penalty for a missed exemption isn't something a well-run sale deserves.

Cleaning up the data: the system can tidy up after itself

The last step is usually the most neglected: the sale system holds personal data — yours, buyers', the purchaser's — and once its purpose is served, it should disappear. Delete the buyer tracking sheet (archive only what relates to the signed contracts), take the website down from Vercel, end the listings (portals like to leave them up "for statistics"), and go through your AI conversations: anything that included paperwork with personal data, delete that too. The property document, on the other hand, is worth keeping — next time you sell or rent anything, you have a ready template and a rehearsed process. That's the quiet payoff of the whole systematic approach: a second sale would take a fraction of the first one's time, because you're not buying outputs, you're building a capability.

What AI can handle — and what needs a lawyer or an authority

The promised table, all in one place. The left column is this guide's territory; the right is the professionals' — and the line between them isn't opinion, it's a mix of law (what only a lawyer, notary, or authority may do) and common sense (where a mistake costs more than the fee).

Notice the pattern in the left column: everywhere, AI drafts, explains, watches for consistency, and prepares questions — and nowhere does it sign, send, pay, or decide. That's not a limit on its abilities, it's a system design: every step with legal or financial consequences passes through a human who's liable for it — you, or a professional you've hired.

When it's still worth hiring an agent

An honest guide to selling without an agent should also say when an agent is the right call — because the argument here was never "agents are pointless," it's "most of their work is a system you can build yourself." An agent (a good one — ask for references and a concrete plan, not promises) makes sense when you don't have the time: viewings, calls, and follow-ups are real hours during the workday, and the system in this guide makes them more efficient, not free. It makes sense for complicated cases: co-ownership disputes, a property on the other side of the country, a sale under foreclosure pressure — there you're buying experience with situations like yours, not a listing. And it makes sense once you've tried the process and the market has said no for three months: an outside read on price and presentation is then worth more than another month of your own certainty.

Even for that case, though, building the system pays off: a property document, an honest list of defects, buyer tracking, and knowing where the legal line sits make you a client who can manage and check an agent's work — who spots a lazy listing, wants to see the buyer log, won't sign exclusivity without a termination clause, and knows the commission is negotiable. The worst position in the market isn't "selling it myself" — it's "I don't know anything and delegated everything."

Common mistakes

• Writing the listing before the property document. Outputs without a source of truth drift apart at the first change, and every reply to a buyer gets invented from scratch. Two evenings on phases 1-2 are the best-spent time of the whole sale.
• A varnished listing. "Investment opportunity" for a flat with mold, a retouched crack, unlabeled virtual furniture. It surfaces at the viewing at the latest, undermining trust in everything else — and an undisclosed defect comes back as a legal problem even years after the sale. AI drafts, the owner vouches.
• A price from neighborhood lore instead of research. An overpriced listing means months of silence, then a discount from a position of weakness — the market reads a listing's history. Price is a hypothesis built from data — and set your bottom line calmly, in advance, not at a viewing.
• No buyer tracking. By the second week you don't know who's had a viewing or what you promised whom; a serious buyer buys elsewhere in the meantime because you didn't follow up. A status sheet and a weekly follow-up review are minutes of work.
• Contracts "based on a template from the internet" or from AI. The one place where cutting corners can turn into a disaster. A lawyer writes the contracts, money moves through escrow — AI helps you understand and ask questions, not sign.
• Deciding under pressure. "Sign today," "send the documents now," "I'll refund the overpayment." Time pressure created by the other side is itself a warning sign; a legitimate buyer can handle "I'll get back to you in three days."

The best tools

• Claude (claude.ai) with a Project — the hub of the whole system: the property document as persistent context, probing in the role of a buyer, listings and their fact-checking, contract breakdowns for a layperson, negotiation sparring, breaking down suspicious messages. Reads scans and photos of documents too.
• Claude Cowork + Affinity or Canva (via connector) — laying out the info sheet as an editable source file: Cowork works over a folder of source material, you approve spreads, and the result is a PDF plus a source file where a price change is one edit.
• Claude Code + Vercel — a one-page property website built from the property document: gallery, floor plan, FAQ, viewing slots; deployment and every change on a single instruction. Covered in the website guide.
• Land registry (online lookup, CzechPoint) — the flat's legal status: ownership, liens, easements. The first stop of the inventory, and the place to verify, not to trust.
• A spreadsheet app (Excel, Google Sheets) — buyer tracking with statuses and follow-ups; that's all one sale needs. Deep research with citations then covers the price research.
• A lawyer with escrow — the one "tool" on this list that can't be replaced: the reservation and purchase contracts, escrow for the purchase price, filing the registration of title. AI turns you into their informed client, not a substitute for them.

What you get out of it

• Time: prep that would otherwise smear across weeks of chaotic searching fits into a handful of evenings; replies to buyers, listing variants, and follow-ups stop being built from scratch each time. The surge of interest right after publishing gets handled with a system instead of panic.
• Money: you trade an agent's commission for a lawyer's fee, a fraction of it — and the system protects you from more expensive mistakes: an underpriced listing from inexperience, a discount forced at a viewing, a scam, a disclosed-too-late defect that comes back to bite.
• Peace of mind: you walk into a viewing with rehearsed answers, into a price negotiation with a bottom line set in advance, into signing with a contract you understand clause by clause. Most of the stress of a self-sale is the stress of improvising — and you've replaced improvising with a system.
• Quality: consistent, honest presentation across every channel — the info sheet, the website, and even the shortest listing all say the same thing, because they all read from the same source. Buyers read that as trustworthy, and trust is the currency that matters when selling the biggest asset of your life.

Pro tip

Once the system is running, feed it feedback: once a week, have AI review the whole funnel — inquiries, viewings, and offers by week, buyer sources, recurring questions — and answer one question: where is the sale stalling, and what single change would fix it? Lots of inquiries, few viewings: weak photos or an off-market price. Lots of viewings, no offers: something at the flat doesn't match the listing, or the price doesn't match the condition. The same question keeps coming up: it belongs in the website FAQ and the info sheet. That's the difference between "the flat's just sitting there and I'm waiting" and a managed sale — a self-seller with data knows more after three weeks than plenty of agencies with thirty listings do.

And the final rule of the whole guide: you're selling the biggest asset of your life — hence the system, hence the honesty, hence the lawyer. AI gave you the leverage that used to belong only to professional agencies: one source of truth, derived outputs, tracking, preparation for every conversation. It left the responsibility with you — and that's exactly as it should be.