Getting Your Property Listings Into AI Answers: A Practical Guide for Agencies

Published on: September 1, 2026

Last reviewed: September 2026. This is a fast-moving area and the specifics below carry a shelf life measured in months. Where a claim is contested, it is flagged as contested.

Quick answer:

• Check your firewall before anything else. The most common cause of invisibility is a CDN or WAF rule silently blocking legitimate AI fetchers, not a content problem.
• Training crawlers and retrieval crawlers are now separate. A blanket block removes you from citations without recovering anything already used in training.
• If the price and description are not in the HTML, your listing cannot answer the question. Server-render the facts.
• Markup buys unambiguity, not a rich result. State the currency and whether the area is gross or usable.
• llms.txt is not a strategy. The evidence for it is currently negative; add it if you like, but not instead of the work above.

A buyer in Toronto who wants a two-bedroom apartment in Alanya no longer opens ten browser tabs. They ask an assistant, get a summary, and open the two or three links it cites. If your agency is not in that summary, you are not in the shortlist, and no amount of portal spend fixes it after the fact.

The uncomfortable part is that most of what is written about optimising for AI answers is either a repackaging of ordinary SEO or a sales pitch for a file that almost nothing reads. This guide separates the two, and is written specifically for property listings rather than for content pages, because listings have technical problems that articles do not.

First, understand the two routes into an answer

Your listing can surface in an AI answer two ways.

Directly. The assistant retrieves your page live, extracts the facts, and cites your URL. This is the outcome you want, because it comes with a link and a click.

Indirectly. The assistant retrieves a portal page, a market report or a directory that happens to contain your listing, and cites that instead. You get the enquiry only if the intermediary passes it on.

Almost everything below is about making the first route possible. The second route is a distribution question, not a technical one, and our guide to international listing syndication covers it.

1. Check that AI systems can actually fetch you

This is the step that most agencies skip and where the most damage is usually found.

The major AI providers now run separate crawlers for training and for search or retrieval. OpenAI runs GPTBot for training and OAI-SearchBot for ChatGPT search. Anthropic runs several bots including ClaudeBot and a separate search crawler. Google separates Google-Extended, which governs AI training use, from Googlebot, which governs Search.

The consequence is significant. The blanket "block all AI bots" advice that circulated in 2023 and 2024 now blocks your eligibility to be cited while doing nothing to recover content already used in training. If you want citations with links, you generally allow the search and retrieval crawlers and make a separate decision about training.

Three things to check today:

• robots.txt. Read it line by line. Someone may have added a blanket AI block, or your CMS vendor may ship one by default.
• Your CDN or WAF. This is where the real enforcement happens, and it is frequently configured by someone other than the person who wrote robots.txt. Many agencies have bot-fighting rules that silently return a challenge page to legitimate AI fetchers. Test it: request one of your own listing URLs with a non-browser user agent and see what comes back.
• Geographic and rate rules. Blocking whole countries or aggressive rate limiting can produce the same result for crawlers arriving from unexpected ranges.

2. Make sure the page contains text, not just a shell

Listing pages are the worst offenders in all of web publishing for this. The common failures:

• The description, price and features are rendered client-side after a JavaScript call, so a fetch of the URL returns an empty template
• The description sits behind a "read more" control that loads content on click
• Key facts live only inside an image or a PDF brochure
• The address is only present inside a map widget
• Price is shown as "price on application"

A retrieval system generally works from what the page returns. If the answer to "how much is it" is not in the HTML, your listing cannot be used to answer that question. Server-render the description, price, currency, location, size and key features, every time.

The same applies to your translated pages. If a language version is generated client-side or is thin relative to the original, it will underperform in exactly the markets where the buyers are.

3. Add structured data, for the right reason

Use schema.org markup on listing pages: RealEstateListing for the page, an appropriate Residence subtype such as Apartment or SingleFamilyResidence for the property, Offer with price and priceCurrency, floorSize as a QuantitativeValue with an explicit unit, numberOfRooms, numberOfBathroomsTotal, address as a PostalAddress, and geo coordinates. Add datePosted and keep it honest.

Be clear about what this does and does not buy you. Google does not publish a real estate listing rich result, so this is not a route to a special search appearance. What it buys is machine-legible, unambiguous facts: a price that is definitely 285,000 EUR rather than a number that might be a price or a service charge, and a size that is definitely 120 square metres rather than square feet. In a retrieval context, unambiguous beats decorative.

Two specific traps in this industry: state whether the area is gross or usable, since the difference between built and usable area is enormous in Turkey, Spain and much of Latin America; and never mark up a price you are not showing.

4. Write pages that can be quoted

An assistant composing an answer needs sentences it can lift a fact from. Long atmospheric paragraphs about lifestyle do not qualify.

• Put a factual summary near the top: type, size, price, location, tenure, condition, key costs
• Use headings that match questions buyers actually ask, in the buyer's own words
• One fact per sentence, with units and currency stated explicitly
• Date the volatile facts: "service charge for 2026," "asking price as at March 2026"
• Include the costs buyers always ask about next: transfer tax, annual charges, community fees, and whether foreigners can buy this property type at all
• Where a restriction exists, state it plainly rather than leaving it out

A listing page that answers "can a foreigner buy this, what will it cost in total, and how long does it take" is far more citable than one that describes the sunset.

5. Feed and inventory hygiene

Retrieval systems reward sources that turn out to be right, and property data is unusually easy to get wrong.

• Remove sold and withdrawn listings promptly, or mark them clearly as sold. A stale listing that a buyer discovers is gone damages trust in everything else you publish.
• Set a canonical page per property. If the same apartment exists at three URLs because of filter parameters, agency branches or language variants, consolidate the signal with canonical tags and hreflang.
• Use a stable reference ID and keep it consistent across your site, your feeds and your portals, so the same property can be recognised as the same property.
• Keep prices synchronised. The same flat listed at three prices across three portals is the most common credibility failure in international real estate, and it is very visible when a system compares sources.
• Publish the update date on the listing page and mean it.

6. The entity layer, which is what actually earns citations

Retrieval systems prefer sources they can corroborate. For an agency that means:

• A real About page with named people, roles and photographs, not stock imagery
• Your licence or registration number, and the body that issued it, stated on the site
• Consistent name, address and phone number everywhere you appear
• A physical address that resolves
• Corroboration elsewhere: portal profiles, industry directories, chambers of commerce, press mentions, and marketplaces where your listings appear alongside an agency profile
• Author attribution on any market commentary you publish

A single-page site with a contact form and no named human is a poor candidate for citation regardless of how well its markup is written.

7. llms.txt, honestly

The llms.txt proposal is a Markdown file at the root of your site listing your most important pages for AI consumption. It is frequently sold as the new robots.txt.

The evidence does not support that framing. A study of 300,000 domains found adoption around 10%, and among the most frequently cited domains in AI answers almost none had the file. Analysis of server logs across hundreds of millions of bot events found only a few hundred requests for /llms.txt, with crawlers overwhelmingly fetching ordinary HTML instead. Google has stated publicly that it does not use the file, and lists it among unnecessary tactics in its generative AI guidance. No major provider has committed to reading it from third-party domains.

The reasonable position: it costs an hour to add and does no harm, so add it if you like, but do not let it displace any of the six sections above. If an agency or consultant is selling llms.txt implementation as an AI visibility strategy, that is the tell.

8. Decide the training question deliberately

There is an economic dimension that publishers care about more than agencies do, and it is worth understanding because it drives a lot of the advice you will read.

Cloudflare's data for the first quarter of 2026 put the ratio of pages fetched to human visitors referred at roughly 1,276 to 1 for GPTBot and roughly 23,951 to 1 for ClaudeBot. For a site funded by advertising, that asymmetry is close to existential, which is why the publisher industry has moved towards blocking and metering, and why Cloudflare has run a pay-per-crawl experiment.

An estate agency is not in that position. You do not monetise page views; you monetise enquiries. One qualified buyer who arrives through a citation is worth more than tens of thousands of unmonetised crawls cost you. For most agencies the sensible policy is to allow search and retrieval crawlers, and to treat the training question as a separate preference rather than a revenue decision.

9. Measure it, or you are guessing

• Server logs by user agent. This is the only source that tells you what actually happened. Segment GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended and Googlebot separately.
• Referral traffic from assistant domains in your analytics.
• Branded search volume, which tends to rise before referral traffic does.
• Manual prompt testing. Run the twenty questions your buyers actually ask, in each of your target languages, monthly, and record which sources are cited. Answers vary by country and by language, so test from the markets you sell to, not only from your own office.

10. The checklist

1. robots.txt permits search and retrieval crawlers
2. CDN and WAF rules do not silently block them
3. Listing content is server-rendered, including description and price
4. Prices are shown, with currency, not "on application"
5. schema.org markup on every listing, with explicit units
6. Area stated as gross or usable, unambiguously
7. Factual summary block near the top of each listing
8. Sold and withdrawn listings removed or marked, promptly
9. Canonical URL and hreflang set correctly across languages
10. Named people, licence number and consistent contact details on the site

Frequently asked questions

Is this different from SEO?
Mostly it is SEO done properly, plus access control and structured data. The genuinely new parts are the crawler split and the emphasis on machine-extractable facts over persuasive prose.

Should I block AI crawlers to protect my listings?
Your listings are already distributed to portals whose whole business is republishing them. Blocking retrieval crawlers removes you from answers without protecting anything you were not already syndicating.

Do I need a separate site for AI?
No. Anyone selling you one is selling you a doorway site with a new name.

How long until this shows results?
Access fixes can change what crawlers see within days. Citation behaviour changes over weeks to months, and depends heavily on whether your entity is corroborated elsewhere.

What matters most if I only do one thing?
Check that your firewall is not blocking the crawlers. It is the most common problem and the cheapest to fix.

Keep reading on JanusHermes

The order of work is the point: access first, then server-rendered facts, then markup, then the entity layer. Everything else in this field is decoration on top of a page that a retrieval system either can or cannot read. JanusHermes publishes listings in eleven languages, with a named agency profile and direct contact details on every listing. If you want your inventory readable by both buyers and the systems they now ask, see JanusHermes for agencies.

Related reading: International Listing Syndication for Agencies, Small Agencies vs Portals, Cross-Border Lead Routing, The Best Property Portal in Every Country, How Buyers Use AI to Find and Vet Property Abroad and Built vs Usable Area in Foreign Listings.

Contested points flagged above: llms.txt effectiveness, where the evidence is currently negative, and whether specific providers read specific files, which none has confirmed for third-party domains.

This guide is general information as of September 2026. Crawler names, provider policies and platform behaviour in this area change quickly; verify the current position with each provider's own documentation before making access decisions.