The Bottom Line
If you want a real estate team to appear in AI answers, start with the same foundation that makes the team useful in search: public pages that can be crawled and indexed, direct answers to real questions, visible evidence, accurate business details, and internal links that identify the authoritative page for each topic.
There is no schema type, llms.txt file, crawler hit, or page format that guarantees a citation. Google says its normal SEO guidance still applies to AI Overviews and AI Mode, with no special AI markup required. OpenAI says inclusion in ChatGPT search depends in part on allowing OAI-SearchBot, but it does not promise placement.
That changes the goal from "use an AEO trick" to "be the clearest, best-supported source for a question that matters."
The Questions a Real Estate Site Should Own
Do not begin with a quota of blog posts. Begin with the questions buyers, sellers, and recruits repeatedly ask.
| Question class | Authoritative page | Evidence the page should contain |
|---|---|---|
| "Who is a strong listing agent in [city or neighborhood]?" | One indexable agent or team page | License and brokerage details, service area, specialties, recent work, attributed reviews, and links to the same public profiles |
| "What is it like to buy or sell in [neighborhood]?" | One maintained neighborhood page | Local trade-offs, housing stock, current market facts, original photos, nearby areas, and a named author |
| "What happened in the [market] housing market?" | One dated market report | Source window, methodology, MLS or public-data attribution, author, and the date it was last checked |
| "Has this agent sold a home like mine?" | Permanent listing or case-study page | Property type, location, marketing work, result, dates, and clear labels for verified versus team-reported claims |
| "How does this real estate service work and what does it cost?" | Service and pricing pages | Scope, exclusions, process, price, ownership, and links to proof |
Each question class needs one source of truth. Ten thin pages that disagree are weaker than one complete page that other pages consistently reference.
The 7-Point Framework
1. Make the important answer crawlable and indexable
Put the name, market, service, proof, and answer in the page's text. Keep the page public, return a successful response, allow crawling, include it in the sitemap, and link to it from another relevant page.
Server-rendered or statically generated text is a dependable cross-provider baseline because the answer arrives in the first HTML response. That does not mean every AI or search crawler is incapable of rendering JavaScript. The practical test is simpler: fetch the page without a browser and confirm the important text is already present.
Crawler roles also matter. OpenAI documents OAI-SearchBot for search discovery and GPTBot for potential model training controls. Perplexity separately documents PerplexityBot for search and Perplexity-User for user-directed fetches. A request from one of these user agents is an access signal, not proof that the page was cited, recommended, or clicked.
2. Give each question one clear answer page
Open with a short answer that stands on its own. Then add the details a buyer or seller would need to verify it: dates, definitions, locations, sources, limitations, and links to deeper proof.
Use descriptive titles and headings. "Granite Bay seller guide: timing, preparation, and local trade-offs" is more useful than "Our Services." A concise answer at the top helps a person scan the page and gives a retrieval system a clear passage to evaluate.
3. Use structured data to describe visible facts
Structured data helps search systems understand what a page describes. It does not create authority by itself.
Use the most specific supported type that truthfully matches the page, such as Organization, Person, Article, BreadcrumbList, or an applicable real-estate type. Connect stable entities with @id and use sameAs only for profiles that represent the same person or business.
Google's structured-data rules require the markup to represent content people can see and explicitly say that valid markup does not guarantee a search feature. Treat schema as an accuracy and extraction layer, not a ranking claim.
4. Make the entity consistent across the open web
The agent's name, brokerage, license information, service area, phone, and primary domain should agree across the owned site and current public profiles. Resolve meaningful conflicts instead of creating more profiles.
The site should remain the source of truth for the full explanation. Third-party profiles can corroborate identity, but they should not be the only place where specialties, market knowledge, or proof exist.
5. Publish demand-led local evidence
Create local pages when first-party evidence shows a recurring question: client conversations, internal site search, Search Console queries, audit findings, or repeated verified crawler requests to a missing path.
Do not turn every city, ZIP code, or 404 into a generated page. A 404 can reveal demand only when the response status is known, the requesting bot is trustworthy, and the pattern repeats. User-agent text alone is not enough.
For market reports, show the reporting window, source, calculation, author, and last-reviewed date. For reviews, use the reviewer's permission, preserve attribution, and keep the visible copy aligned with any review markup.
6. Support claims with primary evidence
AI answers are assembled from sources, so unsupported superlatives are fragile. Prefer first-party results, official records, current local data, and primary research. Label estimates and team-reported observations.
The peer-reviewed GEO study by Aggarwal and colleagues reported visibility gains of up to 40% on its benchmark and found that results varied by domain. That supports testing clear citations, quotations, and statistics. It does not establish a guaranteed citation lift for a live real estate site.
Espo's flagship client, Better Brokers Realty Group, gives us a useful observation, not a controlled experiment. Through April 2026, the team reported agents being named in ChatGPT and Gemini answers, and three callers said ChatGPT recommended a BBRG agent. Separately, Google Search Console verified a position of 1.0 for "selling a home in granite bay." Those facts justify continued measurement; they do not prove which site change caused the AI mentions.
7. Measure citations, human visits, search, and crawl access separately
These are four different signals:
| Signal | What it can tell you | What it cannot prove |
|---|---|---|
| Google Search Console | Clicks, impressions, CTR, and position; Google includes AI-feature traffic in the Web search type | Which clicks came from an AI Overview or AI Mode |
| GA4 referral reporting | Human sessions arriving from reviewed AI referrers; OpenAI says ChatGPT search links include utm_source=chatgpt.com | How often an answer mentioned you without a visit |
| Bing Webmaster Tools AI Performance | Citations, cited pages, sampled grounding queries, and trends across supported Microsoft surfaces | Ranking, authority, or placement inside an individual answer |
| Verified server or CDN logs | Whether a known crawler or user-directed fetcher reached a path and what status it received | A citation, recommendation, model identity, or human visit |
Bing's AI Performance documentation is unusually explicit about those limits: citation counts do not indicate rank or page importance, and grounding queries are sampled.
Where llms.txt Fits
llms.txt can be a concise, maintained directory of a site's authoritative pages and current facts. It should not be the primary AEO investment.
Google says sites do not need a new machine-readable AI file or special schema to appear in its AI features. OpenAI's current publisher guidance focuses on OAI-SearchBot access and does not name llms.txt as an inclusion requirement. We keep the file because it is inexpensive to maintain and useful as a canonical summary, while treating the public HTML, sitemap, robots controls, and internal links as the dependable foundation.
If the file drifts from the site, remove or correct the stale claim. A smaller accurate directory is better than a long document full of volatile percentages.
A Small Test Beats a Site-Wide Rewrite
Pick one or two pages with measurable search demand. Record a baseline, then improve the answer structure, evidence, and internal links on those pages only. Freeze at least three comparable pages as controls.
Review the treated-to-control change weekly for six weeks:
- Google impressions, clicks, CTR, and average position
- AI-referred human sessions and engaged sessions
- Bing citations, cited pages, and sampled grounding queries when available
- Indexing, crawl failures, and verified bot response statuses
- The exact page and claim changes made during the test
Pause or roll back if the treatment creates indexing errors, hides important text, introduces claims without a source, or causes a sustained search decline on treated pages while the controls remain stable. If citation and referral volume are too small to interpret, do not manufacture a win; keep the low-cost foundation and wait for a larger sample.
The One Thing to Do Today
Write down the five questions you most want a buyer or seller to ask about your team. For each question, name the one page that should be trusted as the answer.
If you cannot name that page, that is the gap. If the page exists but the answer is buried, unsupported, or disconnected from the rest of the site, that is the first treatment.
Want to see how this looks in a real build? See Website + IDX, review the AI-discoverable website feature, or request a free AI visibility audit.
