When a traveler asks ChatGPT, Gemini or Perplexity for an apartment in your city, the assistant does not return a list of results to compare. It compresses searching, browsing, evaluating and deciding into a single exchange, and hands back a short list that is already filtered. Only a few properties get named, which is why organic inclusion is now critical (HotelTechReport). There is no position 1 to 10. You are in the answer or you are not.
Being absent is the normal condition, not a personal failure. Victorious tested 177 brands across eight AI platforms in the first quarter of 2026 and found that only 18 of them registered any mention rate above zero. Close to 90% were simply not there.
The volume behind that absence is not theoretical any more. SiteMinder's Changing Traveller Report 2026 puts around 4 in 10 travelers using AI somewhere in their planning, and 8 in 10 wanting AI help while they book. AI travel agents were handling an estimated 3 to 5% of hotel bookings in major markets in early 2026, up from close to zero in 2024, with a projection of 10 to 15% by 2028 (BookingWhizz).
How do I get my apartment to show up when someone asks an AI for a place to stay?
Work on what other people publish about you before you touch your own website. Muck Rack's May 2026 study of more than 25 million cited links found earned media supplying 84% of all AI citations across ChatGPT, Claude and Gemini, while paid and advertorial content accounted for 0.3%. Ahrefs, across 75,000 brands, found branded web mentions correlating 0.664 with inclusion in AI answers against 0.218 for backlinks. In lodging specifically, a Cloudbeds analysis found OTAs supplying more than 55% of AI citations and hotel websites barely 6%, the rest coming from travel blogs, Reddit, TripAdvisor and YouTube. For an apartment, the OTA listings, the reviews, the directories carrying your address and any video of the place are effectively the whole corpus. Make those accurate and specific first. An assistant that never meets your apartment outside your own domain has nothing to weigh. Your website comes sixth, not first.
There is one study that did look at rentals specifically, and it sharpens the problem. CraftedStays analysed 160 AI responses across 40 discovery prompts tied to five vacation rental brands, on ChatGPT, Perplexity, Gemini and Claude, published in May 2026. Operators were cited 97% of the time when the traveler used their name, and only 10 to 15% of the time on unbranded trip planning questions. They called it the discovery funnel cliff. The assistants already know who you are. They just do not bring you up when nobody has mentioned you, which is precisely when the booking is decided. Mention rates also varied almost threefold by engine, with Claude highest at 55% and Gemini lowest at 32.5%.
One note on all of this before the steps. The Muck Rack, Ahrefs and Victorious figures are cross sector, the Cloudbeds split was measured on hotels, and CraftedStays covers five brands and 160 responses, which is a small sample of management brands rather than individual apartments. The direction is consistent across all of them. The exact percentages are borrowed.
Does this work for a small operator, or only for brands?
It favours the small operator, for two separate reasons. The first is that the field is close to empty: in the Victorious data only 18 brands out of 177 had built any AI presence at all, so you are competing against a handful of incumbents in your market rather than against the whole category. The second comes from the Princeton study "GEO: Generative Engine Optimization" (Aggarwal et al., KDD 2024), which tested nine content tactics over 10,000 queries. Adding concrete statistics raised visibility 41%, quoting credible third parties 28%, and clean readable prose another 28%. Citing your sources raised citation by 115.1% for a page sitting around position 5 in Google. The counterintuitive part: when every source in a query optimizes at once, the page that was number 1 can lose up to 30.3% of its share while low ranked pages gain heavily. Generative search redistributes.
The process
1. Audit yourself in three levels, in a clean room. Assistants personalize by history, memory and location, so auditing from your usual account produces flattering nonsense. Use a private window, switch off memory and personalization, write the city and the language inside the question instead of trusting geolocation, and open a new conversation for each question. Confirm the room is clean by asking the assistant whether it has any information saved about you. Then run three levels. Level 1, the market, without naming yourself: apartment for a family near the convention center with parking. Level 2, the competition: what alternatives are there to the listing that just came up. Level 3, direct: what do you know about this property. The reading is diagnostic. Absent at level 1 is a presence problem. Present but low is a priority problem. Present only at level 3 means the assistant knows you exist but associates you with nothing a traveler actually asks for. Close every conversation with the same question: which sources did you use for this? Those links are your work list.
2. Write the listing as facts, including the ones you assume are obvious. The average ChatGPT prompt is 23 words against the 3 or 4 of a Google search (HubSpot). The traveler is handing over constraints, and an assistant can only match constraints it can read: neighborhood, walking minutes to the landmarks guests actually name, bedrooms, sleeping capacity, parking, check-in method, minimum stay. The harder half is what you never wrote down. If the listing never states that check-in is self service and available at any hour, you will not come back for late arrival queries, however true it is. An assistant fills silence with assumption. And drop the adjectives: unsupported superlatives like stunning or unique read as noise to a model and match nothing anybody asked for.
3. One identity, everywhere, identical. The same name, address and phone on every OTA listing, in every directory and on every page you control. This is entity recognition rather than tidiness: the model has to be certain the apartment in one source is the apartment in another, and contradictions tend to get dropped rather than resolved. One correction here, because most advice gets it wrong: an individual rental generally cannot hold a Google Business Profile at all. Google's eligibility rules exclude "rental or for-sale properties such as vacation homes, model homes, or vacant apartments", while allowing that "sales or leasing offices, however, are eligible for verification". A management business with a real staffed address can have one. A single apartment cannot. If you manage several units, do not pool them onto one page either: one page per property, with its own address block and its own map.
4. Reviews that name attributes. Assistants read review text to learn what a property is for, and by 2026 the content of the review matters more than the star rating. When guests repeatedly write "walk to the beach" or "quiet at night", the model learns to recommend you for those trips. Decide in advance which three attributes you want the apartment associated with, and ask for them by name in the post stay message, inviting guests to mention what they actually used. Do not manufacture any of it. Google is explicit that chasing inauthentic mentions works against you.
5. Put the apartment on video. This is the step almost nobody takes, and it carries the strongest single signal in the data: in the Ahrefs work across 75,000 brands, YouTube mentions correlated around 0.737 with AI visibility, higher than any other factor measured. Assistants index the transcript, not the cinematography, so a phone walkthrough is enough. What matters is the title phrased as the question a traveler asks, a description over 300 words carrying the same facts as the listing, chapters, and captions left switched on. Then embed the video on your own page.
6. The one page you own. Answer the target question in the first 150 words, because 55% of AI Overview citations come from the first 30% of the content (Search Engine Land). Write the headings as the questions people ask, not as labels. Add JSON-LD schema: pages combining clear structure with schema get 2.8 times more AI citations (AirOps), and while 72.6% of first page Google results use schema, only around 30% of sites implement it. Use the right type: not Hotel and not the generic LodgingBusiness, but VacationRental, which Google published as dedicated structured data in December 2023 for exactly this case. Its required properties are name, identifier, latitude and longitude to five decimal places, at least eight images and an occupancy block. The Google rich result for it is gated behind Hotel Center, so the markup alone will not put a card in Google's results. Implement it anyway, for the same reason FAQ markup is still worth writing: Google has all but retired the FAQ rich result, but ChatGPT, Claude, Perplexity and Gemini still read that markup, and structured facts are the most legible form your data can take.
7. Technical access, and here the engine matters. ChatGPT's search runs on the Bing index, so submit your sitemap to Bing Webmaster Tools. Claude leans on the Brave index, so verify the site in Brave Webmaster Tools. Both are free. Keep robots.txt permissive for GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot and Google-Extended: blocking them protects you from training and removes you from citation in the same move. Check the CDN separately, because Cloudflare blocks AI crawlers by default on many sites without the owner ever being told. The test takes a minute: ask an assistant about something you know is written on your page and see whether it answers from inside it. For Google there is nothing extra to do at all. Google Search Central states that optimizing for AI Overviews and AI Mode is still SEO, and that Google Search ignores llms.txt files. Anyone selling you a separate GEO package for Google is selling you SEO twice. One caveat worth carrying into the measurement: ChatGPT cites its sources in 96% of answers, Gemini in 82% and Claude in only 55% (Muck Rack), so silence in Claude tells you less than silence in ChatGPT.
Three mistakes that waste the month
- Starting with the website. It is the instinct, and for lodging it is the wrong end. Most of what the assistant knows about you was written by somebody else, somewhere else.
- Publishing a page for every prompt variation. Google classifies that as scaled content abuse. One good page per intent, not a page per phrasing.
- Buying the mentions. A sponsored slot in the listicles that assistants cite looks like a shortcut, and paid or advertorial content accounts for 0.3% of AI citations (Muck Rack), against 84% for earned media. Google is also direct that the benefit of link spam is lost and not recovered. The money buys you the 0.3%.
How you know it worked
- The same questions, same wording, once a month, four assistants, the three levels. Record whether you were named, in what position, how you were described and which sources were cited. Sentiment counts as much as presence: a flat, neutral description means your listing is saying nothing worth repeating.
- Not traffic. AI sends few visits, but they convert far better than search: Seer Interactive and First Page Sage put ChatGPT around 14 to 16% and Claude as high as 16.8%, against 1.76% for Google organic. A fall in visits is not the alarm. Enquiries are. If traffic halves and bookings hold, what changed is the kind of traffic, not the business.
- In GA4: Reports, Acquisition, traffic acquisition, source/medium, and search for chatgpt, perplexity, claude and gemini. Then look at the key events, not the sessions. ChatGPT has appended utm_source=chatgpt.com since June 2025, which makes the attribution straightforward.
- Diary a quarterly update. Pages left untouched for a quarter lose AI citations roughly three times faster.
None of this work is clever. It is writing down what is true about the apartment in a form a machine can read, then making sure the same thing is said in every place it gets said, and getting other people to say it too. That is unglamorous, and it is exactly why the gap is worth going after: with nine out of ten brands absent from AI answers, it is open to anyone willing to be precise.
This content was created with the help of artificial intelligence and reviewed by Rafael Osborne (Profit Guest Services) before publishing.
