ChatGPT has no ranked list of results and you cannot buy your way into its answers. Getting your property quoted needs two things. Your website has to let OpenAI read it, and one of your pages has to answer a specific guest question better than any other site does. Ask your web person to allow OAI-SearchBot, then take a question your guests really ask and answer it plainly in the first two sentences of a page. OpenAI, Google and Microsoft each publish what they need from you, and all three ask for far less than the checklists circulating this year.
What the three companies actually require
OpenAI, Google and Microsoft each publish guidance for site owners, and this is what all three of them say in their own words.
| Company | What it says you have to do | What it says you can skip |
|---|---|---|
| OpenAI | Let OAI-SearchBot read the site. OpenAI publishes the addresses its crawler uses so your web host can stop blocking them, and says a site opted out of that crawler "will not be shown in ChatGPT search answers" | Nothing about markup, special files or page formats appears anywhere on the page. GPTBot is a separate crawler for model training, and blocking it does not remove you from ChatGPT search |
| Be indexed and eligible to appear in ordinary Search with a snippet. Publish writing that comes out of your own experience | Special files, markup, chunked text, rewriting for machines, and buying mentions. Google names all five as things you can ignore | |
| Microsoft | Clear headings, tables and FAQ sections, content kept fresh and accurate, and claims supported with examples and data | Microsoft publishes no list of requirements beyond those, and respects whatever your robots.txt says |
Checked on September 2026 against each company's own documentation.
Access is the only hard technical requirement any of the three publishes. Most of the rest of the advice on this subject is restating Google's ordinary search rules, which have not changed in years, and some of it is a theory an agency invented.
What you can stop doing
Google runs a section of its own guide called mythbusting, and it names five practices you can safely ignore. It does not read the special text files people now write for AI systems, llms.txt among them, and says keeping one will neither help nor harm you. It sets no required page length and handles several topics on a single page, so there is no need to chop your writing into small pieces. It understands synonyms and meaning, so you do not have to write in some special voice for machines. It counts mentions of your name bought across the web as less useful than they sound. And it states that structured data is not required for its generative features, with no special schema to add.
Keep your structured data anyway. It still earns the star ratings and prices that show under an ordinary Google result, which is a real return in its own right.
One caution about scope, because most articles on this search get it wrong. That mythbusting list is Google describing Google. OpenAI and Microsoft have published no equivalent list, so for ChatGPT and Copilot nobody knows whether those same practices do anything. Nobody selling an AI visibility package can show you a vendor page that says otherwise. Google adds a warning of its own here: no outside tool has access to its ranking or AI systems, so be careful with any product promising ranking success.
The five things that actually go on the page
Put the answer in the first two sentences
An assistant does not search for the question your guest typed. It writes several narrower questions of its own, runs them all at once, and quotes whichever page answers one of them most directly. Google describes exactly that behavior in its own guidance for site owners. So if your page opens with a welcome paragraph about your warm hospitality, the answer sits four hundred words down and something else gets quoted instead.
A page about parking at a fictional inn shows the whole pattern in five lines:
Where do I park at the Ashford Inn?
Guests park free in the courtyard behind the building, which holds nine cars and is reached from Mill Lane. A truck with a trailer will not fit through the archway, so we hold two spaces on the street for anyone arriving with one.
The heading is the question a guest types. The two sentences under it settle the matter, including the awkward case. Everything else about parking goes below.
Write the page only you could write
Google contrasts generic writing with writing that comes out of somebody's own experience, and it uses two article titles to make the point. On one side, "7 Tips for First-Time Homebuyers". On the other, "Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line". Google prefers the second one because only the person who lived through it could have written it, and it says that kind of firsthand knowledge counts for more than anything else in its guide.
Your version of the second one is sitting in your inbox already. Whether the lift reaches the top floor and what you do when somebody arrives with a wheelchair. Which rooms hear the market being set up at five. How long the walk to the station honestly takes with a suitcase, which is never what the map says. What a family with a wet dog and a wet Tuesday can do within ten minutes of your door. No travel platform can write a word of that, because they have a whole city to cover in the space you can spend on one street.
Keep every fact in the text of the page
Write your prices, your check in window and your exact address out as words a visitor can read, along with anything else a guest asks before booking. None of it should sit inside a photograph, and none of it should appear a second after the rest of the page, the way a price sometimes flickers in once everything else has loaded. If a floor plan or a rate card currently exists only as an image, type what it says into a paragraph directly beneath it and leave the picture where it is. Google asks that its systems be able to read the page, and Microsoft asks for clear headings, tables and questions with answers underneath, so build the page so that everything a guest needs is already there when it opens, before anybody clicks or scrolls.
Write sentences that survive being lifted out
An assistant quotes a sentence or two, stripped of everything around it, so "It is included in the rate" arrives meaning nothing while "Breakfast is included in the rate in all nine rooms" still tells a reader what they needed to know. Name the subject inside the sentence instead of leaving it up in the heading. Microsoft asks publishers to support their claims with examples and data, which works the same way: a number inside the sentence gives an assistant something exact to repeat.
Date it and sign it
Put the month and year on anything that can go stale, and a real name at the top. Microsoft asks publishers to keep content fresh and accurate and offers IndexNow to tell it when a page has changed. A page that says when it was last checked gives an assistant a reason to prefer it over one that does not.
Let the crawler in, and check the firewall too
Give this ten minutes. Type your own web address followed by /robots.txt into a browser and read what comes back.
| What you may find | What it means |
|---|---|
User-agent: OAI-SearchBot followed by Disallow: / |
ChatGPT search cannot show your pages. Change it today |
User-agent: GPTBot followed by Disallow: / |
You are opted out of training only. ChatGPT search is unaffected |
User-agent: ChatGPT-User followed by Disallow: / |
Blocks the fetch ChatGPT makes when a user clicks through to you |
User-agent: Google-Extended followed by Disallow: / |
A training choice for Gemini. Google states it does not affect inclusion in Google Search and is not a ranking signal |
| No mention of any of them | Everything is allowed, which is what you want for OAI-SearchBot |
Checked on September 2026 against the OpenAI and Google crawler documentation.
Robots.txt is only half of it. Plenty of hosting plans and security plugins block anything that is not a recognized browser, and that block happens before robots.txt is ever read. OpenAI publishes the addresses its search crawler uses, as a list at openai.com/searchbot.json, so a host can let them through. Send your web person or your hosting support desk one line: please confirm OAI-SearchBot is allowed in our robots.txt and that the addresses listed at openai.com/searchbot.json are not being blocked.
What our own record shows
Maxine and I co-own MINT @Naschmarkt, 18 serviced apartments in five apartment types in Vienna, and my team at Ekamoira built its website, which is how Fullstreak started. We applied every step above to those pages.
Since 21 July 2026 the site has carried a piece of tracking code we wrote, which remembers the source a visitor first arrived from and keeps it attached to them through every later visit. Over its first four weeks, 51 people left an email address through a form on the site. Four of them had first landed from an assistant, three from ChatGPT and one from Claude. That code follows each of those contacts onward to the booking, so we can read what became of any one of them by name.
Those pages are ordinary. Each one takes a question a guest asks us and answers it in the first two lines, with the detail underneath.
How to tell whether any of it worked
Start by asking the assistant yourself, which costs nothing and beats any paid visibility tool for this purpose. Open ChatGPT and type the exact question your page answers, in the words a guest would use. Read what comes back and follow the links. If a big travel site is quoted and you are not, your page has not beaten it on that question yet.
Look at the reports the platforms give you. Microsoft's AI Performance report inside Bing Webmaster Tools lists which of your pages were cited across Copilot and the AI summaries in Bing, and shows the phrases the assistant actually searched before it quoted you. Since June 2026 it also shows how much of the citation space on each of those phrases went to you rather than to everybody else answering it. Google's Search Console has a matching report for its own AI features.
Read your server logs. You do not need access yourself: email your hosting support desk and ask for the raw access logs for the last ninety days for that one page address, including the name each visitor reported itself under. Then look down the list for OAI-SearchBot. If it is not there, the crawler has never read the page, which means nothing on it can be quoted. Take that finding back to your host and ask them to unblock the addresses before you touch a word of the writing.
Our AI visibility page for hotels covers the wider picture, including where each assistant gets the property names it uses.
Who should leave this alone for now
Your booking step has to work before any of this earns anything. When an assistant sends somebody to a checkout that fails, you lose the booking and the guest goes back to the platform. Book three nights on your own site with your own card first, and fix whatever that turns up.
If your Google Business Profile is half filled in, start there instead. In my experience it does more for a property of eight to twenty five units than anything else you can finish in a week, and it feeds the hotel listings the assistants read. The full order of work, cheapest first, puts this page's advice where it belongs, which is after the free steps and before paid clicks.
This also needs somebody who will keep at it, so do not start if nobody at the property can. Microsoft asks publishers to keep their content fresh and accurate, which matters at a property because rates and local details change every season. An assistant reading two pages on the same question will take the one that has been corrected recently over the one last touched in March.
If you would rather have somebody read your robots.txt, your logs and your top pages with you, tell us your address and we will look.
