AI visibility

How to get your business recommended by ChatGPT

Content Autopilot team · Updated 14 November 2026 · Sources checked 14 November 2026 · 6 min read

ChatGPT recommends businesses it can find through its search index, read without running JavaScript, and see described on other sites. So the work goes in that order: let its search crawler in, publish pages that answer the questions your buyers ask, and get mentioned on the pages it already reads. Nobody can guarantee a mention, including us, and you will not see a result for weeks.

In short

  • Check that OAI-SearchBot is allowed in your robots.txt. OpenAI says a change takes about 24 hours to register.
  • Write one page per buyer question, with the answer in the first paragraph and specifics a careful buyer wants: prices, limits, who it is not for.
  • Find the pages AI answers cite for those questions and get listed on them. In our check of Google's AI Overviews, comparison pages made up over half the citations for "X or Y" questions.
  • Leave llms.txt, hidden text and mass-published pages until last, or out. None has evidence behind it, and the last two have done harm in tests.
How ChatGPT decides who to recommend. 38 seconds, no sound.
Read the text of this video
  1. How ChatGPT decides who to recommend, and the four things you can change.
  2. A buyer asks: “What invoicing software works for a two-person agency billing in euros and dollars?” That is longer and more specific than a Google search.
  3. Two sources feed the answer. What the model learned is fixed months before the question, and you cannot change it this year. Live search works in three steps: OAI-SearchBot indexes the pages it may read, ChatGPT searches and opens a handful, and it cites some and names businesses. Only the live search path is yours to change.
  4. Step 1: let its search crawler in. If robots.txt disallows OAI-SearchBot, OpenAI says you will not be shown in ChatGPT search answers. Allowed, your pages can be read and cited. Blocking GPTBot is different: it stops training, not citations.
  5. Steps 2 and 3: be on the pages answers are built from. Of the 356 pages Google's AI Overviews cited for 48 buyer questions on 14 November 2026, 29% were articles and guides, 22% YouTube videos, 16% list articles, 13% comparison pages and 12% forum threads. Answer the exact question, near the top. Then get listed where answers already look.
  6. The four steps, in order: let OAI-SearchBot in, answer the questions buyers ask, get mentioned where answers look, and say the same thing everywhere. Nobody can guarantee a mention. Expect weeks, not days.

How ChatGPT decides who to mention

An answer has two sources. One is what the model learned in training, which is fixed months before anyone asks and which you cannot change this year. The other is live search: when a question needs current information, ChatGPT searches, reads a handful of pages and writes its answer from them. That second path is the one you can affect.

1A buyer asks

“What invoicing software works for a two-person agency billing in euros and dollars?”

2What the model learned

Training data, fixed months before the question. You cannot change it this year.

3Live search

  1. OAI-SearchBot has indexed pages it was allowed to read.
  2. ChatGPT searches that index and opens a handful of pages.
  3. It quotes some of them and links them as sources.

4The answer

Names a few businesses. They come from the pages it read: your own, and the comparisons, reviews and forums that mention you.

Two sources feed an answer. Only the live-search path can be changed by what you publish this year.

The search index is built by a crawler called OAI-SearchBot. OpenAI's crawler documentation says sites that opt out of it "will not be shown in ChatGPT search answers", though they can still appear as plain navigational links. Everything else in this guide assumes you have not done that.

Step 1: make sure it can read you

Three things can stop ChatGPT before your content matters.

Robots.txt blocks the search crawler. For a business site this is nearly always an accident: a blanket rule, or a plugin that promised to "block AI bots" and blocked all of them. When we read the robots.txt of 60 well-known sites on 14 November 2026, 56 could be read and 8 of those blocked OAI-SearchBot. All 8 were large publishers and platforms that had decided to. If you sell something, you want it allowed.

To allow the search crawler while keeping out the one that collects training data, two groups are enough:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

Leave out the GPTBot group if you are happy for your pages to be used in training; it has no effect on being cited. If your file keeps crawlers out of private folders under User-agent: *, repeat those lines in the OAI-SearchBot group, because a crawler with a group of its own ignores the rules for everyone. OpenAI says a change takes about 24 hours to take effect for search.

Your text is loaded by JavaScript. A page that builds its content in the browser can look complete to you and arrive almost empty to a crawler that does not run scripts. When Vercel measured this in December 2024, none of the major AI crawlers ran JavaScript. That measurement is two years old, so treat it as the safe assumption rather than a law.

A firewall refuses the crawler. Bot protection can turn away a crawler that robots.txt welcomes. Check your host's or CDN's bot settings.

The checker below tests the first two in one go. It cannot see your firewall.

Try it on your site

Or try one:

Step 2: answer the questions people ask it

People ask an assistant longer and more specific questions than they type into Google. Not "accounting software" but "what accounting software works for a two-person agency that invoices in euros and dollars". The page that gets used is one that answers that question, close to the top, in plain sentences.

So write one page per question. Open with the answer. Then give the details a careful buyer wants and marketing pages leave out: what it costs, what the limits are, who should pick something else.

Evidence helps. In the research paper that coined "generative engine optimization", the changes that raised a page's visibility most were adding citations, quotations from relevant sources and statistics. The abstract puts the gain at "up to 40%". The main test ran on a simulated engine rather than ChatGPT itself, so read it as a direction, not a number to plan around. In the same tests, stuffing in keywords did not help.

To find the questions: ask ChatGPT five to ten things a buyer would ask before choosing a product like yours, with search switched on, and write down who it names and which pages it links. Then read your sales emails and support inbox for the questions people really send.

Step 3: get mentioned where AI answers already look

Open the pages that come up as sources in step 2. Those are pages an assistant already trusts for your question, and being listed on one is usually quicker than writing something to outrank it.

What kind of page they are depends on the question. On 14 November 2026 we searched 48 buyer questions on Google and sorted the 356 pages its AI Overviews cited:

Share of 356 cited pages, by type

  • Article or guide29%
  • Video22%
  • List article16%
  • Comparison13%
  • Forum or community12%
  • Other3%
  • Review or listings site3%
  • Help or documentation1%
  • Home page1%
  • Product or service page0%
Show as a table
Article or guide29%
Video22%
List article16%
Comparison13%
Forum or community12%
Other3%
Review or listings site3%
Help or documentation1%
Home page1%
Product or service page0%
Pages cited by Google's AI Overviews for 42 of 48 buyer questions, United States, 2026-11-14. Types come from fixed rules on each page's address and title. Source: our data
  • "X or Y" questions were answered from comparison pages: 44 of the 81 pages cited for the eight comparison questions were comparisons. If buyers weigh you against a competitor, a fair comparison page is the page to have, and to be in.
  • "How do I" questions drew heavily on video: 31 of 88 citations were YouTube.
  • Local service questions cited the guides of individual plumbers, electricians, movers and photographers next to large sites. A small firm that writes the page answering its customers' question gets a look.
  • Forums made up 42 of the 356 citations, mostly Reddit and Quora.

That is Google's AI Overviews, not ChatGPT. For ChatGPT the best public figures come from Outrank, which reran 48 questions through ChatGPT in October 2026: it read about 36 pages per answer and cited about 4, and only 17% of what it cited was blog content. Outrank found it leaned on brands' own product pages and documentation. So for ChatGPT in particular, your own pricing, product and help pages need to say plainly what a buyer would ask.

For each kind of source:

  • A comparison article or list: email the author with one specific reason you belong on it.
  • A review site: claim your profile and ask recent customers for reviews.
  • A forum thread: answer the question properly, under your own name, and say where you work.

Step 4: say the same thing everywhere

An assistant can only repeat what it reads. If your pricing page says one thing and a two-year-old directory listing says another, it cannot tell which is current. Go through the places that describe you (your own site, LinkedIn, review profiles, directories, app stores) and make the name, the one-line description and the pricing match.

What to leave until last, or out

Otterly, a company that sells AI visibility tracking, publishes its experiments including the ones that failed. Its list of tactics that showed no effect in its tests is worth reading in full. Three stand out:

  • An llms.txt file. On Otterly's test site, 84 of more than 62,100 AI crawler visits in 90 days went to the file. Ahrefs later found that 97% of valid files on 137,210 domains got no requests in a month. It does no harm and takes ten minutes, so it belongs at the end of the list, not the start. Our guide covers it.
  • Hidden text or instructions aimed at AI. No assistant followed them in Otterly's test, and Copilot flagged the page as unsafe.
  • Mass-publishing AI articles. Two test sites with 1,000 posts each were removed from Google's index, one within 8 days and one within 12 weeks.

That last one deserves a plain statement, because we sell a product that publishes an article a day. Volume is not what gets a business cited. A researched, checked article a day on a question buyers ask is a different thing from a thousand pages at once. The useful test for any page is whether it would be worth reading if search engines did not exist.

How to tell whether it is working

Ask the same questions every week and note whether you are named. One check proves little, because the same question can get a different answer an hour later. The free visibility checker runs three buyer questions past each assistant once; How to track AI visibility covers doing it properly over weeks.

We do not have our own figures yet for how long this takes, and we would be wary of anyone who quotes a number of weeks.

Questions

Do the same steps work for Claude, Gemini and Perplexity?
Yes, with different crawlers. Claude's search index is built by Claude-SearchBot and Perplexity's by PerplexityBot. Gemini and Google's AI Overviews draw on Google's own index, so there the crawler is Googlebot and ordinary SEO applies.
Do I need a blog for this?
You need pages that answer questions. A blog is the usual place for them, but a help centre, a comparison page or a detailed pricing page does the same job. For ChatGPT, product and pricing pages matter more than for Google.
Will blocking GPTBot keep me out of ChatGPT's answers?
No. GPTBot collects training data. Search answers come from OAI-SearchBot, and OpenAI says the two settings are independent.
Can I pay to be recommended by ChatGPT?
Not in its answers. What you can do is be mentioned on the pages it reads, which is where most of this guide's effort goes.

Sources

Each one was opened and read on 14 November 2026. Figures in the text say what was counted, how many and when.

  1. 1.crawler documentation (/api/docs/bots) · developers.openai.com
  2. 2.measured this (/blog/the-rise-of-the-ai-crawler) · vercel.com
  3. 3.research paper (/abs/2311.09735) · arxiv.org
  4. 4.Reran 48 questions · outrank.so
  5. 5.Its list (/blog/how-to-appear-in-ai-search-engines) · otterly.ai

Content Autopilot team

We make Content Autopilot, an SEO and AI visibility product

Our guides are drafted with AI from the sources listed above, and every figure is checked against its source before it is published. Where we sell something a guide discusses, the guide says so.