How to get recommended by ChatGPT

To be named by ChatGPT you have to be one of the companies its sources name for that question, and you have to give it a page it can read and quote. Nobody can guarantee a place in its answer, and anyone promising one is selling something they do not control. What you can do is make yourself eligible on both of the paths it uses.

What ChatGPT does differently

It does not always search. For many questions it answers from what the model already holds, with no citations at all. For others it searches, reads a few pages and shows its sources. The same question can go either way depending on the mode, the wording and the day.

It also compresses harder than most. The typical answer is a short list of three to five names with one line each, written as prose rather than a table. That one line is the whole of your presence in the answer.

And it is personal. Chat history, saved memory and custom instructions all shape what a given user sees, which means your own account is the worst place to judge how you appear to strangers.

One more thing that is observable and worth knowing: its crawling is split across separately named bots with separate jobs, so your robots.txt can be open to one and closed to another without anybody intending it. Beyond that, how it chooses what to retrieve is not published, and we will not pretend to know it.

What that means for whether you are named

You have two separate jobs. The first is being part of what the model remembers, which comes from having been written about across independent sites over time. This one is slow and you cannot rush it. The second is being findable at the moment it searches, which depends on indexed pages that answer the question and third party pages that list you. This one you can work on this month.

Being describable in one line is the underrated requirement. If there is no short, consistent sentence about what you do and who you serve, the assistant has to write one, and the easier choice is to name a company that already has one.

It also means the order of your work is decided for you. Fix the live path first, because it is the only one with a schedule, and let the remembered path accumulate behind it.

What to change

• Publish one page per buyer question, answer in the first paragraph, specifics with dates underneath.
• Write one description of your company, in the buyer's words, and use exactly that wording on your site, your profiles and your listings.
• Get onto the third party pages ChatGPT reaches for your category: comparison articles, directories, review sites, active forum threads.
• Make your pages readable as plain HTML and check robots.txt allows the crawlers you want. The bots behind ChatGPT's search are named separately from Googlebot.
• State one honest limit on each page. It makes the rest more credible to a careful summariser.
• Keep every fact identical across your site and every profile, so both paths carry the same version of you.

A worked example

One result from our own work shows both paths at once, and half of it can be checked by a reader. clawlaw.in is our sister company. On 27 July 2026 we asked ChatGPT 78 buyer questions about its category, with no brand named, and ChatGPT confirmed afterwards that it had run no live web search for any of them.

Ten days after that baseline, on 6 August 2026, ChatGPT cited a page on clawlaw.in as a source for a vendor due diligence question. The page answered one narrow question, how to check a company's court cases in India, and it had been published fourteen days earlier. At baseline the same question set had named the company nowhere in that zone.

Read the two together and the two jobs on this page come apart cleanly. The remembered path had not moved: the July reading was made of text that already existed. The live path moved in a fortnight, on one narrow question, because a page now existed that answered it.

That is the shape to plan for. Work the live path, because it is the only one with a schedule, and report it separately from the unsearched readings, so a fortnight of real progress is never mistaken for a change in what the model remembers.

What this cannot do

Five limits, stated plainly.

No guaranteed placement. There is no paid route into the recommendation and no agency control over it. Eligibility is the product on sale, including ours.

No influence over what the model remembers this quarter. That changes with training runs whose dates are not announced.

No fix for a category mismatch. If the question asks for something you do not sell, being left out is correct.

No visibility into the ranking logic. Everything above is inferred from behaviour that can be observed from outside. It can change without notice.

No stable single reading. The same question can name you at noon and not at one. Judge by trend across months.

Common questions

Can I pay OpenAI to be recommended?

There is no product that buys a place in the recommendation. Money can buy presence on third party pages that may be read, which is eligibility, not placement. Treat any vendor selling the second thing as selling something they do not have.

Does having a ChatGPT integration or a listing in its app directory help?

It gives you one more page that exists and can be found, which is worth having if you genuinely integrate. It is not a shortcut into recommendations for your category, and it should not displace the directory and question page work.

How do I check without my own account polluting the result?

Fresh chat, signed out or with memory off, or a separate browser profile kept only for this. Never put your own company name in the question. If you have discussed your company in that account for months, it will name you and that tells you nothing.

Why was I named last month and not this month?

Ordinary reasons first: variation between runs, no search happening this time, or a third party page that changed. Check whether the citation list changed before you change anything on your own site.

Does posting on social media help?

Indirectly and weakly. What helps is a page that exists, is indexed and can be read. A post that leads somebody to write about you on a page that gets cited has helped. The post itself rarely is the source.

How long should I give this before judging it?

Take a baseline now and read monthly for a quarter. In our own runs the first citation of a new page came between 9 and 16 days after publishing, and those are the founder's hand run readings rather than a promise about your site.

What to do first

Ask your buyer questions blind, never naming your company, in a fresh chat while signed out. Do it twice: once with search off to see what it remembers, once with search on to see what it can find. Those two readings move at different speeds and it is worth knowing which one changed.

Log the date, the mode, whether you were named, who was named ahead of you, and every URL cited. Then pick the single most cited third party page you are absent from and get onto it this week. Repeat the reading monthly on the same questions.

How we know this

We read the assistants rather than guessing what they do. In our own audit of 1,260 saved answers from ChatGPT and Gemini across the United States, the United Kingdom and India, 82.0% of every citation landed on a site the business itself owned, and review directories took 0.7%. Counted as one domain per reading, 16,516 of 20,143.

That is the whole asymmetry: the pages you own are the pages an assistant quotes back to a buyer. See the dated readings and the method on our proof page.

Find out whether assistants name your business, on your own domain and your own buying questions.

By Aditya Goel, Founder, AI Knows Us. Reviewed 1 November 2026.