Blog
How do you rank in ChatGPT?
By Bhavesh Barot, Updated

ChatGPT has no fixed ranking to climb, because one question asked ten times on 15 November 2026 got ten different shortlists. You can make sure your website says plainly what you do and for whom, and lets in OAI-SearchBot, the robot that OpenAI, ChatGPT's maker, uses for search. The question was "Which chartered accountant firms in Bengaluru are good for startups?", and the ten answers named 20 firms between them; seven appeared only once.
Is there a ranking in ChatGPT at all?
"Rank" is the searcher's word, the one people type into Google when they ask this question. On Google, a page holds a place in a list of links that anyone can look up. Our ten ChatGPT runs got ten newly written answers instead. No two named the same set of firms, and no two linked the same set of websites.
In ChatGPT, a ranking can only be a pattern across answers: how often you are named when buyers ask, where you come in each list, whether your own website is linked, and what is said about you. In our ten answers, one firm came first nine times, while seven firms appeared only once. A single answer is one sample.
You may also hear this work called generative engine optimization, a name for efforts to get a business into AI-written answers. Our post on that term explains it.
What happened when we asked ChatGPT the same question ten times?
On 15 November 2026 we sent ChatGPT one question ten times: "Which chartered accountant firms in Bengaluru are good for startups?" Each run went to the consumer ChatGPT with web search switched on, in English, with the location set to India. We sent them through DataForSEO, a data company whose service passes a question to ChatGPT and returns the answer. All ten ran within two minutes, between 02:49 and 02:50 UTC (Coordinated Universal Time, the world's reference clock).
We read the ten answers and counted every firm each one recommended, including the shorter lists of other firms that three answers added at the end. We left out headings and the questions ChatGPT asked back.
Each answer recommended between 5 and 9 firms, 68 recommendations in all. Between them, they named 20 different firms. No two answers recommended the same set.
This is how often each of the 20 firms came up:
| Answers that named the firm (out of 10) | Firms |
|---|---|
| 10 | 2 |
| 7 | 1 |
| 6 | 1 |
| 5 | 1 |
| 4 | 2 |
| 3 | 3 |
| 2 | 3 |
| 1 | 7 |
Start at the bottom row. Seven firms appeared in one answer and in none of the other nine. The owner of one of those firms, checking ChatGPT once, could have concluded the firm was invisible, or that it had arrived.
The top was steadier. Two firms appeared in all ten answers. One was named first in nine of them and second in the tenth, and the other came second in seven. Below those two, the list changed from one answer to the next.
What did the ten answers link to?
The ten answers held 94 links between them, from 3 in one answer to 18 in another. Of the 94, 87 pointed to the recommended firms' own websites. The other seven went to google.com (three links) and to two other websites (four links).
| Answer | Firms recommended | Links | Different websites linked | Links to a recommended firm's own website |
|---|---|---|---|---|
| 1 | 5 | 12 | 5 | 12 |
| 2 | 7 | 18 | 7 | 18 |
| 3 | 6 | 6 | 2 | 6 |
| 4 | 7 | 4 | 3 | 4 |
| 5 | 9 | 5 | 5 | 5 |
| 6 | 6 | 3 | 3 | 3 |
| 7 | 8 | 11 | 5 | 11 |
| 8 | 6 | 13 | 5 | 10 |
| 9 | 7 | 5 | 5 | 5 |
| 10 | 7 | 17 | 9 | 13 |
| All ten | 68 | 94 | 17 | 87 |
Every answer linked at least one recommended firm's own website. A link did not come with every recommendation, though. Of the 68, 46 came with a link to that firm's own site, and 22 came without one.
The two firms named in all ten answers had their own website linked in all ten. Six firms were recommended without their own website being linked in any answer, and one of those six was named in seven answers out of ten. In answer 8, two of those six firms got a link to google.com and none to a website of their own.
Four answers showed a rating out of 5 and a review count beside some firms, all between 4.7 and 5. We cannot tell from the answers where those ratings came from. One answer called them snapshots of local listings.
Why would the same question get different answers?
We can't see inside ChatGPT, so we can't say for certain.
No two answers linked the same set of websites. One linked 2 different sites, and another linked 9. Nine answers said what their shortlist was based on, and all nine pointed to services the firms publish or advertise. Seven of those nine also mentioned business information or customer reviews.
Every answer described each firm on its main list with a few lines of services, such as ESOP structuring (setting up share options for staff) or virtual CFO work (a part-time, outsourced head of finance). Six answers added a table matching a startup's situation, such as raising funds or having foreign investors, to the firms to approach first.
Six of the ten said, in so many words, that their list was not an independently verified or audited ranking.
And all ten stopped to ask questions back. Each asked about the startup's stage and its monthly budget, and offered a shorter list in return. We did not reply, so we don't know what those shorter lists would have held.
How can you see what ChatGPT says about your firm?
Do what we did, on a smaller scale. Open a new chat and type the question a buyer would ask, with your city in it. Note the date, every firm named, the order and every link. Then ask again in another new chat, and again on another day. A firm named in one answer out of ten, as seven were in our check, can be missed on any single try.
Seen in Search asks the questions for you. The free report a visitor gets today asks ChatGPT only, using buyer questions written from what your homepage says you sell. Each question is asked once, which makes the report a first reading. The full paid report covers ChatGPT, Google AI Overviews and Perplexity.
Each report turns the answers into a score from 0 to 100, computed in code from five measurements with fixed weights. Presence, whether you are named at all, carries the most weight, at 35%. Citations, whether an answer links to your own website, is another of the five. Our methodology page lists them all, and what your AI visibility score means explains how to read the number. A report is private when it finishes, and you decide whether to publish it.
One free report per business. Use your business email. No card.
How does ChatGPT find your website?
With web search on, as in our check, 87 of the 94 links in ChatGPT's ten answers went to recommended firms' own websites. OpenAI, which makes ChatGPT, publishes a page about the robots that visit websites for it. We read it on 15 November 2026, and for search the one that matters is OAI-SearchBot.
OpenAI says it uses OAI-SearchBot to bring websites into ChatGPT's search results, and recommends allowing the robot, to help make sure your site appears in them. You allow or block it in robots.txt, a plain text file on your website that tells robots which pages they may visit. A site that blocks it, the page says, is left out of ChatGPT search answers, though it can still appear as what OpenAI calls a navigational link, a term the page does not define. For search results, OpenAI adds, a robots.txt change can take about 24 hours to reach its systems.
GPTBot is a different robot, for training. It collects content that may be used to train OpenAI's AI models, and blocking it tells OpenAI not to use your content that way. OpenAI says the two settings work independently, so a site can block GPTBot and still let OAI-SearchBot in.
The page tells site owners how to use robots.txt. It does not ask them for an llms.txt file (a plain text file that tells AI tools what a site is about); our post on llms.txt covers that file.
We did not test whether blocking or allowing OAI-SearchBot changed any answer.
What can you change?
- Check the robot setting. In our ten answers, 87 of the 94 links went to firms' own websites. Ask whoever runs your website whether robots.txt blocks OAI-SearchBot. If it does, OpenAI says your site will be left out of ChatGPT search answers, though it can still appear as a navigational link. Allowing the robot comes with no promise of a place in an answer, and OpenAI's page makes none.
- Say what you do, and for whom, on your own website. Nine of our ten answers said they based their shortlist on services the firms publish or advertise. If you want to be suggested to startups preparing to raise money, a page on your site should say you do that work, in the words a founder would use. We have not measured how long a new page takes to change an answer, or whether it does.
- Check what appears beside your firm's name. Six firms were recommended without their own website being linked in any answer, and four answers showed a rating out of 5 and a review count beside some firms. In answer 8, two of those six firms got a link to google.com and none to a site of their own. We can't tell where the ratings came from. For Google's own local results, its Business Profile help says that more reviews and good ratings can help a business rank locally. If you are a chartered accountant, company secretary or advocate in India, your professional body restricts advertising and soliciting work; check its current rules before you list your practice, publish fees, advertise or ask clients for reviews.
- Count the samples behind any result. In our ten answers, seven of the 20 firms appeared only once, so a single answer can show you in or out by chance. Seen in Search's report asks each question once, so one report is one sample; check the questions that matter most to you by hand on other days as well. Anyone selling a fixed place in ChatGPT is selling something our ten answers did not have.
What did our check not measure?
Ten answers to one question make a small sample.
- The question was about chartered accountants for startups in Bengaluru. A dentist in Pune or a law firm in Mumbai may see a different pattern.
- All ten ran within two minutes on 15 November 2026, in English, with the location set to India. We did not ask again on other days or with other locations.
- Web search was on for every run. Answers with it off, or shaped by a person's own chat history and settings, were not tested.
- Every answer asked about the startup's stage and budget. We did not reply, so the shorter lists they offered are unknown.
- We did not visit the firms' websites, or check whether the answers described them correctly.
- Ten answers cannot show why one firm came first in nine of them. Its website was linked in all ten, but that pattern does not prove a cause.
- We asked ChatGPT only. Google's AI Overviews work separately; our post on them covers how.
Questions owners ask about ranking in ChatGPT
How to rank on SearchGPT?
How do you make your website show up on ChatGPT?
How can ChatGPT recommend my business?
How to make ChatGPT recommend your brand?
How to get your brand recommended by AI?
How to be found on ChatGPT?
How to get listed on ChatGPT?
How do I get my business to pop up on ChatGPT?
How to appear higher in Google search?
Is it free to put my business on Google?
One free report per business, private unless you publish it. Use your business email. No card. Questions about the report? Contact us.
Sources
- OpenAI, Overview of OpenAI Crawlers, read 15 November 2026. The old address, platform.openai.com/docs/bots, now forwards to this page.
- Google, Add or claim your Business Profile, Google Business Profile Help, read 15 November 2026.
- Google, Tips to improve your local ranking on Google, Google Business Profile Help, read 15 November 2026.
- Seen in Search, How the score works, read 15 November 2026.
- Our own check: ChatGPT (with web search) asked "Which chartered accountant firms in Bengaluru are good for startups?" ten times, in English, with the location set to India, through DataForSEO, 15 November 2026, 02:49 to 02:50 UTC. Answers change; these are the answers it gave then.
- Our own keyword data: Google Ads monthly search volume for "how to rank in chatgpt", 170 in the United States and 110 in India (average of the last 12 months), through DataForSEO, 15 November 2026.