- There's no fixed position to win. In a SparkToro and Gumshoe study, ChatGPT and Google's AI returned the same brand list for a prompt less than 1 time in 100, so the target is mention rate across many runs.
- ChatGPT showed just 3.10 links per answer in Slate's 1,000-question B2B test, and 372 answers had no link at all. Most of what it says about your category comes from sources you don't control.
- Blocking OAI-SearchBot keeps you out of ChatGPT search answers entirely. OpenAI says robots.txt changes take about 24 hours to apply.
- Only 12% of URLs cited by ChatGPT, Gemini and Copilot ranked in Google's top 10 for the same prompt in an Ahrefs study of about 15,000 prompts, so a page-one Google ranking doesn't carry over.
- You can't buy the organic answer. OpenAI says product results "are not ads" and that ads "do not influence the answers ChatGPT gives you."
What does it mean to rank in ChatGPT?
To rank in ChatGPT is to be one of the brands it names when someone asks a buying question, and ideally to be named first or described as the best fit. It's a different goal from being cited as a source. A citation means ChatGPT linked your page; a recommendation means it put your product in front of a buyer, with or without a link.
That distinction changes the work. If you want your articles linked as sources across ChatGPT, Perplexity and Google, our guide on how to get cited by ChatGPT and other AI engines covers each engine's index and crawler. This page is narrower: the commercial prompts where ChatGPT picks winners, and what moves you into that list.
Commercial prompts fall into five shapes, and each one is won in a different place:
| Prompt type | Example | Where you usually win it |
|---|---|---|
| Category | "best expense management software for a 200-person company" | Listicles, review sites and roundups that ChatGPT retrieves for the rewritten query |
| Comparison | "Ramp vs Brex for a startup" | Your own comparison page plus neutral head-to-head reviews |
| Alternatives | "alternatives to Expensify" | "Alternatives" articles, review-site category pages, Reddit threads |
| Problem | "how do I stop employees losing receipts" | Your how-to content, if it names your product as one fix among several |
| Branded | "is [brand] good for multi-entity accounting" | Your pricing, docs and review profiles, which have to agree with each other |
Most teams track only the last row because it looks good. The first three rows are where buyers who don't know you yet get a shortlist, so weight your effort there.
How does ChatGPT decide which brands to recommend?
ChatGPT answers a buying question in one of two ways: from what the model learned in training, or by searching the web first and writing from the pages it retrieved. You need to show up in both.
When ChatGPT searches
OpenAI's help center says that when ChatGPT uses third-party search providers, it typically rewrites your message into one or more targeted queries, sends them out, and may follow up with more specific queries after reading the first results2 (OpenAI). It also shares a general location, derived from IP, with those providers. OpenAI's launch post described the source mix as third-party search providers plus content from partners3, Microsoft had announced Bing as ChatGPT's default search back in 202315, and OpenAI runs its own search crawler, OAI-SearchBot. Sites that opt out of OAI-SearchBot "will not be shown in ChatGPT search answers"1 (OpenAI crawler docs).
So for "best expense software for a 200-person company," ChatGPT is probably not searching that exact sentence. It's searching something closer to "best expense management software mid-size companies," reading a short stack of results, and naming the brands those pages agree on. When it does cite, it favors reference and review sources: Wikipedia was 47.9% of ChatGPT's top 10 cited domains in Profound's analysis of 680 million citations9. That mix moves fast, too. Semrush watched the share of ChatGPT responses citing Reddit fall from about 60% to about 10% between early August and mid-September 202510.
When it doesn't search
Plenty of answers carry no link. In Slate's August 2026 test of 1,000 B2B software questions, 372 ChatGPT answers had none8 (Slate). Those recommendations come from training data, which is why the brands that dominate older comparison articles, review sites and forum threads keep appearing even when they've lost ground. You change that answer slowly, by being named in more of the places future models learn from, which is the same off-site work as the search path.
Product and shopping answers
For physical products ChatGPT may show a product carousel. OpenAI says products appear "when ChatGPT perceives it to be relevant to your intent," taking into account the query and context such as memory or custom instructions, and that product results are selected independently and are not ads4 (OpenAI). Merchant listings draw on product metadata from third-party providers or the merchants themselves. Shopify stores are covered through Shopify Catalog, and other merchants can apply for direct feed access. If you sell physical goods, a clean product feed is part of ranking. For software, it isn't.
Why your own test can mislead you
Memory and custom instructions shape what a logged-in user sees, so your own account is a bad test bed. It knows you work in your category and may know your brand. Test logged out, or in a Temporary Chat set to unpersonalized, which OpenAI says ignores memory and custom instructions6.
The 8-step plan to get recommended by ChatGPT
Run these in order. Steps 1 and 2 take a day, step 3 a few hours, and steps 4 to 7 are the ongoing work. Each step ends with a test, so you know when to move on.
1. Write 30 to 50 commercial prompts
Pull the wording from places buyers actually type: sales call notes, the "best" and "vs" queries in your Search Console data, your G2 or Capterra category name, and the competitor names prospects mention. Write each prompt the way a buyer would, with a constraint ("for a 200-person company," "that integrates with NetSuite"). Constraints are where smaller brands win, because the generic list is usually locked up by the category leaders.
Done when: you have at least 30 prompts, each tagged with a type from the table above, no more than 20% branded, and at least a third carrying a buyer constraint.
2. Take a baseline, three runs per prompt
Run every prompt three times in ChatGPT, logged out or in an unpersonalized Temporary Chat, and log what you see. Answers vary run to run, so a single screenshot tells you very little: SparkToro and Gumshoe found a less than 1 in 100 chance of getting the same brand list twice13. Count a brand as "recommended" for a prompt only if it shows up in at least two of three runs.
Done when: 90 or more runs are logged, with your mention rate, your average position when mentioned, and the competitors named, broken out by prompt type.
3. Make sure ChatGPT search can reach you
Check robots.txt for OAI-SearchBot (our robots.txt AI crawler checker does it in seconds), then check the firewall, which is where most blocks hide. Grep 30 days of access logs for OAI-SearchBot and match the IPs against OpenAI's published searchbot.json. Then verify the site in Bing Webmaster Tools and submit the sitemap. Leave GPTBot allowed too, unless you have a reason to opt out of training, since 37% of ChatGPT answers in the Slate test carried no link and drew on what the model already knew.
Done when: OAI-SearchBot gets 200 responses on your pricing, product and comparison pages in the last 30 days, and every one of those pages is indexed in Bing.
4. Map the pages ChatGPT reads for your prompts
From the baseline, list every URL ChatGPT cited, then count how many prompts each domain appears for. You'll usually get a short head (a review site, two or three listicles, a Wikipedia page, a Reddit thread) and a long tail. For each of the top 20 URLs, record whether you're named and who owns the page. That list is your real competition, and it's usually not your competitors' homepages. Remember that Ahrefs found only 12% of AI-cited URLs in Google's top 10 for the original prompt11, so don't skip this step by looking at the Google SERP.
Done when: you have a sheet of the 20 most-cited URLs for your prompts, each marked "named / not named," with an owner or contact for the ones that aren't.
5. Get named on those pages
Work the "not named" list. Pitch the listicle authors with something they can verify (a pricing page, a trial login, a customer quote they can check). Fill out and refresh your review-site profiles in the categories ChatGPT pulled. Answer the cited Reddit threads only where you have something specific to add, and say who you are. For comparison queries, ask neutral reviewers to include you in head-to-heads. The evidence that this matters more than links: in Ahrefs' study of 75,000 brands, branded web mentions correlated with AI Overview visibility at 0.664, against 0.218 for backlinks12. That was measured on Google's AI, not ChatGPT, and big brands have more of everything, so read it as direction.
Done when: you're named on at least half of the top 20 URLs. Our working target is 90 days for that, because listicle updates and review moderation run on other people's schedules.
6. Make your own pages say who you're best for
ChatGPT can only describe you the way the web describes you. Put a one-sentence positioning line in the first 100 words of your homepage and product pages ("[Product] is expense management software for mid-size companies with multiple entities"). Publish pricing, or at least a range; "contact sales" gives a model nothing to compare. Build an honest comparison page for each of your top three competitors and an alternatives page for the category leader, each with a table and a clear "choose them if" section. The original GEO paper found that adding statistics, quotations and cited sources raised visibility in generative answers by up to 40%14, so give those pages numbers with sources. Run each one through the citability grader before publishing.
Done when: every money page opens with a positioning sentence, pricing is public, and you have a comparison page per top-3 competitor that names a case where the competitor is the better pick.
7. Make the facts about you agree everywhere
Check that your category label, founding year, headquarters, pricing model and main integrations match across your About page, LinkedIn, Crunchbase, G2, Capterra and, if you're notable enough to have one, Wikipedia and Wikidata. Mismatches give a model two versions of you to choose from. Add Organization schema with sameAs links to those profiles (the schema generator builds it), mostly as housekeeping: OpenAI hasn't said it reads schema for recommendations.
Done when: the same category label and core facts appear on all of those profiles, word for word where possible.
8. Re-run the prompt set every four weeks
Run the same prompts, the same way, on a fixed day. Compare mention rate by prompt type, and check which cited URLs changed. When a number moves, look at the cited URLs first: a new listicle that names you, or a refreshed one that dropped you, explains most swings.
Done when: it's on the calendar, and every report shows mention rate, share of voice against named competitors and the top cited URLs, next to last month's.
Worked example: a mid-market expense tool
Here's the plan applied to a hypothetical company, "Ledgerly," an expense management product for companies with 100 to 1,000 employees. We haven't run these prompts; the example shows the method and the decisions it leads to, not results.
Prompt set (excerpt). Ledgerly writes 40 prompts. Twelve are category prompts with constraints, eight are comparisons, six are alternatives, eight are problems and six are branded:
id,type,prompt
C01,category,best expense management software for a 300-person company
C02,category,expense software that integrates with NetSuite for multi-entity companies
C07,category,cheapest expense tool with corporate cards for a mid-size company
V01,comparison,Ramp vs Ledgerly for a 300-person company
V03,comparison,Expensify vs Ledgerly for international teams
A01,alternatives,alternatives to Expensify for a mid-size company
P02,problem,how to enforce per diem policy across countries
B04,branded,is Ledgerly good for multi-entity accountingBaseline cost. 40 prompts x 3 runs = 120 runs. At about two minutes per run to paste, read and log, that's four hours by hand, or a few days of runs in a tracker.
Reading the baseline. Each pattern in the log points to a different next move:
- If Ledgerly is absent on category prompts but cited URLs are review sites and listicles that don't name it, the work is step 5. Rewriting Ledgerly's homepage won't change those answers.
- If Ledgerly is named on constrained prompts (C02, NetSuite) but not generic ones (C01), it's winning on a real differentiator. Write more constrained prompts and build pages around each constraint.
- If comparison prompts cite the competitor's "vs Ledgerly" page, Ledgerly has no page of its own for that pair. Build it (step 6).
- If branded answers get pricing or features wrong, check which cited page carries the old fact, then fix that page or the profile it came from (step 7).
- If an answer recommends Ledgerly with no link at all, it came from training data. That's good news that's slow to change either way, so leave it alone and keep working steps 4 and 5.
The citation map. After the baseline, Ledgerly sorts every cited URL by how many of its 40 prompts it appeared on and works the list from the top:
url | prompts_cited_on | named_ledgerly? | page_type | owner_contact | action | statusWhat doesn't get you recommended
These come up in every planning meeting. None of them moves the commercial answers on its own.
Buying ChatGPT ads
OpenAI began testing ads in ChatGPT in 2026 and lists answer independence among its ad principles: ads "do not influence the answers ChatGPT gives you," and they're labeled and kept separate from the answer5 (OpenAI). That's a company commitment, not an audited fact, but there's no product that sells a place in the organic list.
Calling yourself "the best" on your own site
A model reading five independent pages that agree, plus your page saying you're the leader, sides with the five. Self-description helps on branded prompts. On category prompts it barely registers until third parties say the same thing.
llms.txt and other quick fixes
No major AI search engine has said it uses llms.txt to choose sources, and the request data in our explainer shows almost no AI traffic to those files. It's cheap to add, but it won't put you in a shortlist.
Chasing one screenshot
"We're #1 in ChatGPT" based on one run is noise, given how rarely lists repeat. Report a mention rate over at least three runs per prompt and say how many runs sit behind every percentage.
Blocking GPTBot to "protect" content
Blocking GPTBot doesn't remove you from ChatGPT search, since that's OAI-SearchBot's job1. It does mean new content won't reach future training data, and that's the source of every no-link answer. For a software company that wants to be recommended, that's usually the wrong trade. Our AI crawler list covers when blocking makes sense.
How do you measure your ChatGPT ranking?
Track four numbers on your fixed prompt set, monthly, with run counts shown:
- Mention rate: the share of runs that name you. This is the headline metric.
- Share of voice: your mentions divided by all brand mentions across the set, against named competitors.
- Average position when mentioned: where you appear in the list. Useful as a trend only.
- Citation rate: the share of runs that link one of your URLs.
Then connect it to traffic. ChatGPT tags many referral links with utm_source=chatgpt.com7, so a GA4 exploration filtered to that source shows which landing pages it sends visitors to. That matters more each quarter for B2B: in Goodie's panel, ChatGPT's share of AI referral sessions fell from 89.1% to 62.6% while Claude's rose from 1.4% to 18.5%16, so track the others too. Our guide to measuring AI visibility has the sample-size math and a reporting template.
By hand, 40 prompts x 3 runs is an afternoon. If you need it weekly or across several engines, a tracker pays for itself; our comparison of ChatGPT rank trackers covers what each one measures and what it costs. For a free first read, the Arobis AI visibility checker shows whether ChatGPT names your brand for category questions.
How long does it take to rank in ChatGPT?
The technical part is fast and the rest isn't. OpenAI says robots.txt changes reach search within about 24 hours1, so an access fix shows up in the next month's baseline. Getting named on third-party pages depends on when their owners update and when ChatGPT's search step picks up the new version, which is weeks to months. Training-data answers change only when a new model ships. Plan on a quarter before judging the off-site work, and judge it on mention rate by prompt type, not on any single prompt.
If you'd rather hand the off-site and content work to a team, that's what agencies in this space sell. Arobis AI (the publisher of AI Ranked) does it for B2B SaaS, and our list of the best AI SEO agencies compares others. The GEO guide covers the broader discipline, and our GEO audit checklist turns steps 3, 6 and 7 into pass/fail checks.
Frequently asked questions
Can you pay to rank in ChatGPT?
No. OpenAI says ChatGPT's product results are selected independently and are not ads, and its ad principles state that ads do not influence the answers ChatGPT gives. Ads, where they run, are labeled and shown separately. The organic list of recommended brands comes from what ChatGPT retrieves and what the model learned, so the only way in is to be named in those sources.
How do I get my business recommended by ChatGPT?
Make sure OAI-SearchBot and Bing can reach your key pages, then find the pages ChatGPT cites for your buyers' prompts and get your business named on them: review sites, roundups, comparison articles and forum threads. Back that up with your own pages stating who you serve, what you cost and how you compare. Measure with a fixed prompt set, three runs per prompt.
Does ChatGPT use Google or Bing rankings?
Neither directly. OpenAI says ChatGPT search uses third-party search providers, its own OAI-SearchBot crawler and partner content, and Microsoft announced Bing as the default provider in 2023. Ahrefs found only 12% of URLs cited by ChatGPT, Gemini and Copilot ranked in Google's top 10 for the same prompt, so a strong Google position doesn't guarantee a ChatGPT recommendation.
Why does ChatGPT recommend my competitors and not me?
Usually because the handful of pages ChatGPT retrieves for that query name your competitors and not you. Log the cited URLs for the prompt across three runs and check whether you appear on each one. If you don't, that's the gap to close. If you do appear on them and still aren't named, look at whether those pages describe you for the same use case.
Does schema markup help you rank in ChatGPT?
There's no public evidence that it does for recommendations. OpenAI hasn't said ChatGPT reads structured data to choose brands. Organization schema with sameAs links to your profiles is still worth adding, because it keeps your facts consistent and costs an hour. Spend the real effort on third-party mentions and pages that answer buyer questions directly.
How often do ChatGPT recommendations change?
Constantly at the level of a single answer. SparkToro and Gumshoe found less than a 1 in 100 chance of the same brand list twice. Underneath that noise, how often a brand appears is far steadier, which is why you track mention rate over many runs. Real shifts follow changes in the cited sources, such as a listicle refresh, or a new model release.
Sources
- OpenAI. "Overview of OpenAI crawlers."
- OpenAI Help Center. "ChatGPT search."
- OpenAI. "Introducing ChatGPT search." 31 October 2024.
- OpenAI Help Center. "Shopping with ChatGPT search."
- OpenAI. "Testing ads in ChatGPT." 2026.
- OpenAI Help Center. "Temporary Chat FAQ."
- OpenAI Help Center. "Publishers and developers FAQ."
- Slate. "How to get cited by ChatGPT, Perplexity, Claude and Gemini" (1,000-question study). 27 August 2026.
- Profound. "AI platform citation patterns" (680 million citations, August 2024 to June 2025).
- Semrush. "The most-cited domains in AI: a 3-month study." 10 November 2025.
- Ahrefs. "Only 12% of AI cited URLs rank in Google's top 10 for the original prompt." 11 August 2025.
- Ahrefs. "An analysis of AI Overview brand visibility factors (75K brands studied)." 26 May 2025.
- Search Engine Journal. "AI recommendations change with nearly every query: SparkToro."
- Aggarwal et al. "GEO: Generative Engine Optimization." KDD 2024.
- CNBC. "Microsoft says Bing can be default search engine for ChatGPT users." 23 May 2023.
- Goodie. "2026 AI search traffic report." 21 May 2026.
AI Ranked Editorial. "How to rank in ChatGPT and get recommended for buyer prompts." AI Ranked, October 11, 2026. https://airanked.ai/guides/how-to-rank-in-chatgpt