How to Rank in ChatGPT and Perplexity: What 95 Citations Told Us

Most advice on ranking in ChatGPT is written from theory. We wanted to know what the engines actually pick, so we asked them. On 10 September 2026 we ran ten questions a business owner might ask before hiring an agency, through both ChatGPT search and Perplexity, and logged every source each one cited. This article is what came back, and what it means if you want your own site in those answers.

If you are new to the topic, our pillar guide on GEO vs AEO vs SEO explains the landscape. This piece is narrower: two engines, ten questions, and the pages that won.

How Does ChatGPT Decide What to Cite?

OpenAI has published more about this than most people realise. Three facts from its own documentation shape everything else in this article.

ChatGPT search does not have its own web index. OpenAI’s help centre says that to answer questions, “ChatGPT search sometimes partners with other search providers,” and links to the privacy policies of Microsoft and Shopify as those providers. In practice that means Bing’s index is the candidate pool. A page that Bing has not indexed cannot be retrieved, no matter how good it is.

ChatGPT rewrites your question before it searches. Again from OpenAI: it “typically rewrites your query into one or more targeted queries that it sends those providers,” then may send further, more specific queries after reviewing the first results. You are not ranking for the user’s words. You are ranking for the model’s reformulation of them, which tends to be shorter and more literal.

ChatGPT knows roughly where the user is. OpenAI states that ChatGPT “collects general location information based on your IP address and may share that general location with third-party search providers.” Our test was run from Pakistan, and 6 of 10 answers came back localized, including one about how much a B2B company should spend on SEO, a question with no local intent at all.

Two crawler rules complete the picture. OpenAI’s bot documentation says OAI-SearchBot “is used to surface websites in search results in ChatGPT’s search features,” and that sites blocking it “will not be shown in ChatGPT search answers.” It is separate from GPTBot, which is the training crawler. You can block training and still allow search.

Perplexity is simpler. It maintains its own index through PerplexityBot, which its docs describe as “designed to surface and link websites in search results on Perplexity,” and every answer carries numbered citations back to the pages it used. There is no partner index in the middle.

What We Tested and How

We chose ten questions that sit close to a purchase decision for the kind of business Digital Rhodium serves, then ran each one through both engines with no account, no history, and no memory. For Perplexity we recorded the full source list from its Links tab. For ChatGPT, which shows fewer citations inline, we recorded every named source chip in the answer.

Two honest limits. Logged-out Perplexity capped us after a few queries and one query never completed, so the Perplexity sample is nine questions and 95 sources, not ten and 100. And our IP location skewed ChatGPT toward Pakistani results on commercial questions, which is itself a finding but means the specific sites it named will differ for a user in Houston. The patterns are the point, not the individual domains.

What Got Cited in Perplexity

The 95 sources came from 75 different domains, so Perplexity spreads its citations widely. What it cites is remarkably consistent, though.

Vendor content dominates. 52 citations were agency blog posts and 29 were blog posts from software companies such as Semrush, HubSpot, Yotpo and Conductor. That is 86% of the sample. Publishers and reference sites (Search Engine Land, Coursera, Wikipedia, the Cambridge Dictionary) accounted for 9. Directories and review platforms (DesignRush, G2) for 5. Notably, DesignRush was cited four times for the same query with four different paginated listing URLs, which suggests Perplexity is retrieving by page-level relevance rather than picking one canonical page per site.

The URL restates the question. 65 of 95 cited URLs contained most of the query’s words in the slug. On “how to get my business cited by ChatGPT,” every single one of the ten titles was a rewording of the query: “How to Get Cited by ChatGPT for Your Business,” “How Do I Get My Business to Show Up in ChatGPT?,” “How to Get Your Business Recommended by ChatGPT.” On “what is answer engine optimization,” eight of ten titles began with the same four words.

Blog posts, not service pages. 61 of 95 URLs sat on a path like /blog/, /insights/, /news/ or /academy/. Only 4 were a homepage or top-level page. Not one was a city landing page, a pricing page or a “services” page, even on the query about PPC management cost, where an agency’s own pricing page would be the obvious answer. The pages that won were the ones written to answer the question, not the ones written to sell.

Recency is a visible signal on commercial queries. On “best PPC management agency for small business,” 12 of the 15 cited titles included “2026.” Across the whole sample it was closer to one in four. Year-stamping helps most where the user is comparing options and wants a current list.

A few domains get cited again and again. Clicks Geek appeared five times across three queries, Semrush and HubSpot four times each, Thrive three times. None of these are household names outside marketing. They are sites that have published a separate, literal answer to dozens of adjacent questions.

What Got Cited in ChatGPT

ChatGPT is a different animal, and treating it like Perplexity is the first mistake we see.

It cites very little. Across ten answers ChatGPT named between one and three sources each. There is no long tail of citations to squeeze into. If you are not one of the two or three pages it leans on, you are not in the answer.

It prefers the primary source for definitions. For “what is generative engine optimization,” “what is answer engine optimization” and “is schema markup important for AI search,” ChatGPT’s only cited source was Google Search Central. For “how to get my business cited by ChatGPT,” it cited OpenAI’s own help centre. Perplexity, on the same questions, cited ten vendor blogs and no documentation at all. If your article is a paraphrase of the official docs, ChatGPT has no reason to cite you instead of them.

It cites small, specific sites when the question is commercial. This is the most useful finding for anyone without a big brand. On the cost and agency questions, ChatGPT named seven agencies we had never heard of, several with a handful of reviews, because they had published a page stating their actual prices for the user’s region. A page that says “PPC management starts at X per month” beat every large agency that says “contact us for pricing.” Specificity, not authority, won those citations. We later checked what those agencies actually charge in our PPC management pricing teardown.

It localizes whether you asked or not. Six of ten answers were tailored to our IP location. That means the same query produces a different set of cited pages in Houston, Richmond or Orem, and it means consistent, specific location information on your site is a ranking input for ChatGPT in a way it never was for classic blue-link SEO.

How Does a Small Website Rank in ChatGPT?

Put the two datasets together and the playbook is short. None of it requires a large domain, and all of it is checkable.

1. Be in Bing. Verify the site in Bing Webmaster Tools, submit the sitemap, and check that the pages you care about are indexed there. This is the single most skipped step, and it is a hard gate for ChatGPT search.
2. Allow the search crawlers explicitly. Your robots.txt should not block OAI-SearchBot or PerplexityBot. If you block GPTBot to keep content out of training, do it by name so search access survives.
3. Write one page per question, and use the question as the title. The cited pages restated the query in the title and slug. A single long service page that touches twelve questions did not get cited once in our sample.
4. Say the specific thing. Prices, ranges, timelines, named tools. The ChatGPT citations on commercial queries went to pages with numbers on them. If you cannot publish a number, publish a range.
5. Put the year in titles that compare options. Where the user is choosing between options, cited titles carried the year 12 times out of 15. Where the user wants a definition, it barely mattered.
6. Make your location unambiguous. ChatGPT will localize. The pages it picked for local questions had a city and a price on them. Our guide to schema markup for AI search covers the Organization and Service markup that supports this without faking an address.
7. Add something the docs do not have. On definitional questions ChatGPT went straight to Google and OpenAI. To be cited instead of them you need original data, a test, or a position the documentation does not take. That is the reason this article exists.

What did not show up as a factor: domain authority in the classic sense. The Perplexity sample was full of small agencies. The ChatGPT commercial answers were built on businesses with 28 reviews. Being findable and literal beat being big.

How Do You Track ChatGPT Rankings Over Time?

There is no Search Console for ChatGPT, and the interface changes month to month, so you need your own panel. Here is the lightweight version we use.

1. Fix a query set. Fifteen to twenty questions your buyers actually ask, written the way they would type them. Do not change the wording between runs.
2. Run it on a schedule, logged out. Monthly is enough to see movement. Logged-out removes memory and personalisation, which would otherwise make your own results look better than a stranger’s.
3. Record four things per query. Whether you were cited, which URL, which competitors were cited, and how many sources the answer used. Perplexity’s Links tab gives you the full list. In ChatGPT, open the Sources panel.
4. Watch the referral data you already have. Visits from AI engines arrive in GA4 with chatgpt.com or perplexity.ai as the referrer, and GA4 now groups them into an “AI Assistant” default channel. Small numbers, but real, and they are the only conversion-linked signal you will get.
5. Treat Google’s AI features separately. Google says traffic from AI Overviews and AI Mode is already “included in the overall search traffic in Search Console” under the Web search type, so it cannot be isolated there. Your panel covers ChatGPT and Perplexity; Search Console covers Google. We ran the same kind of panel on Google itself in our AI Overviews test.

If you would rather buy a tool, several of the pages Perplexity cited on the measurement query were tool vendors, which tells you the category is real. The manual panel costs an hour a month and produces the same core numbers.

Does Ranking in ChatGPT Affect Google Rankings?

Not directly, and not in either direction. ChatGPT search reads from Bing’s index and its partners; Google’s AI features read from Google’s index. Google’s own guidance says there are “no additional requirements to appear in AI Overviews or AI Mode,” and that a page simply needs to be indexed and eligible for a snippet. The overlap is that the same page qualities, a literal answer to a specific question with verifiable detail, get retrieved by all three systems. Build for that and you are building for Google at the same time, which is the argument our GEO guide makes at length.

Frequently Asked Questions

← Back to all articles