How to Rank in ChatGPT: A Measured Playbook

How do you rank in ChatGPT?

Ranking in ChatGPT means getting into the small set of pages its retrieval layer pulls before the answer is written, which is a different job from winning a position on a results page. ChatGPT retrieves dozens of URLs per question and cites roughly half of them. The work is making your page one of the ones retrieved, then one of the ones that survives.

Four things carry most of the weight, and they are cheap relative to the tactics this subject usually attracts. Allow OAI-SearchBot, which is the crawler that populates ChatGPT Search. Server-render every word that must be found. Open each section with its answer, naming the subject rather than using a pronoun. And write about the sub-questions your topic decomposes into, because that is what retrieval scores against.

1. Allow OAI-SearchBot, which is the crawler that populates ChatGPT Search.
2. Server-render every word that must be found.
3. Open each section with its answer, naming the subject rather than using a pronoun.
4. Write about the sub-questions your topic decomposes into, because that is what retrieval scores against.

The mechanics underneath those four belong to the generative engine optimization guide and to query fan-out, and the comparison across all five assistants belongs to the platform-by-platform pillar. This article stays on what is specific to ChatGPT, including the two tactics its own search results recommend most and measurement supports least.

How does ChatGPT choose which pages to cite?

ChatGPT chooses citations by scoring retrieved passages against synthetic sub-queries rather than against the question a person typed. The largest published measurement is Ahrefs' analysis of 1.4 million ChatGPT prompts (opens in a new tab) from February 2025, run with data scientist Xibeijia Guan. It found ChatGPT cited 49.98 per cent of the URLs it retrieved, so half of everything pulled into the process is discarded before the answer appears.

That dataset is from February 2025, and the reversals described later are the reason it is quoted for the shape of the retrieval channels rather than for their current sizes.

Where a URL came from predicted its fate better than anything else in that dataset. Citation rates split sharply by source type, and the distance between the top and the bottom of the table is a factor of more than forty.

Read that table as odds conditional on retrieval, not as a promise. A page in ChatGPT's search channel still has to be retrieved before any of it applies. Once retrieved, that page is far more likely to be cited than one arriving from anywhere else.

Two smaller findings from the same study point at what to write. Title similarity to the fan-out sub-query scored 0.656 against 0.602 for similarity to the prompt itself, so the heading that earns a citation answers a question nobody typed. And descriptive natural-language URL slugs appeared on 89.78 per cent of cited pages against 81.11 per cent of uncited ones.

Does Reddit actually get you cited in ChatGPT?

Less than the volume figures suggest, and as of August 2026 much less than they suggested a month ago. Reddit spent most of 2026 as simultaneously ChatGPT's most-cited domain and one of its least-cited source types, and almost every guide on this subject quotes only the first half. Then in August the first half changed. All three numbers below are worth holding at once.

The volume figure is real and it is what every guide quotes. Ahrefs' guide to ranking on ChatGPT, published on 13 March 2026 and ranking on page one for this query, says Reddit is the single most-cited domain, with 847,338 mentions against Wikipedia's 431,710 and Amazon's 97,457. Ahrefs' own most-cited-domains tracker still had Reddit first on July 2026 data, at a 16.7 per cent mention share against Wikipedia's 8.9. Its third recommendation is to engage authentically on Reddit.

Ahrefs' own 1.4 million prompt study measures the other thing. Reddit URLs were retrieved 16,182,976 times and cited 1.93 per cent of the time, and 67.8 per cent of every uncited URL in the dataset was a Reddit URL. Reddit tops the citation count because ChatGPT pulls an enormous volume of Reddit, not because a given Reddit thread is likely to be quoted.

The difference decides where an hour goes. A comment you leave is one URL entering a pool with roughly 1 in 50 odds of citation. A page of your own that reaches ChatGPT's search channel enters a pool where 88 per cent of retrieved URLs get cited. Contribute to Reddit because your buyers read it and because the AI citation you might earn is real. Do not fund a seeding programme on a volume statistic.

The same caution applies to YouTube

YouTube produces the identical reading error. Ahrefs reports that YouTube mentions showed the strongest correlation with ChatGPT visibility of any factor it tested, and that correlation is real. YouTube's measured citation rate in the 1.4 million prompt dataset is 0.51 per cent, the lowest of the five source types. Large brands make videos and large brands get cited, and no published test has separated the two.

What changed when ChatGPT stopped citing Reddit in August 2026

Anyone who did fund a Reddit programme on that volume statistic watched it break in four days. Otterly (opens in a new tab) tracked Reddit citations in ChatGPT falling from 497 a day to 132, a 73.4 per cent drop arriving in two steps on 8 and 14 August 2026. Promptwatch measured the same event as a fall from a 3.83 per cent average share of ChatGPT Search citations between 18 July and 7 August to 0.52 per cent between 14 and 17 August, a relative decline of 86.4 per cent.

Otterly's control is what makes it credible rather than an artefact. Across 16 brand reports in unrelated industries, ChatGPT's total citation volume rose 3.5 per cent over the same window while its Reddit citations fell, and Google AI Mode, Gemini and Perplexity moved 0.6 per cent and carried on citing the exact threads ChatGPT dropped. Three of the 16 reports recorded no Reddit citations at all after 14 August.

The most useful fact about the episode is that it was not about Reddit. Petra Labs (opens in a new tab) analysed 113,500 ChatGPT runs between 1 and 16 August 2026 and found every major social platform lost citation share at once: YouTube fell 88 per cent, Reddit 81, TikTok 72, LinkedIn 36 and Facebook 28. Reddit took the largest absolute loss because it started from the largest share, and the steepest percentage fall was YouTube's.

That breadth rules out most of what was written in the first week. A Reddit robots.txt block, reported on 20 August, cannot make YouTube citations fall 88 per cent. What fits is a change in how ChatGPT selects sources, which is Otterly's reading of a site: operator scoping retrieval toward sources it already trusts, and what Lily Ray's analysis of ChatGPT fan-out queries describes independently. OpenAI has not commented. Reddit's chief communications officer, Adam Collins, said the change has no meaningful impact on its business.

Petra Labs also measured where the citations went, and the answer is the reason this section sits in this article. Brand visibility in ChatGPT answers did not fall with the social platforms. Citations shifted toward companies' own websites and review platforms instead, so the same brands kept being recommended while the sources behind the recommendation changed.

Has ChatGPT done this to Reddit before?

Three times in a year, which is what the episode above does not support reading as the end of Reddit. Reddit peaked above 14 per cent of ChatGPT answers in September 2025 and fell to 2 per cent within weeks. Kevin Indig attributed that drop to Google removing the num=100 search parameter, which cut access to the deeper results where Reddit threads sit, rather than to any decision by OpenAI or Reddit.

Its share fell again from roughly 7 per cent to under 1 in mid-June 2026. Separately, seoClarity measured ChatGPT citations falling 86 to 94 per cent across five markets between February and April 2026, across millions of interactions, and rebounding in May. Different causes, the same shape.

Three collapses in a year is the finding, and it is worth more than any single figure in this section. Recovery is not clean either: after September 2025 the share fragmented rather than returning, growing in some categories and staying flat in others. A tactic whose value can halve because a platform changed a retrieval rule, or because Google changed a search parameter, is not a strategy you can plan a quarter around. Own the pages you control, and treat third-party surfaces as upside you cannot bank.

Do brand mentions across the web help you rank in ChatGPT?

Brand mentions across the web correlate with AI visibility about three times more strongly than backlinks do, which is the clearest off-page finding available and the one most often over-read. Ahrefs ranked 75,000 brands and published Spearman correlations of 0.664 for branded web mentions, 0.527 for branded anchors and 0.218 for number of backlinks. Domain Rating came in at 0.326.

Two limits sit on that result and both matter here. It measured Google AI Overviews rather than ChatGPT, so carrying it across is an extrapolation, and the evidence for that surface is in how to appear in Google AI Overviews. And a correlation of this kind is confounded by size, because large brands earn mentions and get cited, and no published intervention has separated the two. Nobody has run the controlled test that would settle it.

What the direction supports is ordinary earned coverage rather than a link-buying budget. Being written about, reviewed and listed on the authoritative third-party sites your buyers already read is what a mention is, and digital PR is the honest name for the work. Those are also the pages ChatGPT's search channel retrieves when somebody asks it for a recommendation.

ChatGPT is still where most of that attention pays off for business audiences, though its lead is narrowing. Goodie tracked 2.8 million AI referral sessions across 41 business-facing brand sites and recorded ChatGPT supplying 89.1 per cent of AI referrals in its May to August 2025 wave and 62.6 per cent by March to April 2026. A quarter of the B2B market moved in under a year, and ChatGPT still carries most of it.

Which ChatGPT crawler do you need to allow?

OAI-SearchBot is the crawler that decides whether you can appear in ChatGPT Search, and it is not the one most sites think about. OpenAI runs four bots with separate purposes and separate robots.txt tokens. Blocking the wrong one is the most common self-inflicted wound in this field, usually committed by a team intending to opt out of model training.

Two consequences follow directly from that table. A site can block GPTBot, stay out of training data, and remain fully eligible for ChatGPT Search. And robots.txt is not an access control on this platform, because ChatGPT-User fetches on a person's behalf regardless. Confirm the outcome in your own logs rather than in your configuration file, and how AI crawlers read your site covers the checking.

OpenAI's crawlers also waste an unusual share of their fetches, which makes ordinary hygiene worth more here than elsewhere. Vercel and MERJ measured OpenAI spending 34.82 per cent of fetches on 404 pages and 14.36 per cent following redirects, against Googlebot's 8.22 per cent and 1.49 per cent. No major AI crawler executes JavaScript, though GPTBot downloads JavaScript files in 11.50 per cent of its requests.

The volume is real enough to plan capacity around. Alli AI logged 24,411,048 requests across 69 websites and more than 78,000 pages between 14 January and 9 March 2026, and ChatGPT-User made 3.6 times more requests than Googlebot over that window. GPTBot, the training crawler most robots.txt files name, ranked seventh.

How is ChatGPT ranking different from traditional SEO?

ChatGPT ranking differs from traditional SEO in the unit that competes and in the index being searched. A traditional search engine ranks pages and returns a list for a person to choose from. ChatGPT retrieves passages from its own index and returns one answer, so a section competes rather than a page, and being quoted replaces being clicked.

Most traditional SEO still applies underneath, which the more excitable guides on this subject understate. Indexing, crawl access, internal links, clear structure and content worth reading all decide whether a page can be retrieved at all. Search engine optimisation is the floor under ChatGPT visibility rather than a discipline ChatGPT has replaced.

ChatGPT Search launched on Microsoft's search infrastructure and OpenAI has since built an index of its own, which is what OAI-SearchBot populates. That is why search-channel URLs were cited 88.46 per cent of the time in the Ahrefs study, against 1.93 per cent for Reddit and 0.51 per cent for YouTube. Bing indexation remains a useful floor check, and a Bing position is a poor lever.

Two habits transfer badly. Targeting one head term is the first, because the retrieval layer scores against sub-queries a person never typed. Saving the conclusion for the end is the second: Kevin Indig matched 18,012 verified citations to their source sentences across 3 million ChatGPT responses and found 44.2 per cent came from the first 30 per cent of the page.

Does ChatGPT prefer fresh content?

ChatGPT does lean fresher than every other assistant measured, and the lean is smaller than the advice around it implies. Ahrefs analysed nearly 17 million cited URLs and found AI-cited content averaged 1,064 days old against 1,432 days for the organic top 10, 25.7 per cent fresher. ChatGPT sat at the front of that field.

958 days is two years and seven months. ChatGPT prefers content 458 days newer than the organic results for the same queries, which is a genuine and measurable preference, and it still cites material older than most content calendars assume is dead. A definitional page two years old is not losing ChatGPT citations for its age.

So update the pages that genuinely decay and leave the rest alone. Pricing, tool comparisons and anything carrying a year in its title belong on a short cycle. Restamping a stable page to manufacture a freshness signal trains a team to ignore its own dates, and Google decides whether to trust a site's modification dates per site rather than per page.

Does structured data help you rank in ChatGPT?

Structured data does not measurably move ChatGPT citations, on the only controlled test anyone has published. Ahrefs added JSON-LD to 1,885 pages (opens in a new tab) between August 2025 and March 2026 and compared them against roughly 4,000 matched control pages. ChatGPT citations moved 2.2 per cent, Google AI Mode 2.4 per cent, and AI Overviews fell 4.6 per cent. Four separate statistical approaches agreed the result was indistinguishable from zero.

searchVIU tested the same question from the opposite direction, with a page carrying a price that existed only in JSON-LD. No engine extracted it, ChatGPT included. Retrieval pipelines convert HTML to text and strip script tags before anything is embedded, so a fact living only in markup is a fact ChatGPT never sees.

Keep your schema anyway, for rich results in Google and for resolving your organisation into an entity. Move any fact that has to be quotable into visible body text, which is the actual fix. The wider list of tactics tested and found not to work sits in the GEO guide, and how LLMs pick their sources covers the selection stage in more detail.

Why a ChatGPT citation is not a brand mention

ChatGPT cites far more often than it names, and the gap is wider on ChatGPT than on any other engine measured. Semrush, working with Kevin Indig, logged 3,981 domain appearances across 115 prompts, 14 countries and four engines, publishing on 9 June 2026. ChatGPT cited a domain in 87 per cent of its appearances and named the brand in the answer text in 20.7 per cent.

Gemini behaves close to the reverse, naming the brand in 83.7 per cent of appearances while citing it as a source in 21.4 per cent. Across all four engines, 62 per cent of citations produced no brand mention at all. A team reporting one blended AI visibility number is averaging two behaviours that move independently.

Track both on ChatGPT specifically. Counting only AI mentions understates ChatGPT by roughly four to one, and counting only source links understates Gemini by about the same margin in the other direction. Measuring AI search visibility sets out the two metrics and why they need separate reporting.

What ChatGPT visibility looked like on our own site

RankX AI measured its own ChatGPT visibility rather than only quoting other people's, and publishes the number whether or not it flatters. Across the 30 days to 21 August 2026, ChatGPT named the brand in 10 of 44 analysed answers, a 22.7 per cent mention rate, against 20.2 per cent across all five assistants and 218 answers.

The method is four prompts run nightly against five assistants, with each stored answer analysed for whether it names the brand. Tracking began on 10 August, so the window holds eleven nights of runs. That is a baseline on a small panel rather than a benchmark, and it supports exactly one claim: this is where the site started, and next month's figure is comparable to it.

ChatGPT sat level with Gemini at 22.7 per cent and just behind Grok at 23.3 per cent, while Perplexity named the brand in 5 of 44 answers, 11.4 per cent. The full table, the prompts and the limits are in our AI visibility baseline. No external benchmark exists for these numbers, because a mention rate depends entirely on which prompts were chosen. Perplexity sits lowest for reasons specific to how it retrieves, which is why Perplexity rewards Google rankings more than the others.

How to tell whether any of this worked

Measuring ChatGPT visibility requires a fixed panel of prompts run repeatedly, because a single check cannot separate a result from noise. SparkToro and Gumshoe had 600 volunteers run 12 brand-recommendation prompts 60 to 100 times each across three assistants in November and December 2025, 2,961 runs in total. Two runs of the same prompt returned the same brand list under 1 time in 100, and the same order roughly 1 time in 1,000.

Three grounded measures survive that volatility. Your server logs answer whether OAI-SearchBot is fetching your pages and receiving 200 responses. Your analytics answer whether ChatGPT referrals arrive and convert. And an aggregate mention or citation rate across dozens of prompts, tracked month on month, answers whether anything changed.

A screenshot of one good ChatGPT answer is not evidence, and neither is a screenshot of one bad one. Set the panel before making changes, so the comparison exists when you need it.

RankX AI builds the third of those measures, and the disclosure belongs on a page making this argument. AI Visibility runs a prompt panel across ChatGPT, Claude, Gemini, Grok and Perplexity, and reports mentions and citations separately for the reason above. Two checks are free and need no account: the AI Readiness Score grades a single page against the extraction rules in this article, and the AI Crawler Access Checker reads the robots.txt half of the OAI-SearchBot question in seconds.