How to Get Cited by ChatGPT: Sources, Signals and Tests

How do you get cited by ChatGPT, and how do you test whether it is working?

Updated October 31, 2026

To get cited by ChatGPT, a page must first be retrieved in a web search, then chosen from the results, and then measured. OpenAI documents the retrieval side and its crawlers. The selection side comes from third-party studies, so treat those figures as evidence to test, not as rules.

Ahrefs found that ChatGPT cited about half of the URLs it retrieved, so retrieval alone does not earn a citation.

Getting Retrieved Comes Before Getting Cited by ChatGPT

Decide first which sub-questions your page must match. ChatGPT search sometimes partners with other search providers, and it rewrites your question into one or more targeted queries that it sends to them.

After it reviews the first results, ChatGPT search may send more specific queries to other providers. A page that matches only the head term can miss those follow-up queries.

ChatGPT may search the web on its own when a question would benefit from current information. Test with prompts that a buyer asks now, not only with evergreen definitions.

OpenAI's documents name third-party search providers. Its help article links the privacy policies of Microsoft and Shopify, but it does not say in so many words that Bing supplies the results. OpenAI does not name Bing, while Ahrefs calls Bing ChatGPT's built-in provider. Treat the Bing role as unconfirmed, and plan around the providers OpenAI documents.

The search ecosystems page covers the Bing and Google index background, which matters when you decide where retrieval effort goes.

Which OpenAI Bots to Allow and Which to Block

Allow OAI-SearchBot if you want your pages in ChatGPT search. Set GPTBot separately, because it controls training, not search appearance.

OpenAI says OAI-SearchBot is the robot used to surface websites in ChatGPT search features. A site that opts out of it is not shown in ChatGPT search answers, though it can still appear as a navigational link.

OpenAI recommends allowing OAI-SearchBot in robots.txt and allowing requests from its published IP ranges. The ranges come as a JSON file, so a log line that claims to be OAI-SearchBot can be tested against them.

A robots.txt change can take about 24 hours to take effect for search. Plan a fix that depends on crawl access with that lag in mind.

The two settings are independent. OpenAI's worked example allows OAI-SearchBot so the site can appear in results, and disallows GPTBot to signal that its content should not be used for training. This is the pattern from that example.

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

If a site allows both bots, OpenAI may use one crawl for both purposes. GPTBot is the training crawler, so disallowing it signals that content should not be used to train generative AI foundation models.

ChatGPT-User is a separate agent for visits that a user starts in ChatGPT or a Custom GPT. OpenAI says it is not used to determine whether content may appear in Search. OAI-AdsBot visits only pages submitted as ads, so it is not a citation signal.

Do not combine a robots.txt Disallow and a noindex tag on the same URL. The crawler has to fetch a page to read its meta tag, so the Disallow blocks the tag. OpenAI says it may still surface just the link and page title of a disallowed page, if it gets the URL from a third-party search provider or from other pages.

OpenAI says any public website can appear in ChatGPT search. Its launch post names 14 publisher partners and says any website or publisher can choose to appear. No sign-up step for ordinary sites appears in OpenAI's documents, so for most sites the lever is OAI-SearchBot access.

The GEO and AEO guide covers the wider crawler list for GPTBot, ClaudeBot and CCBot.

The SEO fundamentals page and the complete SEO checklist cover the crawl and index baseline that these settings sit on.

Retrieved Does Not Mean Cited: What the Studies Measured

Do not set a citation target from one study's rate. Each study uses a different base: all retrieved URLs, the top 10 Google results for the original prompt, or the general search channel.

Ahrefs' April 2026 study of 1.4 million prompts found that ChatGPT ends up citing about 50% of the URLs it retrieves. Ahrefs also says some non-cited URLs were likely never opened, so the rate depends on what counts as retrieved.

Ahrefs also reports citation rates by retrieval channel. URLs taken from general search were cited 88.46% of the time across 25,563,589 data points, and Reddit URLs were cited 1.93% of the time.

Reddit makes up 67.8% of the non-cited URLs in that study. Ahrefs says the gap between cited and non-cited URLs is a composition effect of that imbalance, so it cannot say whether snippet or publication-date fields matter.

Ahrefs' August 2025 overlap study found that 8.00% of ChatGPT in-text citations overlapped Google's top 10 for the original prompt. The 88% search-channel figure measures something else: presence in the search channel after query fan-out. The 8.00% comes from early-July 2025 data, and the 88% comes from 1.4 million ChatGPT 5.2 prompts dated February 2025 in the Ahrefs 2026 study, so the prompt sets and model versions also differ. Do not put the two figures in one sentence as the same metric.

AirOps, in a separate analysis of 548,534 retrieved pages across 15,000 prompts, reports that only 15% of retrieved pages were cited. Its sample and its definition of retrieved differ from the Ahrefs study, so the two shares do not measure the same thing.

Ahrefs explains the low overlap with query fan-out. A page that ranks lower for several variants of a query can be cited over a page that ranks first for only one variant.

Ahrefs' fan-out study of 66.6 million free, logged-out ChatGPT responses found that 14.4% contained fan-out queries. Responses with fan-outs averaged about 1.6 fan-out queries.

Rank still matters, because most citations come through the search channel. Ahrefs concludes that you need to be in the search selection pool, which means ranking. AirOps reports that pages ranking first in Google were cited 3.5 times more often than pages outside the top 20.

Which Page Traits Get Cited by ChatGPT, and How Strong the Evidence Is

Treat each trait as a hypothesis to test on your own pages. The title and slug findings come from Ahrefs' large study. The capsule and original-data findings rest on a small audit.

Trait	What the study found	Sample and limit
Title match to sub-questions	Titles of cited URLs scored 0.602 similarity to the prompt, against 0.484 for non-cited URLs. The best match to any fan-out query scored 0.656 for cited URLs.	Ahrefs, 1.4 million prompts. The scores come from open-source embedding models, an approximation of ChatGPT's scoring.
Natural-language URL slug	In the search channel, URLs with natural-language slugs had an 89.78% citation rate, against 81.11% without.	Ahrefs, 1.4 million prompts. A comparison inside one channel. It does not show that changing a slug causes a citation.
Answer capsule	Most cited blog posts in one audit had an identifiable answer capsule, placed directly after a title or a question-framed H2.	Search Engine Land audit of cited blog posts. Small sample, built from referral sessions, with the capsule definition set by the author. It had no matched group of uncited posts.
Links inside the capsule	Most capsules in that audit had no links. The author reads this as links diluting quotability.	Same audit. The data is a share of cited posts, not a controlled test.
Original data or owned insight	The audit tracked original data and owned insight as a separate trait. Other write-ups report different shares for it, because they define the trait differently.	Same audit, with the same limits.
Length	Longer pages earned more citations on average. Finance ran the other way, with shorter, denser pages doing better.	Search Engine Land reporting on Kevin Indig's study, across several verticals.
Position on the page	ChatGPT cited heavily from the upper part of a page. The bottom of a page earned few citations, and conclusions were largely ignored.	Same study, as reported by Search Engine Land.
Page age	The median cited page in the search channel is about 500 days old (about 1.3 years). Some cited pages are over 2,700 days old.	Ahrefs, 1.4 million prompts. Ahrefs cannot say whether date fields matter, so do not treat age as a lever on its own.
Recurring pages	Most cited URLs were cited only once. Pages cited again were mostly category roundups, comparison pages or broad guides that answer several related questions.	Search Engine Land reporting on Kevin Indig's study.

Across all pages, Similarweb reports Wikipedia at 13.15% and Reddit at 11.97% of ChatGPT web-browsing citations. The sample is nearly 600,000 citation events in the US, from January to February 2026.

That is a share of all citations, not a rate for any one page. Do not read it as a tactic to copy.

A Repeatable Test to Check Whether ChatGPT Cites a Page

Run one fixed prompt set, log every run the same way, and compare rates before and after each change. One run cannot show a citation rate.

Build the prompt set from the head term and the sub-questions a buyer asks. AirOps reports that most prompts triggered two or more follow-up searches.
Set clean-room conditions: signed out or a clean account, memory off, and one location. Memory can change the search query, and general location based on IP address can be shared with search providers.
Keep the same account state for every run. Signed-out users can use web search, so a signed-out run is a valid baseline.
Run each prompt many times for each tool. The brand-list study ran each prompt 60 to 100 times per tool. Use that range as a starting point, not a proven citation threshold, because the study measured brand recommendations, not links.
Log each run with the fields in the table below.
Calculate the citation rate: runs where your URL is cited, divided by total runs for that prompt. Run the same prompts for competitor URLs and compare the rates side by side.
Repeat the test on a schedule. Semrush reported Reddit in close to 60% of ChatGPT responses in early August 2025, and about 10% by mid-September 2025. That is a share of responses, not citations, but it shows how fast a figure can move.

Log field	What to record	Why it matters
Date and time	When the run happened.	Answers can change between runs and over weeks, so a single date is not a baseline.
Prompt and run number	The exact prompt text, and the number of that run.	Lets you count the runs behind each rate.
Account state	Signed in or out, and memory on or off.	Memory can change the search query that ChatGPT sends.
Location	The one location you used for every run.	General location from the IP address can shape the search.
Search used	Yes or no.	The help article says responses that use web search may include citations. Log runs with and without search as separate groups.
Cited URLs and domains	Every cited link in the answer, with its domain.	Shows which competitors are cited for the same prompt.
Your URL cited	Yes or no, for your page.	The count that feeds the citation rate.

Which Logs and Analytics Show ChatGPT Visits

Use server logs for crawler visits and GA4 for visits from ChatGPT links. Do not use Microsoft's AI report as a ChatGPT measure.

OAI-SearchBot hits: match the IP address against OpenAI's published ranges before you count the visit.
robots.txt fetches: OpenAI says these carry an extra robots.txt marker in the user agent, so you can separate them from search visits.
ChatGPT-User hits: these are visits a user starts in ChatGPT or a Custom GPT. They do not decide search appearance.
GA4 referrals: publishers who allow OAI-SearchBot can track ChatGPT referral traffic in Google Analytics. ChatGPT adds utm_source=chatgpt.com to referral URLs.
Landing pages with ChatGPT referrals: group GA4 sessions with UTM parameters, custom channel groupings and referrer path segments, then review each page by hand.
Expected size: Ahrefs' tracker of 74,752 websites found that all AI chatbots combined sent 3.5 million visitors in March 2026. That is 0.28% of total web traffic, so expect a small signal.

Microsoft's AI Performance report in Bing Webmaster Tools is in public preview. It covers Microsoft Copilot, AI summaries in Bing and select partner integrations. ChatGPT is not among the surfaces it lists.

The grounding queries in that report are a sample of citation activity. Its page-level counts do not show page importance, ranking or placement.

What to Fix First, in Order

Fix crawl access first. A page that a bot cannot fetch, or that carries a noindex tag the bot cannot read, gains nothing from better copy.

Allow OAI-SearchBot in robots.txt for the sections you want in ChatGPT search. The change can take about 24 hours to take effect.
Remove Disallow rules from pages that carry a noindex tag. Allow the crawl, and keep the noindex tag on pages that must stay out of the index.
Match each title and question-framed H2 to a sub-question a buyer asks. Titles of cited URLs score closer to the prompt than titles of non-cited URLs.
Put a direct answer straight after each question-framed H2, then test whether citations move.
Add original data or owned insight where you can support it, then test the effect on the same prompts.
Build the prompt set and log a baseline before you change more pages.
Set up the GA4 referral view and the server log check, then review both on a schedule.

Repeat the test after each change. Keep a change only when the citation rate on the same prompts moves. Use the SEO guide for practitioners to find the next page for each fix.

Common Questions

If a page ranks in the top 10 for my keyword, why might ChatGPT still not cite it?

ChatGPT rewrites the question into targeted queries before it searches, so your exact keyword is only one input. Ahrefs found that 8.00% of ChatGPT in-text citations overlapped Google's top 10 for the original prompt, in a study of 15,000 long-tail prompts. Ahrefs also found that 80% of the cited links across ChatGPT, Gemini, Copilot and Perplexity do not rank anywhere in Google for the original query. Match your titles and headings to the sub-questions a buyer asks.

Does blocking GPTBot hurt my chances of being cited in ChatGPT search?

Not according to OpenAI. Its bot documentation says each setting is independent. OAI-SearchBot controls appearance in ChatGPT search, and GPTBot controls use in training. OpenAI gives a worked example of allowing OAI-SearchBot while disallowing GPTBot. ChatGPT-User handles visits that a user starts, and OpenAI says it is not used to decide Search appearance. Keep OAI-SearchBot open, then set GPTBot as a separate content policy.

Can I use robots.txt Disallow and noindex together to keep a page out of ChatGPT?

Not reliably. OpenAI's publisher FAQ says it may still surface just the link and page title of a disallowed page, if it learns the URL from a third-party search provider or from other pages. A crawler reads a noindex tag only when it may fetch the page, so a Disallow on the same URL blocks the tag. To keep a page out, allow the crawl and add a noindex tag to the page.

How many times should I run a prompt before I trust a citation result?

One run tells you little. SparkToro and Gumshoe.ai ran each prompt 60 to 100 times per tool, and found less than a 1 in 100 chance that ChatGPT or Google's AI gives the same brand list twice. That study measured brand lists, not links, so use 60 to 100 runs as a starting point, not a proven citation threshold. Log a citation rate across many runs and prompts, with memory off and one location.

Does answer-capsule formatting really raise citation odds?

The evidence is thin. A small Search Engine Land audit of cited blog posts found that most had an identifiable answer capsule, and that few capsules contained links. The audit counted cited posts, not a matched group of uncited posts, and the capsule definition came from its author. Treat the capsule as a cheap hypothesis. Add one to a set of pages, then compare citation rates on your test prompts before and after the change.

Does the Bing Webmaster Tools AI Performance report show ChatGPT citations?

No. Microsoft describes the report as covering Microsoft Copilot, AI summaries in Bing and select partner integrations. It shows a sample of grounding queries, and it does not list ChatGPT. Page-level counts do not show page importance, ranking or placement. For ChatGPT visits, read your server logs for OAI-SearchBot and your GA4 referral data for utm_source=chatgpt.com.