ChatGPT cites a page when three things line up: OAI-SearchBot can actually reach it, Bing’s index returns it for a query ChatGPT bothers to search, and the page itself reads as a direct answer rather than a summary of one. The middle part matters more than most advice admits — ChatGPT only runs a live web search for a minority of its answers, so being crawlable and indexed outranks any single on-page trick.
Key takeaways
- ChatGPT searched the web for only about 14.4% of 66.6 million sampled responses, Ahrefs, 2026 — most answers never get a chance to cite anyone.
- GPTBot and OAI-SearchBot are different crawlers with different jobs, and OpenAI’s own documentation, 2026 says blocking one has no effect on the other.
- Citations concentrate hard: Wikipedia and Reddit alone took roughly a quarter of all ChatGPT citations in early 2026, per Similarweb, Jan–Feb 2026.
- That concentration is already shifting: Reddit’s and Wikipedia’s citation share roughly halved between March and September 2026 as ChatGPT leaned more on primary sources like PubMed and the FDA, per the same Ahrefs study.
- Adding schema markup, on its own, produced no statistically meaningful citation gain on ChatGPT, Ahrefs, 2025–2026 — the common advice to “add schema to get cited” doesn’t hold up on its own.
Does ChatGPT actually search the web before answering?
Rarely, by default. In the largest public sample to date, Ahrefs found that only about 14.4% of 66.6 million ChatGPT responses collected in August and September 2026 triggered a live web search at all, averaging roughly 1.6 background search queries when it did. That means the first question isn’t “how do I get cited” — it’s “does this specific query even make ChatGPT search.”
ChatGPT decides to search based on the question itself: anything time-sensitive, local, or about a specific named thing (a product, a company, a recent event) is more likely to trigger a search than a general “how does X work” question the model can answer from what it already learned during training. If a prompt doesn’t make ChatGPT search, no amount of on-page optimization changes the outcome — there’s no retrieval step for your page to win.
What’s the difference between GPTBot, OAI-SearchBot, and ChatGPT-User?
They’re three separate crawlers with three separate jobs, and controlling one in robots.txt does nothing to the others. Here’s how OpenAI’s own developer documentation describes each:
| Crawler | What it does | Blocking it means |
|---|---|---|
| GPTBot | Gathers content used to train OpenAI’s models | Your content stays out of future model training — no effect on ChatGPT search |
| OAI-SearchBot | Indexes pages so ChatGPT can retrieve and cite them in search answers | You can disappear from ChatGPT search results while staying eligible for training |
| ChatGPT-User | Fetches a page live when a user’s own prompt or a plugin asks for it in real time | Real-time browsing requests on your pages stop working |
This split dates back to when OpenAI first added source links to ChatGPT, on November 1, 2024, Search Engine Roundtable — the three-bot structure hasn’t changed since, only the scale of what each one does. OpenAI’s current documentation also notes that robots.txt changes can take roughly 24 hours to take effect, so a fix you make today won’t show up in ChatGPT search results immediately.
Which sites does ChatGPT actually cite, and why?
A small number of large, trusted domains take an outsized share, and that share is shifting toward sources that originate information rather than summarize it. Similarweb’s analysis of citation data from January and February 2026 found Wikipedia took 13.15% of citations (118,285 instances) and Reddit took 11.97% (107,680), with OpenAI’s own site third at 6.21% and Walmart fourth at 2.90%. Citations rarely land on homepages — most come from pages several folders deep that cover one specific topic in detail.
That picture moved during 2026. Ahrefs’ later August–September sample found Reddit’s and Wikipedia’s share of answers citing them roughly halved compared with March — Reddit from about 1 in 5 answers down to about 1 in 43, Wikipedia from about 1 in 7 down to about 1 in 35 — while original sources climbed fast: PubMed went from the 64th most-cited domain to 4th, and the FDA entered the top 200 at 8th. WebMD, a summarizer of medical research rather than a primary source, dropped from 18th to 79th in the same window. The direction is consistent: ChatGPT is increasingly rewarding the page that did the work over the page that explained it.
Does adding schema markup get you cited?
Not by itself. Ahrefs tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026 against 4,000 matched control pages that didn’t, using a difference-in-differences design to isolate schema’s own effect. The result on ChatGPT citations was a +2.2% change — statistically indistinguishable from random noise. Google’s AI Overviews actually saw a measurable −4.6% decline on the same pages. The study’s own caveat matters: it only covered pages that already had baseline citation activity, so it says nothing about whether schema helps a page go from invisible to indexed in the first place — just that it won’t move the needle once you’re already in the pool.
How do you check whether ChatGPT can even reach your page?
Run through this before touching the content itself — it catches the cases where no amount of rewriting would have worked anyway.
- Read your live robots.txt. Look for a
User-agent: OAI-SearchBotblock specifically — a rule aimed at GPTBot doesn’t cover it, and the reverse is also true. - Confirm the page isn’t behind a login, a paywall, or a client-rendered shell with no content in the initial HTML — crawlers that don’t execute JavaScript the way a browser does will see an empty page.
- Check whether the page is in Bing’s index (not just Google’s) — ChatGPT’s live search runs on Bing’s index, so a page that ranks well on Google but has never been crawled by Bing has nothing to be retrieved from.
- Submit the URL through Bing Webmaster Tools and IndexNow if it’s new or recently changed, rather than waiting for an organic crawl.
- Open the page and find the one paragraph that answers a specific question completely on its own — if every claim needs the paragraph above it for context, there’s nothing a retrieval system can lift out cleanly as a citation-worthy passage.
Frequently asked questions
Does blocking GPTBot hurt my chances of being cited by ChatGPT?
No. GPTBot only feeds model training, which is a separate process from the live search ChatGPT runs when it decides to look something up. Blocking GPTBot keeps your content out of future training data, but OAI-SearchBot — the crawler that actually matters for citations — is controlled independently in robots.txt.
Is ChatGPT’s citation behavior the same as Google’s AI Overviews?
No, they’re separate systems with separate indexes and separate crawlers. ChatGPT’s live search draws on Bing’s index through OAI-SearchBot, while Google’s AI Overviews draws on Google’s own index and crawler. The Ahrefs schema study found opposite directional effects on the two platforms on the same set of pages, which is itself evidence they don’t behave identically.
Why does ChatGPT sometimes answer with no citations at all?
Because it didn’t search the web for that answer. In the 2026 Ahrefs sample, roughly 85.6% of responses had no live search behind them at all — the model answered from what it learned during training, which produces no sources to cite because none were consulted.
Do I need FAQPage or Article schema to be cited by ChatGPT?
Schema makes a page easier for any system to parse correctly, which is worth doing for its own sake, but the 2025–2026 Ahrefs study found it produced no statistically significant citation increase on ChatGPT by itself. Treat it as good hygiene, not a citation strategy.
How long does it take for a robots.txt change to affect ChatGPT search?
OpenAI’s own documentation gives roughly 24 hours for crawl-permission changes to take effect. Whether the page then gets cited is a separate question that depends on Bing indexing it and ChatGPT choosing to search for a query it matches.
Does writing for Reddit or Wikipedia-style pages help, since they get cited most?
Not directly — you can’t publish on Wikipedia or Reddit as a business. What the 2026 shift shows is that ChatGPT is moving its citations toward primary, first-hand sources on specific topics rather than general summarizer sites, which is a better argument for publishing your own original data or documentation than for imitating a forum’s format.
None of this is something you have to work out alone from a blog post. If you want to know specifically whether OAI-SearchBot can reach your own pages today and where your retrieval gaps actually are, TheMarkElite runs a free, no-obligation 30-minute diagnostic that checks exactly the crawler-access and indexing questions covered above against your own site, with a prioritized list sent back within 48 hours — start a diagnostic.
For the parallel question on Google’s side, see our look at how to appear in AI Overviews, and for general ChatGPT search visibility (not just citations), our piece on ranking in ChatGPT search results. For the brand-recommendation angle instead of citation mechanics, see how to get your business recommended by ChatGPT. Browse more in TheMarkElite’s insights hub.
Last updated: November 12, 2026. Written by the TheMarkElite team.