How to get cited by ChatGPT

To get cited by ChatGPT, your page must be reachable by OpenAI's OAI-SearchBot crawler, must answer a question ChatGPT would search for, and must contain a passage clean enough to lift and attribute. Everything in this guide serves one of those three conditions.

There are two separate ways your brand can show up in a ChatGPT answer, and they work completely differently. The first is retrieval: ChatGPT searches the web, reads pages, and cites the ones it used. The second is memory: the model answers from its training data, mentions brands it learned about, and cites nothing. You can influence both, but the tactics differ, so it pays to keep them apart.

When ChatGPT actually cites sources

This is the part most advice skips. ChatGPT only produces citations when it decides to run a web search. For questions it believes it can answer from memory, it simply answers, and no website gets credit. Observed behavior, and OpenAI's own descriptions of its search feature, point to retrieval firing most often for queries that are fresh, specific, local, price-sensitive, or comparative: things the model cannot safely answer from a training snapshot.

The practical consequence: you earn ChatGPT citations on the kinds of questions that trigger search. Pages that answer timeless, encyclopedic questions face memory, not retrieval. Pages that answer current, specific, decision-stage questions face retrieval, where a well-structured page can win. That is where to aim your content.

Let the right crawler in

OpenAI documents three separate crawlers, and mixing them up is the most common technical mistake in this space:

• OAI-SearchBot powers ChatGPT search results and link citations. If it cannot fetch your pages, you cannot be cited. Allow it.
• GPTBot collects training data. Blocking it opts you out of training but, per OpenAI's documentation, does not remove you from search citations.
• ChatGPT-User fetches a page when a user directly asks about a specific URL. Blocking it breaks those on-demand lookups.

Check your robots.txt today. A blanket rule added during the 2023 wave of AI-crawler blocking may still be silently excluding you from a surface you now want to win. The minimal safe configuration is:

User-agent: OAI-SearchBot
Allow: /

Beyond robots.txt, the same technical hygiene that serves Googlebot serves OAI-SearchBot: server-rendered content, fast responses, no login walls in front of your best answers, and a firewall that does not challenge crawlers. A technical audit catches most of these in one pass.

How ChatGPT finds its sources

OpenAI has published little about how ChatGPT search ranks pages, so honesty requires separating what is documented from what is observed. Documented: OAI-SearchBot crawls the web to build search results, and answers cite the pages the model actually used. Observed by practitioners: cited sources skew toward pages that already rank in traditional search, and OpenAI has worked with Bing as a search partner, so presence and standing in the classic indexes appear to carry over. Nobody outside OpenAI can tell you the exact weighting.

The takeaway is reassuring rather than mysterious: classic SEO fundamentals still feed the machine. A page that is indexed, authoritative, and ranking for a query is the raw material ChatGPT retrieves. ChatGPT SEO is not a replacement discipline, it is an extension of one you already practice.

Write passages ChatGPT can lift

Retrieval gets you into the candidate pool. Structure decides whether your passage is the one quoted. The model needs a span of text it can extract, trust, and attribute without dragging along surrounding context, so build every important section to stand alone:

• Open each section with a direct two or three sentence answer, before background and caveats.
• Use question-phrased headings that mirror real prompts, then resolve them immediately below.
• Name entities explicitly. "SmashSERP tracks AI citations" is quotable; "our tool does this too" is not.
• Include specifics a model can repeat with confidence: numbers, dates, definitions, and steps.
• Add an FAQ with FAQPage schema so the question and answer pairs are machine-readable.

This is the same answer-first pattern that wins Google AI Overviews and Perplexity citations, which is the quiet good news of GEO: one disciplined structure serves every engine.

Build the model's memory of your brand

The second channel is slower and less controllable, and it is worth being straight about that. When ChatGPT answers from training data, it can only mention brands the model learned about. What plausibly shapes that memory is the corpus of text about you across the web: reviews, directories, comparison posts, press, forum threads, and consistent descriptions of what you do. This is assumption territory, because training data selection is not public, but it is the only lever that exists for memory-based answers, and it compounds with everything else you do for authority.

Two practical rules follow. Describe your product the same way everywhere, so the model learns one clear sentence about you rather than five conflicting ones. And invest in being mentioned by third parties, because a model trained on the web trusts the web's consensus more than your homepage.

Verify citations instead of guessing

Most teams doing GEO have no idea whether it is working, because checking means manually prompting ChatGPT and squinting at the sources. SmashSERP does this as a product feature: it runs real checks against ChatGPT, Perplexity, Gemini, and Google AI Overviews for the prompts that matter to you, records whether your site is cited, and captures the exact passages the AI quoted. That last part changes behavior, because once you see which paragraph got lifted, you know exactly what shape of writing to produce more of. See how it fits the broader practice on our GEO and AI visibility page, or start with the full playbook for getting cited by AI.