Can AI use my content without citing me?

AI can use your content without citing you, and it happens in three distinct ways.

• Retrieved but not cited. The engine fetches your page for a query, reads it, and then cites a different subset of sources.
• Training data. Your content was in a model's training corpus. The model answers from memory, with no live lookup and no link.
• Answer-time fetches. A user asks the assistant to open a page. That fetch may feed the answer without appearing as a citation.

The reason is structural. A citation is something the model emits while writing the answer. It is not a receipt for every document that shaped it. Our guide on cited vs retrieved sources walks through the difference in detail.

What this means for you: citation counts undercount influence. The next move is to decide which of the three channels you want to allow, then set the controls that govern each one.

OpenAI's own documentation

OpenAI's Responses API returns two lists for web search: the sources the model consulted and the citations it placed in the answer. OpenAI's documentation states that "the number of sources is often greater than the number of citations." In other words, some pages are read and then dropped from the credit line. This is the set we define as retrieved sources.

Mechanistic research

A 2026 study of how models cite, run on Llama-3.1-8B (arXiv:2606.28358), found citation is produced by a distributed set of internal components. Amplifying specific attention heads and MLPs repaired over 90% of missed citations and removed 69% of spurious ones, without changing the sources and without harming answer accuracy. Citation behaviour can move while the evidence the model used stays fixed.

Platforms cite in different units

As of October 2026, Anthropic's API cites short text spans, while Perplexity cites whole sources by number. The Claude web search tool documentation covers the span format. A citation on one platform is not the same unit as a citation on another, so counts do not compare cleanly.

When an answer comes from the model's memory, there is nothing to cite. The content shaped the model's weights, not the answer's source list.

That memory has a cutoff. OpenAI lists a February 16, 2026 knowledge cutoff for the GPT-5.6 family and August 31, 2025 for the Instant path ChatGPT also uses. Anything you published after the cutoff cannot reach a retrieval-off answer from that model.

Training crawlers are separate bots with separate rules:

• GPTBot (OpenAI). Disallowing it "indicates a site's content should not be used in training", per OpenAI's bots page.
• ClaudeBot (Anthropic) collects training data. Claude-SearchBot and Claude-User do other jobs.
• Google-Extended covers training and grounding for Gemini Apps and Vertex AI. Google says it does not affect inclusion in Google Search.

Be plain about the limit: nobody, including the vendors, has measured whether allowing or blocking training crawlers changes what a model says about your brand. Blocking expresses a preference for future training. It does not remove what an existing model already learned.

Every control below is per bot or per page. None of them makes an engine cite you. They decide whether your content can be used at all.

A training opt-out that keeps search open looks like this:

User-agent: GPTBot
Disallow: /

User-agent: OAI-SearchBot
Allow: /

OpenAI states each setting "is independent of the others." On Bing, content marked data-nosnippet is still indexed but excluded from snippets and AI summaries.

• User-triggered fetchers. OpenAI says robots.txt rules "may not apply" to ChatGPT-User. Perplexity says Perplexity-User "generally ignores robots.txt rules." When a user pastes your URL, robots.txt is not a reliable block.
• Google has no AI-only robots.txt token. Googlebot rules govern AI Overviews and AI Mode along with regular Search. The Search Console generative AI control is opt-out and is still reaching only some site owners.
• Your firewall may override you. A CDN or bot-management rule can block a crawler your robots.txt allows. Then you lose the credited path while answer-time and training channels may continue.
• Crawler visits are not citations. Heavy crawling with no credit usually reflects training bots, which have no referral surface. See why AI crawlers visit but never cite.

The trade-off is real. Blocking a search bot to stop uncredited use also removes your chance of being cited on that surface.

1. List which bots your robots.txt allows, and separate training bots from search bots.
2. Decide per bot. A common choice is to block training and allow search, since search is the channel that can carry a link.
3. Check that your firewall lets the allowed search bots through. A 403 to OAI-SearchBot silently removes you from ChatGPT search.
4. Audit snippet directives for accidents. A template-wide nosnippet or data-nosnippet can hide content you wanted shown.
5. Measure the gap. The retrieved vs cited logging playbook shows how to capture both lists through the API.

To see whether ChatGPT names you at all for real buyer questions, the free checker on this site runs up to three and scores the answers. Treat it as a snapshot, not proof of use.