What does an AI citation tell you?
An AI citation tells you that a particular answer presents a source as support. To verify the observation, open the cited link and check whether the page supports the statement beside it. Record the answer and URL together. A source reference alone does not establish that the answer is accurate.
In this glossary, an unlinked mention of a business is recorded separately as a brand mention. That distinction matters when someone reports that your business “appeared in AI.” Was it named in the answer, linked as a source, or recommended as a supplier? Those observations answer different questions.
This is also different from a local SEO citation: a listing or mention of a business's name, address and phone details. An AI citation concerns a source in a generated answer, not a directory listing.
Keep citations, mentions and outcomes separate
| Observation | Evidence to keep | What it does not prove |
|---|---|---|
| Source citation | The answer, source reference and exact linked URL. | That anyone clicked the link or endorsed the business. |
| Brand mention | The passage naming the business and its context. | That the business's own website was used as a source. |
| Referral visit | An observed visit and the available referral information. | Every citation impression, or the exact answer the visitor saw. |
| Qualified enquiry or signup | A recorded outcome with available attribution. | That a content edit caused the outcome, without further evidence. |
Report missing attribution as unknown. Do not substitute a citation count for visits, or assign a universal click-through rate to estimate leads. This entry does not claim that AI referrals convert better than other traffic sources.
Access rules depend on the platform
There is no single crawler switch covering every use of content by AI products. Search discovery, model training and a user-requested page visit are separate purposes. Check the documentation for the specific platform and agent before changing access controls.
Google AI Overviews and AI Mode
Google requires a page to be indexed and eligible to appear in Search with a snippet before it can be a supporting link in AI Overviews or AI Mode. Meeting those requirements does not guarantee inclusion. Google also says there is no special schema.org markup or new AI text file required for these features. See Google's AI features documentation.
Google reports traffic from AI features within Search Console's Performance report under the Web search type. Do not label all clicks to a page as AI Overview clicks. For eligibility and preview-control details, read our AI Overviews definition.
ChatGPT search and OpenAI crawlers
OpenAI documents OAI-SearchBot for search discovery and GPTBot for content that may be used in model training. Their settings are independent: allowing search access does not require allowing training. ChatGPT-User supports user-triggered actions, not automatic search crawling, and robots.txt rules may not apply to those actions. See OpenAI's crawler documentation.
Perplexity search and user requests
PerplexityBot is intended to surface and link websites in Perplexity search results, not to collect content for foundation-model training. Perplexity-User handles user-triggered visits and generally ignores robots.txt. Perplexity recommends checking both user-agent strings and its published IP ranges when allowing bots through a firewall. See Perplexity's crawler documentation.
These access rules describe documented behavior, not a recipe for earning citations. Do not remove security protections across a whole website or change its training permissions solely because a citation report shows a gap.
A hypothetical citation check
Hypothetical example, not a customer result: an agency asks an AI search tool how to review a draft before publishing. The answer links to a software company's guide. The agency opens the link and finds that the guide explains fact-checking and approval, but not the automatic compliance check claimed in the answer.
The agency can record that the guide was cited in that response. It cannot honestly describe the citation as proof that the product provides automatic compliance checks. It should preserve the mismatch in its notes. Without referral and outcome records, it also cannot claim the citation produced a visit or a lead.
Build a citation log you can review
Use a repeatable sample of real customer questions to investigate how your content appears. This is a proposed measurement workflow, not a benchmark we have performed or a claim of complete visibility across an AI platform.
- State the question: save the exact prompt, intended reader task and any follow-up context.
- Record the conditions: platform, search mode, date, language, market and account settings available to you.
- Preserve the answer: capture the response and source references, not just a dashboard's mention total.
- Check the source: open the exact URL and compare its content with the claim it supposedly supports. Record redirects or inaccessible pages.
- Separate results: report citations, unlinked mentions, referral visits and qualified outcomes in distinct fields. Keep failed or unavailable observations separate from observed absence.
A citation share calculated from a chosen prompt set describes that sample, not every query or user. If you change the prompt set, platform mode or market, document the change before comparing results. A competitor appearing in one response is a research lead, not proof that your page is losing all relevant demand.
Improve the source page, not a citation quota
Start with what the reader needs to decide or do. Correct unsupported statements, identify original evidence, explain practical limitations and make important information available as text. Google specifically recommends useful content, internal links and structured data that matches the visible page.
Do not add statistics, fake tests or FAQ markup just to look authoritative. Publishing more versions of the same answer is not evidence of increased usefulness. If an existing page already owns the reader's task, improve it before proposing another URL. Nor does placing an answer within an arbitrary word limit establish that a platform will select it.
Frequently asked questions
Does a citation mean an AI system recommends my business?
No. A source reference and a supplier recommendation are different observations. Read the passage around the citation and open the linked page. The answer might cite your definition without evaluating your product at all. Record the wording and context before describing the appearance as a recommendation, endorsement or customer outcome.
Do I need FAQ schema to get cited?
Google says no special schema.org markup is required for AI Overviews or AI Mode. Use structured data only when it accurately describes visible content. That Google rule does not establish how every other platform selects sources. A valid FAQ block is not evidence that a page will receive a citation.
Does blocking GPTBot prevent ChatGPT search citations?
OpenAI documents GPTBot and OAI-SearchBot as independent controls. GPTBot concerns content that may be used in training; OAI-SearchBot concerns search discovery. Review both robots.txt rules and infrastructure access for the relevant agent. Allowing a search crawler is not a promise that the page will be selected for an answer.
Connect the audit to content work
Choose an important source page and review the claim that a generated answer cited. If the page needs correction, fix its evidence and explanation before increasing publication volume. For draft production and CMS publishing, explore theStacc's content SEO workflow. Check that its controls fit your review process.
Sign up for free → or book a demo to discuss the workflow. Neither action guarantees search inclusion, citations or leads.
Sources and correction note
- Google Search Central: AI features and your website.
- OpenAI: overview of crawlers.
- Perplexity: crawler documentation.
Correction: this revision removes unsupported traffic figures, customer examples presented as real results, fixed retrieval counts and claims that publishing volume or specific markup guarantees citations. It distinguishes search crawlers from training controls. This automated revision is not attributed to a named human who has not approved it.