How to Check Whether ChatGPT Cites Your Website

What counts as a ChatGPT citation

ChatGPT responses that use search may show inline citations beside claims and a Sources panel with cited sources and other relevant links. For measurement, count your site as cited when the answer visibly attributes supporting information to your domain or page through that citation interface.

Keep other outcomes separate. A brand mention without a source link is a mention, not a confirmed citation. A link shown only as a related or navigational result should not automatically be labeled as evidence for the answer. Similar wording is also not proof that the system used your page. Conservative labels make before-and-after reporting credible.

How to run a manual ChatGPT citation check

Start with a real question that matters to the page and audience. Use ChatGPT search or a response that clearly searches the web, then keep the query wording unchanged for the baseline. After the answer appears, inspect every inline citation and open the Sources panel. Record the exact page URL rather than counting only the root domain.

Run the check in a clean conversation so earlier messages do not reshape the task. Record any context that can affect reproducibility, such as location, language, account state, search mode, or follow-up instructions. The result describes that response at that time; it is not a universal rank across every user or future answer.

1. 1

   Choose one target query tied to a specific page and business need.

2. 2

   Open a new conversation and use a response mode that searches the web.

3. 3

   Submit the exact query without naming your brand unless brand visibility is the test.

4. 4

   Inspect inline citations beside the relevant claims.

5. 5

   Open the Sources panel and check the full source URLs.

6. 6

   Save the answer, sources, timestamp, result label, and target URL.

Record a baseline that another person can inspect

A screenshot alone is weak evidence because it can omit the query, date, or source details. Keep the response and the measurement fields together. A reviewer should be able to understand exactly what was asked, which system answered, what was cited, and why you assigned the result label.

Store direct citation and brand mention as different fields. This avoids turning any positive-looking appearance into a citation win. It also lets you report meaningful intermediate movement, such as moving from absent to mentioned before a direct source link appears.

Recheck variability without pretending it is a ranking

Generated answers can change with time, query phrasing, location, context, source freshness, and product behavior. One successful check proves that a citation appeared in that observed response; it does not establish a permanent position. One failed check also does not prove that the site can never be cited.

Use a consistent prompt set and a repeatable schedule. Compare the same provider and query after a meaningful page change or enough time for retrieval systems to refresh. Keep separate histories by provider instead of combining ChatGPT, Google AI features, Gemini, Claude, and Perplexity into one invented score.

• Keep the exact query stable for before-and-after checks.
• Record provider, mode, location, language, and timestamp where available.
• Save the answer and source list, not only a yes-or-no label.
• Track target queries that map to real pages and customer decisions.
• Report provider-specific observations rather than a universal AI rank.
• Recheck after a meaningful change, not repeatedly until a preferred answer appears.

What to check when your site is not cited

First confirm access. OpenAI documents OAI-SearchBot as the crawler used to surface websites in ChatGPT search and treats it separately from GPTBot, which is used for content that may support model training. A publisher can allow search crawling while disallowing training crawling. Also confirm that the host or CDN allows OpenAI's published search-bot IP ranges.

Access is only eligibility, not a placement guarantee. Review whether the target page directly satisfies the query, offers original or first-hand information, supports claims, uses a stable canonical URL, and is linked from relevant pages. Improve the useful page instead of publishing a thin version for every prompt variation.

1. 1

   Confirm the target page returns a successful status and is publicly accessible.

2. 2

   Check robots.txt and infrastructure access for OAI-SearchBot.

3. 3

   Verify that the target URL is canonical and internally linked.

4. 4

   Compare the page with the sources the answer actually cited.

5. 5

   Add missing original evidence, clarity, sourcing, or intent coverage.

6. 6

   Record the change and rerun the same check after a reasonable refresh period.

When AI citation tracking is worth automating

Manual checks are useful for learning the protocol and reviewing a small set of valuable queries. Automation becomes useful when the prompt set, providers, pages, and review schedule are too large to record consistently by hand. The automation should preserve evidence rather than hiding it behind one percentage.

A useful tracker stores the baseline answer, visible sources, target URL, provider, timestamp, and classification, then attaches the follow-up result to the page change that was shipped. That turns citation monitoring into a workflow: observe, diagnose, improve, recheck, and report without claiming guaranteed placement.

Primary sources

Official documentation used to review and update this guide.