On this page
When ChatGPT recommends a competitor, the answer is only the beginning of the investigation. The linked sources can show which explanations, comparisons and independent references are supporting that response.
A citation source audit records those pages across a defined set of questions. It helps you decide whether to improve your own content, investigate an external publication or leave a result alone. The useful output is a short list of evidence-backed actions.
Begin with one buying decision
Choose a decision your customers actually make. For a booking product, that might be choosing software for a small tour company. For a hotel, it might be finding somewhere suitable for a family without a car.
Write a small set of questions covering different requirements within that decision. Ten is a manageable starting point, not a measurement standard. Keep questions that name your business in a separate group from questions that ask for options generally.
Record the intended market, language, date and the environment used to collect answers. Keep a manual ChatGPT session separate from an API-based monitoring run. Changing the collection method halfway through makes the source comparison harder to interpret.
Record explicit citations at page level
For each completed answer, save the question, answer text and explicitly cited URLs. A list of pages returned by a search tool is different from links actually cited in the response.
Open the cited page and record what it contributes. Does it establish a product feature, compare several businesses, supply a statistic or explain the wider topic? If you cannot open it, mark it unavailable rather than guessing from its title.
Use a simple worksheet:
| Field | What to record |
|---|---|
| Question | Exact wording and intended customer |
| Observation | Collection date and provider or interface |
| Cited URL | The specific page, preserving the original link |
| Source type | Vendor page, directory, review, publication or research |
| Contribution | The claim or topic the page appears to support |
| Brand relationship | Your business, a competitor or neither |
| Next action | Investigate, improve, approach or leave alone |
Preserve original URLs before grouping redirects or tracking variants. Two URLs that look similar can point to different material. Only merge them after checking their destinations and content.
Count appearances without inflating them
Use one consistent counting rule. For a first audit, count whether a domain appears at least once in each answer. Count specific pages separately.
Suppose a completed answer links to three pages from the same directory. That is one answer containing the directory and three distinct cited pages. Treating it as three separate answer appearances would exaggerate that directory's coverage.
Also retain answers with no explicit citations. They are part of the completed sample. Failed collections belong in a separate completeness count.
An illustrative result could read: “This directory appeared in 6 of 20 completed answers, using four distinct pages.” That tells a reviewer far more than “the directory has strong AI authority.” It describes your sample without claiming a universal ranking signal.
Read the recurring pages closely
Start with pages that recur across several relevant questions. Compare their information with what a customer can find on your website.
Imagine that answers about tour booking repeatedly cite a comparison explaining deposits, cancellation rules and group capacity. Your own page says only that the product makes bookings easier. The actionable gap is the missing explanation of those specific capabilities, assuming your product supports them.
Do not copy the comparison's wording or reproduce its structure mechanically. Ask what additional evidence you can contribute: a worked booking example, an actual setup guide or a clear description of a limitation.
A source audit reveals an information gap to investigate. It does not establish why the model selected that page or prove that matching it will earn a citation.
Turn findings into three kinds of work
Improve an owned page when the cited material answers an important question your website leaves unresolved. Choose the closest existing destination before planning another article.
Investigate an external source when a relevant directory or publication describes your business inaccurately or has a legitimate editorial submission process. Recurrence is a reason to inspect it, not a reason to buy a placement.
Keep watching when the page is relevant but there is no clear information gap or practical action. A competitor's appearance does not automatically justify another content project.
For each selected action, write down the original observation and what you expect to learn. “Explain our deposit options on the booking page” is reviewable. “Improve AI authority” is too vague to assign.
Repeat the audit on a comparable sample
Revisit the same questions after the work has been published and had time to be discovered. Save both sets of answers. Look for changes in cited pages, brand descriptions and source types, while checking whether the collection conditions changed.
Readable content remains a prerequisite worth checking. OpenAI identifies OAI-SearchBot as its search crawler; GPTBot has a separate training role. OpenAI crawler documentation
Use Prerender Buddy AI Visibility to inspect recorded answers and citations. Start with one important question and follow the sources behind it.