How to Investigate Duplicate Pages in AI Search Citations

If an AI answer cites your old campaign page instead of your current guide, investigate the pages before blaming the answer engine. Duplicate or near-duplicate URLs may make it less clear which page represents the content. But a missing citation—or a citation to the wrong URL—does not establish that duplication caused the choice.

Search Engine Journal reports Microsoft guidance that LLM experiences grounded in search indexes may group near-duplicate URLs and choose one representative, potentially an outdated or unintended version. That is a reason to inspect overlapping pages, not a demonstrated selection rule for every AI-search product or mode.

Map the URL family first

Start with the cited URL and the URL you wanted cited. Then add pages that answer the same question with substantially similar copy: old campaign pages, regional versions and URLs with parameters are plausible candidates. Read the pages rather than relying on their titles alone. A short table makes the differences—and the gaps in your evidence—visible.

Here is a hypothetical returns-policy investigation. The entries describe what a publisher might find, not observed search or AI-answer behavior:

URL	Purpose	Overlapping answer text	Material differences	Redirect destination	Declared canonical
example.com/returns	Current general policy	Explains the return window and process	Current terms and exceptions	None	Self
example.com/spring-returns	Expired campaign explanation	Repeats most of the general process	Old promotion dates	None	None
example.com/returns?campaign=spring	Campaign-tracked visit to the general policy	Same answer as /returns	Tracking parameter only	None	/returns
example.com/uk/returns	UK policy	Shares the basic process	Different local terms and exceptions	None	Self

For each real URL, note the precise passage that answers the query. Does the old page still state a valid exception, or does it merely repeat the current page with expired dates? Does the regional page explain a difference readers need? Those questions separate a substantively equivalent page from a useful variant. Search Engine Journal’s report of Microsoft guidance makes the same distinction for localization: meaningful differences can include terminology, examples, regulations or product details.

Check each URL’s current destination and declared canonical separately. A redirect tells you where a request goes; a canonical declaration tells you which URL the retained page identifies as preferred. Neither check tells you which page an answer engine retrieved or why it displayed a citation.

Keep indexing findings in their own column

Where applicable, review the URL family in Google Search Console’s Page Indexing report, including duplicate findings and cases labeled “Duplicate, Google chose different canonical than user.” Record the inspected URL, the finding and the date. The duplicate-content investigation advice here points publishers to those Search Console findings; Ahrefs’ explanation of canonicalization describes Google selecting a representative URL for search results.

That can answer a useful, narrower question: how Google treated those URLs in its index. It cannot, by itself, explain a citation in another product. Keep four records distinct: the pages’ actual overlap, their redirect and canonical behavior, Google’s indexing findings, and the URL visibly cited in the AI answer. For example, “Google selected /returns as canonical” and “the answer cited /spring-returns” are observations worth preserving together, not proof that one caused the other.

Fix the pages according to their purpose

Choose a primary page only after deciding what each variant is for. If an expired campaign page has no distinct job, move any still-useful explanation into the current page and redirect the old URL to it. If a variant must remain available but repeats the primary answer, review its canonical declaration. Search Engine Journal reports Microsoft’s recommendation to choose a primary page for slightly different campaign variants, use canonicals for retained variants and consolidate older pages without a distinct purpose. The implementation advice in this duplicate-content guide likewise reserves permanent redirects for a version that definitively supersedes another.

In the hypothetical table, that might mean redirecting /spring-returns to /returns, retaining the tracked URL with its declared canonical, and keeping /uk/returns distinct because its local exceptions answer a different reader need. Do not erase a meaningful regional difference merely to make the URL list shorter. Equally, do not preserve an expired answer just because it once had its own campaign URL.

After making changes, request the superseded URLs and confirm they reach the intended destination. Check the canonical declarations on variants you retained, and reread the primary page to ensure it contains the answer you meant to preserve. These are implementation checks, not confirmation that an index has consolidated the pages or that an AI product will cite the primary URL.

If you check the answer again, record the product, mode, query, date and displayed citation URL. Compare that observation with your earlier record without assigning a cause the record cannot show. The useful outcome of this investigation may be a cleaner, more purposeful URL family—even if the next answer still cites something else.