How Can I Find Thin Pages That AI Assistants Cite Even Though Google Analytics Shows No Traffic?

Why would a page be cited but get no traffic?

Several reasons, most of them normal:

• The answer satisfies the question. If an assistant takes a fact from your page, such as a price or a definition, many people never click.
• Referrers are lost. Visits from AI apps and copied links often arrive as direct traffic, so GA4 does not credit the page to AI.
• Supporting citations are rarely clicked. A page listed as the fifth source under an answer gets few visits even when it is used.
• Consent and bot filtering. Visitors who refuse analytics cookies do not appear in standard reports.

None of that is a problem in itself. The problem is when the cited page is thin: a tag archive, an old announcement, a stub glossary entry, an auto-generated location page or a PDF. Then assistants build answers about you from your weakest material.

Which data sources show cited pages?

User-triggered fetches are particularly telling. When a user asks ChatGPT, Claude or Perplexity about something and the assistant reads your page, the request comes from ChatGPT-User, Claude-User or Perplexity-User. A URL that receives many of these requests is being used in answers, whatever GA4 says. The report behind the first two rows is explained in whether Bing Webmaster Tools shows AI visibility data.

How do you build the comparison?

A spreadsheet with one row per URL:

1. Export cited pages from Bing's AI Performance report and AI impressions by page from Search Console's generative AI report, for the same period.
2. Count user-triggered fetches per URL from your logs for ChatGPT-User, Claude-User and Perplexity-User.
3. Add cited URLs from any tracker or manual checks.
4. Add GA4 sessions by landing page, and AI Assistant channel sessions if you have them.
5. Add content signals from a crawl, such as word count and last-modified date. Crawlers such as Screaming Frog report word count per page.
6. Flag rows where citations, AI impressions or fetches are high but sessions are near zero and the page is short or old.

Sort the flagged rows by citation volume. The top twenty are usually where the effort pays.

What counts as a thin cited page?

Judge by usefulness, not word count alone. Warning signs:

• Little unique content: a paragraph and a list of links, or text copied from another page.
• Out of date: old prices, discontinued products, former staff or old addresses.
• Auto-generated: location, tag or filter pages that differ only in a name.
• Wrong format for the job: a PDF spec sheet cited for a buying question that a product page should answer.
• Not meant for buyers: press releases, job adverts or internal documents that were never retired.

Some short pages are fine. A clear, current page that answers one question precisely, such as an opening-hours page, can be short and correct. The test is whether a buyer who read only that page, or only the answer built from it, would come away with an accurate and complete picture. If not, the page needs work, however few visits it gets.

What should you do with each page?

Decide by what the page is cited for, which Bing's grounding queries and your tracker's prompts reveal:

• Cited for an important question, and correct: strengthen it. Add the detail buyers need, current facts, a clear answer near the top and links to the pages where people act, such as product or booking pages.
• Cited for an important question, but wrong or thin: fix it, or redirect it to a better page that answers the same question.
• Cited for questions you do not want: tighten the page so it matches its purpose, or noindex it if it should not be in search at all.
• Obsolete: retire it with a 404 or 410, or a redirect to the current page, and remove internal links and sitemap entries.

Note that noindex does not always stop AI use of a page, as discussed in whether a noindex page can still be cited by AI, and that crawlers find unlinked pages through sitemaps and external links, as covered in how AI crawlers find unlinked pages.

How do you know it worked?

Recheck the same sources a month later. For pages you strengthened, look for citations holding or rising and for more clicks from AI referrals or branded searches. For pages you retired or redirected, confirm that answers now cite the replacement. Keep the spreadsheet and rerun it each quarter, because new thin pages appear as sites grow. Wider signals of citation decline are covered in how to detect falling AI citation authority when traffic stays steady.

What might this look like in practice?

Consider an illustrative accounting software company, Ledgerbrook. GA4 showed its 2022 "tax deadlines" page with almost no visits, but Bing's AI Performance report listed it among the site's most-cited pages, for grounding queries about quarterly filing dates. Its logs showed hundreds of ChatGPT-User requests for it a month. The page still listed 2022 dates. Ledgerbrook rewrote it as a current deadlines guide with this year's dates, a short explanation of each deadline and a link to its filing reminders feature, and redirected three other old deadline posts to it. The next month, citations held steady, answers showed current dates, and the page began receiving referral visits from Copilot.

What do people ask about cited pages with no traffic?

Can a page be cited by AI without getting visits?

Yes. Many AI answers satisfy the question without a click, and some AI visits arrive without a referrer.

Which tool shows which of my pages AI cites?

Bing Webmaster Tools shows cited pages for Microsoft's AI; Search Console shows AI impressions by page for Google; logs show assistants fetching pages.

What is a thin page?

A page with little unique, current or useful content, such as a stub, an old announcement or an auto-generated listing.

Should I delete thin pages that AI cites?

Only if they are obsolete. If they are cited for important questions, improve them instead.