TL;DR
- Google rankings are useful context, but they do not show which pages an AI answer actually cites.
- Same-page overlap and whole-domain correlation use different units. Strong findings at one level cannot become a citation probability at the other.
- A top-10 position is not the same as Google index eligibility. Evidence from AI Overviews should not be treated as a rule for every assistant.
- Keep rank, cited URL, answer context and scope in separate fields. Use their differences to choose a useful next check.
Keep the ranking record beside the saved answer
This distinction matters when a buyer asks why a well-ranked product page is absent from an AI comparison. It also matters when a cited guide ranks poorly for the original query. Neither case, on its own, shows that SEO has stopped working or that AI visibility is fully explained by one rank.
Read Same-Page Overlap With the Right Denominator
Ahrefs' March 2026 analysis compared cited URLs with Google results for the same query. Its standard blue-link test found 37.1% of cited URLs in the top 10, 26.2% at positions 11–100, and 36.7% without a top-100 match. The study analyzed 863,000 keyword result pages and four million AI Overview URL records overall.
| Same-query standard organic position | Share of AI Overview cited URLs |
|---|---|
| Top 10 | 37.1% |
| Positions 11–100 | 26.2% |
| No top-100 match | 36.7% |
Source: Ahrefs, March 2, 2026, standard blue-link test. Ads and other search-result features were excluded from this version. Shares describe cited URLs in that dataset, not the probability that a top-10 page will be cited. No top-100 match does not mean a URL was unindexed.
The direction of the count is important. The analysis starts with cited URLs and asks where they rank. To estimate how often top-10 pages get cited, you would need a base of top-10 pages and their citation outcomes under a defined protocol. Reversing the conditional gives a claim the study did not measure.
The source also reports a separate test that counts result blocks, including features. Its headline rounds that result to 38%. The chart here uses blue links only. Mixing the two tests would make a tidy-looking report with mismatched fields. Nor should a comparison with the older study be called a clean trend: Ahrefs says its citation parsing improved between studies.
Ahrefs, March 2026. Standard organic blue links only, excluding ads and other features. Shares start with cited URLs; they are not a top-10 page’s citation probability or a measure of index status. Original study and tests.
A Domain Correlation Can Be Strong While Page Overlap Is Limited
Semrush's domain-level analysis selected the top 10,000 domains in its AI Visibility Index. It reported a 0.86–0.87 correlation between organic keyword footprint and AI citation count across the platforms analyzed. Keyword footprint means the number of keywords a domain ranks for. This is an association across selected domains, not a test of one page's rank.
These findings can coexist. A large domain can rank across many topics and earn many citations, while an answer cites a different URL from the page ranking for the original query. The domain finding does not tell us that moving one page into the top 10 will cause a citation. The overlap finding does not tell us that a site's wider search presence has no value.
| Evidence | Actual result and unit | Inference to avoid |
|---|---|---|
| Ahrefs, same-query blue-link test | 37.1% of cited URLs matched the top 10; 26.2% matched 11–100; 36.7% had no top-100 match | “A top-10 page has a 37.1% citation chance” |
| Semrush, selected top 10,000 AI-visible domains | Organic keyword count and citation count correlated at 0.86–0.87 across the analyzed platforms | “Raising one URL's rank, or adding keywords across the domain, will cause citations to rise” |
| Method contrast | URL/query overlap versus a whole-domain association | Treating the studies as rival measurements of one effect |
Source: Ahrefs study and Semrush analysis, checked November 8, 2026. The Semrush page does not clearly give collection dates or a full query protocol. Both firms sell search tools; neither study is a controlled B2B ranking intervention.
Selection matters in the second study. Domains already prominent in an AI index are not a random sample of all B2B suppliers. The coefficient is not a success rate and is not a share of buyers reached. Keep it as evidence of an association in that selected group. Nor does it show that expanding a domain's keyword footprint will raise its citations. Site size and brand prominence could influence both counts.
Eligibility, Rank and Citation Are Separate Checks
Google's current AI features guide says a page must be indexed and eligible to appear with a snippet to be a supporting link in AI Overviews or AI Mode. It also says that meeting requirements does not guarantee serving. Index eligibility is a different question from a top-10 position for a particular query.
That distinction prevents a common overreach. A source outside the tracked top 100 may still meet Google's index requirements. A page in the top 10 still has no promise that an answer will cite it. This is Google guidance, not a rule for ChatGPT, Perplexity or every other product.
The tracking record should preserve the URL, question, market, language, product surface and time. Keep the full answer so a reviewer can see what the citation supports. A link to the domain's guide is not a link to its product page. A source link is also not proof that the brand was recommended.
Our citation diagnostic guide covers that source-support check. Here the task is narrower: keep the ranking observation and the cited page distinct so the report does not claim more overlap than the record shows.
| Field | Fictional B2B software observation | What the reviewer can say |
|---|---|---|
| Original question | Compare access-control options for a small document-search team | This is the chosen buyer task, not market-wide demand |
| Scope | Fictional: a Google AI Overview saved for that query, US English, same day as the rank check | Google's eligibility guidance applies to this surface; another assistant needs its own record |
| Rank record | The vendor's product page is in the tracked Google top 10 for that query, at the recorded time | A rank was observed under those conditions |
| Saved AI answer | Links the same vendor's access-control guide, but not the product page | The domain was cited; the two URL observations do not match |
| Answer context | Uses the guide to explain a permission model | A source role was observed; recommendation needs its own label |
| Next check | Review whether the guide supports the claim and gives a useful route to the product facts | No need to claim that the product page was cited or to abandon its search value |
This is a fictional record, not a live test or a SemShift client result. The draft report says “Our top-10 product page was cited by AI.” The saved answer actually links another page on the same domain. The evidence supports a domain citation and a separate product-page ranking, not the draft's same-page claim.
The corrected note names both URLs: “The product page ranked in the tracked top 10. The saved AI answer cited the access-control guide.” The reviewer retains the time and conditions for each record, checks the cited passage and labels any brand recommendation separately. The match check passes when the report preserves both facts without combining them.
A useful response is to inspect the guide's accuracy and its handoff to the product detail. Do not assume it needs to be replaced by a sales page merely because that page ranks well. The buyer's question and the source role should guide the content decision. A ranking report alone cannot make that decision.
Rankings and AI Citations: Common Questions
Does no top-100 match mean a page is not indexed?
No. It means no match was found within that tracked result range and test. It does not establish index status, permanent rank or behavior for related queries. Use the relevant system's evidence for those separate questions.
Can I use 37.1% as my expected citation rate?
No. That share starts with cited URLs, not all top-10 pages. It is a result for one study's AI Overview dataset. Your own citation rate needs a defined answer sample, a numerator and a denominator that match the question.
Should we stop SEO work if a lower-ranked page is cited?
That observation does not settle the value of search traffic or the site's broader coverage. Review the buyer task and the page's role. Track search performance and AI answers separately, then judge what each adds.
Can we pool all assistants into one overlap number?
Only if the report clearly defines the products, samples and weighting, and explains what that pooled figure answers. It should not hide platform results or be presented as a universal rank rule. Our measurement guide covers those broader sampling choices.
Match ranking and citation claims to their records
If an AI visibility report uses Google positions as a substitute for saved citation evidence, SemShift's AI visibility audit can help review the mapping. Bring the rank export, full answers, cited URLs and buyer questions so each claim can be checked at the right level.