How Far Does Search Rank Affect AI Citations? A 135-Query Analysis

This article was first published in Korean by Intrix Lab on September 15, 2026.

Search rank played a role in AI source discovery, but fine-grained rankings alone did not distinguish which sources were cited after discovery. An analysis of 135 queries as of 2026-09-01 found that sources ranking 1–5 in search results for the original query entered the reference set at a rate of 29.7%, compared with 10.7% for ranks 21–30. In contrast, the conditional citation rate among referenced sources was 44.8% for ranks 1–10 and 40.5% for ranks 11–30, with substantially overlapping confidence intervals. Search rank was a gateway to discovery, but it was not a sufficient condition for citation selection.

How did we distinguish discovery from selection?

We defined discovery as the stage when AI adds a web source to its reference set, and selection as the stage when it links to one of those sources in its answer. The analysis covered 135 market queries across 6 industries and a single model, Model A. The AI answers were from 2026-08-31, and search results for the original queries were from 2026-08-15. We compared reference-set entry rates and citation rates after referencing across search ranking bands.

How did rank relate to discovery?

Reference-set entry rates declined as search rankings fell. The rates were 29.7% for ranks 1–5, 26.9% for ranks 6–10, 16.8% for ranks 11–20, and 10.7% for ranks 21–30. At least in this sample, higher rankings for the original query were associated with a greater likelihood of discovery by AI.

These figures do not mean that search rank directly produces AI citations. They show an observed difference in the opportunity to enter the reference set first. Actual citation must be examined separately at the selection stage.

Why was the distinction between fine-grained rankings less clear at selection?

Among referenced sources, the citation rate for ranks 1–10 in the original query's search results was 44.8%, with a confidence interval of 39.2–49.9. For ranks 11–30, it was 40.5%, with a confidence interval of 32.5–47.6. The two intervals overlapped substantially, so we could not identify a difference in selection attributable to fine-grained rankings after sources entered the reference set.

The conditional citation rate for sources outside the top 30 search results was 15.2%, with a confidence interval of 14.0–16.5. Being inside versus outside the top 30 was associated with a roughly 3-fold difference. However, this result is also an observational correlation. A third factor may contribute to both search relevance and citation relevance.

Why did 90% of references come from outside the ranking range?

Of 12,443 total reference occurrences, 11,162, or 90%, came from outside the top 30 search results for the original query. This is because AI performs fan-out searches, issuing additional rewritten queries instead of searching the original question just once. Looking only at rankings for the original query misses most of the actual paths used to collect candidates.

Being outside the ranking range does not mean a source was absent from all search results. It means it was outside the top 30 for the original query. It may have ranked highly for other search terms generated by AI, so rankings for the original query and sources found through rewritten queries should not be combined into a single value.

Which types of sources were selected?

After classifying all domains by type, citation rates after referencing were 23.9% for platforms (out of 1,024 occurrences), 19.9% for individual businesses (out of 5,611 occurrences), 18.6% for institutions, 9.0% for media, and 4.5% for directory sites. Businesses, institutions, and platforms had rates of 19–24%, compared with 4.5–9.0% for media and directories, a 2–5-fold gap. Directory sites readily entered the reference set but were discarded most often at the selection stage.

When only a subset had been classified, individual businesses appeared to have the highest rate (30.6%). Expanding classification to the full set brought that rate down to 19.9%. This article itself provides an example of how comparing types using a partial sample can reverse a conclusion. This lab's approach is to be open about its corrections.

Cited sources also tended to have titles that more closely matched AI-generated search terms than uncited sources did. Because this was not a prespecified hypothesis, we retained it only as a candidate for the next measurement.

What should performance reports separate?

Search rank, entry into the AI reference set, and answer citation should appear in separate columns. Reporting only high rankings for the original query cannot explain the 90% of references that came through fan-out searches. Conversely, reporting only final citations hides sources that were discovered in search results but dropped out at selection.

This separation allows discovery problems and selection problems to be addressed individually. You can first check your site's search and crawling conditions with Intrix's free SEO and GEO audit, then follow up with repeated AI citation measurements using a fixed question set.

How broadly can these results be applied?

The population consists of a market question set of 135 queries and a single model, Model A. We cannot generalize that the same proportions will occur with other models or industries. Domain types may also have been misclassified.

There is a 16-day gap between the search results and the AI reference data. If rankings changed during that interval, the overlap between the two sets may have been underestimated. Because this is an observational analysis without random assignment, it also cannot establish that changing rankings causes citations to increase.

Frequently asked questions (FAQ)

Q. Do higher search rankings lead to more AI citations?
At the discovery stage, reference-set entry rates declined as rankings fell. However, after sources were referenced, the confidence intervals for citation rates at ranks 1–10 and 11–30 overlapped substantially, so we could not identify a selection effect from fine-grained rankings.

Q. Will AI fail to discover a source outside the original query's top 30 search results?
No. Of all reference occurrences, 90% came from outside the original query's top 30. AI explores other search results through rewritten queries, so rankings for the original query alone cannot determine the candidate set.

Q. What should we check first in a GEO report?
Check search rank, reference-set entry, and final citation separately. This allows you to distinguish whether discovery or selection is the bottleneck.