Guide · Practitioner

AI Overviews vs AI Mode vs Classic Google Results: Same Sources?

AI Overviews, AI Mode and classic Google results all draw on Google's Search index, but they don't necessarily show the same sources. Google says both AI features may run multiple related searches ("query fan-out") and may use different models, so their links vary from each other and from the classic top ten. The only reliable way to know your overlap is to test your own queries.

Key takeaways

  • Same index, different selection: ranking well helps, but doesn't guarantee a citation.
  • Fan-out means pages ranking for related sub-questions can be cited.
  • Measure overlap on your own query set; don't assume someone else's study applies.

What are the three surfaces?

SurfaceWhat the user seesWhen it appears
Classic resultsRanked links, snippets and SERP featuresEvery search
AI OverviewA generated summary at the top with supporting linksOnly when Google's systems judge it adds value; often doesn't trigger
AI ModeA separate, conversational AI search with longer answers and linksWhen the user chooses AI Mode

Google's site-owner guide explains that AI Overviews "are only shown when our systems determine that it is additive to classic Search", and that AI Mode suits queries needing "further exploration, reasoning, or complex comparisons" (Google Search Central).

What does Google say about how sources are chosen?

From the same documentation (Google Search Central):

  • Both features may use query fan-out, "issuing multiple related searches across subtopics and data sources".
  • While responses are generated, Google's models "identify more supporting web pages", which lets them show "a wider and more diverse set of helpful links" than a classic search.
  • AI Mode and AI Overviews may use different models and techniques, so responses and links vary.
  • To be eligible as a supporting link, a page must be indexed and eligible to be shown with a snippet. There are no extra technical requirements.

Put plainly: the classic ranking for your exact query is one input. Pages that answer the sub-questions can be pulled in too.

How do you test overlap on your own queries?

You need about two hours and a spreadsheet.

  1. Pick 20 queries where you care about visibility. Mix definitional ("what is X"), comparison ("X vs Y") and decision queries ("best X for Y").
  2. Classic: for each, record the top 10 organic URLs. Use a clean browser profile and the same location each time.
  3. AI Overview: note whether one triggered. If yes, record every linked URL.
  4. AI Mode: run the same query and record every linked URL.
  5. Normalise to domains and URLs. Compare at both levels; domain overlap is usually higher than exact-URL overlap.
  6. Calculate overlap per query:

- AIO ∩ top 10 / AIO links - AI Mode ∩ top 10 / AI Mode links - AIO ∩ AI Mode / AIO links

  1. Repeat in two weeks for the same queries to see how stable the sets are.
QueryAIO shown?AIO linksIn top 10AIM linksIn top 10AIO∩AIM
example: best crm for agenciesY631142

The row above is an illustration of the format, not real data.

How should you read the results?

PatternWhat it suggestsWhat to do
High overlap with top 10AI features lean on the main ranking for these queriesClassic SEO work pays off directly; protect rankings
Low overlap, cited pages answer sub-questionsFan-out is pulling in specialist pagesCreate focused pages for the sub-questions your buyers ask next
You rank but aren't citedYour page may be hard to quote or snippet-limitedCheck for nosnippet/max-snippet; rewrite answer-first
You're cited in AI Mode but not AIO (or vice versa)Different models and selectionTrack both separately; don't merge into one metric

Does ranking still matter?

Yes. Eligibility for AI features starts with being indexed and snippet-eligible, and the AI features sit on top of the same index. What changes is that you can win citations for queries where you aren't in the classic top three, if your page answers a sub-question better than anyone else.

How do you measure this in Search Console?

Google includes AI Overview and AI Mode appearances in the overall search traffic in the Performance report, under the Web search type (Google). There's no filter for "AI Overview clicks". Useful proxies:

  • Watch impressions and CTR for queries where you know an AI Overview triggers (from your test sheet).
  • Compare page-level clicks before and after you were first cited.
  • Pair with GA4 engagement metrics on those landing pages.

What should you change on your pages?

  1. Answer the main question in the first 40-60 words.
  2. Cover the next three questions with their own H2s so fan-out searches can land on you.
  3. Use tables for comparisons and specs.
  4. Remove unnecessary snippet limits on pages you want quoted.
  5. Keep facts current and dated. Out-of-date pricing is a common reason a competitor's page is preferred.

How often should you re-test?

Monthly is enough for most teams, plus an extra run after any big Google update or a major change to your own pages. Keep the query list fixed for at least a quarter so trends are comparable. If you add queries, report them as a new cohort rather than mixing them into the old average.

Common mistakes

  • Concluding "AI Overviews only cite the top 3" from a handful of searches.
  • Mixing locations or logged-in states between tests.
  • Counting "no AI Overview shown" as "not cited". Record it separately.
  • Ignoring AI Mode because it's a separate tab; it's where complex comparison queries go.

Sources

Questions

Do AI Overviews cite the top-ranking pages?

Often there is overlap, but not always. Google says AI features may run several related searches (query fan-out), so cited pages can include ones that don't rank in the top ten for the original query.

Are AI Overview clicks counted in Search Console?

Yes. Google reports traffic from AI Overviews and AI Mode within the overall Web search type in the Performance report, not as a separate line.

Why does AI Mode show different links from AI Overviews?

Google says AI Mode and AI Overviews may use different models and techniques, so the responses and links they show will vary.