Google AI Mode Ranking Factors: What the Evidence Supports

Google's AI features page gives one eligibility rule for AI Mode and AI Overviews. A page "must be indexed and eligible to be shown in Google Search with a snippet." The same page says "There are no additional technical requirements" and no special optimizations are needed. It also warns that indexing and serving isn't guaranteed.

Google adds two details that matter for AI Mode specifically:

• AI Overviews and AI Mode "may use different models and techniques", so the two surfaces can return different links for the same query.
• AI Mode may use a "query fan-out" technique, issuing multiple related searches across subtopics. The searches it actually runs are not shown to you.

As of October 2026, Google's Search I/O post names gemini-3.5-flash as AI Mode's default model since May 2026. Models change, so treat that as a dated fact.

What Google does not publish: a list of selection factors, their weights, or any position data for AI Mode.

Less AI Mode data than AI Overviews data

Most independent studies measure AI Overviews, not AI Mode. Ahrefs' top-10 overlap figure is an AI Overviews number. Ahrefs' schema study is one of the few that reports AI Mode separately. Where this page grades a factor, check whether the evidence came from AI Mode or was borrowed from elsewhere. For the AI Overviews picture, see what evidence supports for AI Overviews.

The pattern matches every other surface: access evidence is strong, content tactics are not. For how AI Mode turns this into links, read how Google's AI Mode picks links.

These factors work by removing blocks. None makes a page more likely to be picked. Each one can make a page ineligible.

• Googlebot rules in robots.txt. There is no robots.txt token that blocks AI Overviews or AI Mode. Googlebot is the control for all Google Search surfaces.
• Google-Extended is not the AI Mode switch. It governs Gemini Apps training and grounding. Google says it does not affect inclusion in Search.
• On-page snippet controls. noindex, nosnippet, max-snippet:0 and data-nosnippet keep content out of AI answers. A template that hides a cookie banner with data-nosnippet can hide more than intended.
• The Search Console generative AI control. Inclusion is opt-out: it "is the default control for all properties." It covers AI Overviews, AI Mode and Discover generative AI features. Child properties default to "Inherit control from parent", so a parent setting can silently exclude a subdomain. As of the last check, the toggle was still rolling out to a subset of owners.

Otterly's 2,000 AI blogs experiment shows the eligibility rule in action. Google deindexed both test sites, and Google's AI surfaces, AI Mode included, almost never cited them. Bing kept the sites, and Copilot kept citing them.

Rendering is less of a risk here than on ChatGPT or Claude, because Googlebot renders JavaScript. Googlebot fetches up to 2MB per URL, but Google says most pages never hit that limit.

Next move: confirm Googlebot is allowed, scan templates for snippet directives, check index status in Search Console, and screenshot the generative AI control on every property.

A 2026 critical survey of 45 GEO studies (arXiv 2607.14035) found that topical relevance and context position are the most reproducible levers. A 252,000-trial factorial study reached the same conclusion. Both measured documents already placed inside a model's context. Neither queried AI Mode or a live index.

So the honest grade is moderate, and it transfers to AI Mode by assumption, not measurement.

Topical relevance is not keyword coverage. Google says you don't need to worry about long-tail keywords or every variation of a query. The original GEO paper found keyword stuffing lowered its own visibility metric (17.8 against a 19.5 baseline). Fan-out makes lexical matching even less useful, because you can't see which queries AI Mode runs.

Organic rank has no AI Mode number at all. Ahrefs found 37.9% of AI Overviews citations came from the first 10 results (March 2026), but that measured AI Overviews, which Google says may use different models. Treat any rank-to-AI-Mode claim with the caution in correlation vs causation in AI search studies.

Next move: make sure each page plainly answers the questions it targets, and states specific facts (prices, numbers, definitions) in fetchable text.

Schema

Ahrefs' matched-control study of 1,885 pages found "no major uplift in citations on any platform" after adding schema: AI Mode +2.4%, and AI Overviews a statistically significant 4.6% drop. Every treated page already had 100+ AI Overviews citations, so whether schema helps a page enter the candidate set is untested. Google says existing markup does no harm. It just isn't a lever. FAQ rich results were deprecated in May 2026. More in what evidence shows on schema markup.

llms.txt

Google's AI optimization guide says Google Search ignores AI text files, and that creating one "will neither harm nor help" visibility in Search, including its generative AI features.

AI-friendly writing and chunking

The same guide says you don't need to write in a specific way for generative AI search, and that there's no ideal page length.

E-E-A-T scores

Google says E-E-A-T "isn't a specific ranking factor." There is no score to improve. See Google's helpful content guidance and E-E-A-T.

If a vendor sells any of these as an AI Mode fix, check it against GEO claims that don't hold up.

Google's generative AI report in Search Console shows impressions for AI Overviews and AI Mode. As of October 2026 it shows no clicks, no CTR and no position. A Google-side bug depressed impressions from about 12 to 13 August 2026, so check current status before reading a trend across that window.

That means any "AI Mode rank" has no ground truth behind it. Prompt-based checks are samples of what AI Mode showed for chosen queries, on one day, in one location.

What stays in your control:

1. Keep pages indexed and snippet-eligible.
2. Remove accidental noindex, nosnippet, max-snippet and data-nosnippet directives.
3. Check the Search Console generative AI control, including inherited settings.
4. State key facts plainly in the HTML.
5. Spot-check real buyer questions, for example with the Google AI Mode Checker, and treat results as a snapshot.

What you can't control: which model Google runs, which fan-out queries it issues, and which eligible page it picks. For the sibling surface, compare Gemini visibility factors.