E-E-A-T beyond YMYL: what really matters in 2026

I am Louis Choquet, founder of Pharelia. What we have observed since Google's official AI guide (May 2026) and the arrival of AI Overviews in France: teams overinvest in pointless tactics (llms.txt, chunking) and underinvest in genuine signals of experience. This article draws the distinction, backed by figures.

No. Google evaluates E-E-A-T across all queries, with varying levels of scrutiny. For a YMYL topic (health, money), the bar is high and an error can lead to a downgrade. Outside YMYL, the same framework applies in a lighter form: experience and reliability remain signals, even for a DIY tutorial or a software comparison.

The Creating helpful content documentation mentions no exemptions by topic. The first E — Experience — was added specifically to cover topics that are not sensitive: a product review carries more weight if it comes from someone who actually tested the product. In an audit, a 'how to clean a coffee maker' article written by someone who operates one beats a generic article optimized for a keyword, because lived experience comes through.

The practical decision: do not use non-YMYL status as an excuse to neglect author signals. The difference is gradual, not binary.

Four concrete signals: firsthand experience visible in the text, a verifiable author identity, originality (information not found elsewhere) and the site's consistency within its subject area. These are what matter outside YMYL, rather than the medical certifications or financial legal disclosures specific to YMYL.

The most underestimated lever is information gain: contributing a data point, a test or a figure missing from other pages. Google's Quality Rater Guidelines (Sept. 2025 version, ~182 pages, ~16 000 raters) assign the lowest rating to 'low-effort' content with little originality. For our clients, adding a proprietary data table or a screenshot of a real test changes rankings faster than a cosmetic rewrite.

Yes, and Google says so explicitly. The official AI guide (15 May 2026) sums it up: 'optimizing for generative AI means optimizing for search — so it is SEO'. Search Engine Journal reinforces the point: AEO and GEO remain SEO. They are not a separate discipline.

The guide's mythbusting section is explicit. Google Search DOES NOT USE llms.txt: it neither helps rankings nor incurs a penalty. There is no requirement to split content into chunks. There is no need to write specifically 'for AI'. Structured data is NOT required for generative AI — it remains useful for traditional rich results. One nuance: llms.txt may serve other crawlers, but there is no evidence of an effect on Google. I have seen teams spend a week on an llms.txt file that changed nothing on Google's side.

To distinguish the approaches clearly, see our AEO vs SEO vs GEO comparison. The operational conclusion: invest in quality and E-E-A-T, not in technical artifacts that Google ignores.

Only in part. AI surfaces are fragmented: ranking well on Google does not guarantee a citation from ChatGPT or Perplexity. ChatGPT citations overlap with only ~12 % of the Google SERP, and around 80 % of LLM citations do not appear in Google's top 100. Earning those citations requires broader distribution, not just a good ranking.

The share of AI Overviews citations coming from Google's top 10 has collapsed: ~76 % in mid-2025, ~38 % in early 2026, according to Ahrefs data on the decline in clicks and Higoodie. In other words, AI is increasingly drawing from outside the top 10. Reddit, disproportionately cited by LLMs because its full text can be retrieved, is becoming a GEO distribution channel in its own right.

Google announced the launch of AI Overviews and AI Mode in France on July 22, 2026. AI Overviews appears when Google considers a generated answer useful for the query; it does not replace all conventional results. Track changes in impressions and clicks on your pages; an AI presence does not prove a CTR decline or a commercial gain.

Search Console provides a generative AI performance report for AI Overviews and AI Mode impressions. Google reports a global rollout as of August 31, 2026, with access limitations depending on the property and impression volume. This data is also included in the Web report: do not add it to that report's total. Compare identical periods, countries, devices and aggregations. The chart is aggregated by property, or by URL when a URL filter is applied; the page table may therefore have a different total. Google's counting rules do not establish automatic double-counting of AI and conventional impressions. A CTR decline calls for an analysis of traffic and queries, not a blanket adjustment. In GA4, check the AI Assistant channel and source/medium data for identifiable sessions. Traffic with no identifiable source may appear as Direct — a custom channel group does not reconstruct a lost source. Do not therefore reattribute all Direct traffic to AI.

Check the available reports now and keep a dated baseline measurement. If earlier data exists, compare identical scopes; otherwise, start monitoring without artificially reconstructing the past. Our guide to AI Overviews in France explains the steps to follow.

Not if your content provides real value. The May 2026 core update finished on 2 June (~12 days), and the June 2026 spam update launched on 24 June, worldwide and across all languages. They target low-value content produced at scale, not AI itself.

Google's position (Danny Sullivan): 'we judge quality, not the production method'. There is no blanket penalty against AI. The Quality Rater Guidelines list three spam categories — Expired Domain Abuse, Site Reputation Abuse, Scaled Content Abuse — plus a 'low-effort' section. Content generated at scale without supervision receives the lowest rating. An AI-written article that is reviewed, enriched with your experience and fact-checked passes; a content factory falls.

The decision: keep people in the loop and document genuine experience. To audit your pages systematically, use our SEO audit checklist.