What 2 Million AI-Cited Pages Teach About Which AEO Tactics Actually Work

The GEO industry sells a checklist: add FAQ schema, publish llms.txt, chunk content for extractability, cite .gov domains. Most of that advice was written before anyone had enough citation data to test whether it works.

SE Ranking's September 7, 2026 AEO guide synthesizes 2 million pages and 400,000 domains across ChatGPT and Google's AI answers. Some widely recommended tactics hold up. Others do not. And there is no single formula that works equally across every answer engine.

This article distills what the dataset says — and what it means for how you allocate AEO budget in late 2026.

Why this dataset matters now

Prior AEO advice was largely theoretical: plausible mechanisms, anecdotal wins, vendor case studies. SE Ranking's Visible product now tracks citations at scale across ChatGPT, AI Mode, AI Overviews, Gemini, and Perplexity — and their September guide is the first public synthesis large enough to test the checklist items against real citation outcomes.

The headline: authority and existing visibility dominate. Cosmetic GEO tactics — llms.txt, FAQ schema markup, linking to .gov domains — show negligible or negative returns in their models.

What works for ChatGPT citations

SE Ranking analyzed 129,000 domains, 216,524 pages, and responses to 100,000 prompts.

This aligns with Indexably's finding that domain factors explain 77% of predictive importance and with Victorious' Q2 2026 correlation between referring domains and AI mention rates.

What did NOT work for ChatGPT

ChatGPT citation visibility does not hinge on one technical AEO trick. Authority, existing visibility, useful depth, freshness, and external validation matter more.

What works for Google AI Mode

SE Ranking analyzed 2.3 million pages from nearly 300,000 domains across 20 niches, using 500,000+ prompts.

The pattern is straightforward: authority and existing Google visibility create the strongest advantage. Content depth, freshness, and structure can improve odds further — but as compounding factors, not substitutes for footprint.

AI Mode self-citations are surging

One September datapoint worth isolating: SE Ranking's research into AI Mode citations found google.com itself accounted for 17.42% of all citations in February 2026, up from 5.7% in June 2025 — more than tripling in under a year.

Google is increasingly citing its own properties inside AI Mode answers. That shifts the competitive frame: your content competes not only with other publishers but with Google's own surfaces for citation slots. See Google AI Mode's self-citation shift for the broader pattern.

What works for AI Overviews

AI Overviews sit closest to Google's traditional search ecosystem:

Google's own guidance confirms: no separate technical checklist for AI Overviews. Pages need to be indexed, eligible for snippets, crawlable, and competitive in organic search first. Then structure content for extractability.

This is consistent with Google's May 2026 statement that AEO and GEO are still SEO — and with SE Ranking's explicit rejection of "AI-specific markup" or special AI files as requirements.

Platform divergence: no universal formula

SE Ranking's central warning: what works for ChatGPT may matter less in AI Mode, AI Overviews, Gemini, Claude, or Perplexity.

Only 2.7% of domains are cited by all five engines. Optimize per platform or accept that a win on one surface may not transfer.

The llms.txt verdict — again, with numbers

SE Ranking tested llms.txt across nearly 300,000 domains. Only 10.13% used the file. They found no correlation with AI citation frequency.

This adds quantitative weight to Google's explicit rejection of llms.txt and to Ahrefs' finding that 97% of llms.txt files were never read by an AI crawler. Publishing llms.txt is low-cost insurance for developer docs consumed by coding agents — not a citation lever for content sites.

FAQ schema: content yes, markup no

The distinction matters:

• FAQ content embedded in pages — modest positive association in AI Mode (4.9 vs 4.4 citations)
• FAQ schema markup — no lift in ChatGPT; no meaningful impact in AI Mode

Write FAQ sections because buyers need answers, not because schema tags unlock AI citations. Structured data still helps for product, organization, and consistency — but it is not the citation differentiator GEO vendors claim.

Practical playbook: sequenced by what the data supports

Tier 1 — Authority and footprint (all platforms)

1. Grow referring domain diversity — not volume, but independent corners of the web
2. Maintain review platform presence (G2, Capterra, Trustpilot, Yelp as relevant)
3. Build Reddit/Quora/community mentions in categories where those surfaces get cited
4. Keep pages fresh — 3-month update cycles correlate with citation advantage

Tier 2 — Content structure (compounds with Tier 1)

1. Answer-first sections with 120–180 words between headings
2. Depth where warranted — longer pages correlate with more citations, but clarity beats word count
3. In-page FAQ content where buyers ask real questions
4. Technical fundamentals — crawlability, canonical tags, meta descriptions (see Indexably)

Tier 3 — Deprioritize unless you have evidence

1. llms.txt for citation purposes
2. FAQ schema as an AEO tactic
3. Linking to .gov/.edu for citation juice
4. "Chunking" content into artificial AI-friendly blocks (Google explicitly rejects this)

Tier 4 — Platform-specific tuning

1. ChatGPT: earned media on cited domains; review presence; authority
2. AI Mode / AI Overviews: organic competitiveness first; watch self-citation competition
3. Perplexity: freshness and primary sources
4. Measure each separately — do not assume transfer

How Obsurfable fits

Checklists tell you what to change. Obsurfable tells you whether anything moved.

Define buyer Prompts, run repeated observations, and track citation share and mention rate over time. When you skip llms.txt and invest in review platform presence instead, Obsurfable shows whether your category mention rate actually shifts — on ChatGPT, AI Overviews, and Perplexity separately.

The Visibility Director maps prompt-level gaps to the tactics this data supports: authority building when retrieval fails, extractable structure when you are in the pool but not cited, earned media when citations exist but mentions do not.

FAQ

Does this mean I should stop all on-page GEO work?

No. On-page structure compounds once you have authority — Indexably showed page signals only lift the top quartile. Fix fundamentals everywhere; expect measurable citation lift primarily when footprint is already strong.

Is AEO just SEO renamed?

For Google surfaces, largely yes — SE Ranking and Google agree. For ChatGPT and Perplexity, additional factors (community presence, freshness weighting, retrieval index differences) create platform-specific layers on top of SEO fundamentals.

Why did FAQ content help but FAQ schema did not?

Models extract answers from visible page content, not from schema tags they may not parse for citation decisions. Write FAQs for users; do not expect markup alone to move the needle.

How often should I refresh the measurement?

Citation sets are volatile — rank-style tracking fails. Run prompt observations weekly or biweekly, not monthly snapshots.

Bottom line

Across 2 million AI-cited pages, authority, organic visibility, and third-party validation dominate. llms.txt and FAQ schema show negligible returns. FAQ content, freshness, and clear structure compound with authority. AI Mode self-citations are tripling. No single checklist works across all engines. Measure citations and mentions per platform — then sequence work to the bottleneck the data actually shows.