I Reviewed Tested AEO and GEO Experiments and Case Studies

I have been doing SEO since 2019. Now AEO and GEO are trending. So I thought I’d look at case studies and experiments. AEO and GEO advice is full of confident claims and thin evidence. Three of the most repeated claims are that you need an llms.txt file, that schema markup gets you cited by AI, and that being cited sends you traffic. Several large studies have now tested AEO and GEO, so here is what they found as of October 2026.

  • No strong evidence that major AI systems read llms.txt when answering questions.
  • The largest controlled test found no citation lift from adding Schema markup to pages already visible in AI answers. It is still worth having for rich results.
  • Cited brands get more clicks than uncited ones, but the cause is unproven, and overall clicks on AI Overview results fell.

Note: Original AEO case studies and Experiments are linked at the bottom of this article.

Myth 1: “You need an llms.txt file”

Ahrefs analyzed 137,000 domains. 28% publish an llms.txt file, but almost all of them (97%) received zero requests. Of the requests that did arrive, 96% came from bots, and 77% of those bots were not AI tools (think scanners and catalogers). Named AI bots made up 19.5% of the requests to the small group of files that got read.

Other trackers point the same way:

  • Originality.ai’s tracking of over 3 million sites found that adoption has grown sharply while evidence that AI systems read the files stays close to zero. The same coverage, citing Ahrefs, says AI retrieval bots accounted for only 1.1% of requests.
  • Senthor, a bot-detection vendor, reported that none of 10 million AI requests it analyzed attempted to read the file.
  • Google’s John Mueller has publicly said no AI system currently uses llms.txt.
  • A WordPress plugin’s opt-in dataset found that OpenAI’s bots engage most, while Common Crawl’s bot never fetched the file in 30 days.

The counterargument: iPullRank’s Mike King has argued that systems like Claude, Perplexity, and vertical agents have different retrieval architectures and may process the file, and there is no consolidated guidance on which ones do.

Limitations of this case study: Ahrefs did not check whether files were well-formed and studied only the index file. Fetch logs also can’t show whether the content influences answers some other way.

My Opinion: It’s harmless and cheap, but don’t expect traffic or citations from it. It belongs at the bottom of your priority list.

Myth 2: “Schema markup gets you cited by AI”

The myth started with a real correlation. Ahrefs found that schema is much more common on AI-cited pages than on uncited ones across 6 million URLs. But big, well-maintained sites tend to have both schema and citations for unrelated reasons.

So Ahrefs ran a controlled test. It tracked 1,885 pages that added JSON-LD between August 2025 and March 2026 and compared them with 4,000 control pages. The results:

PlatformChange in citations
Google AI Mode+2.4%
ChatGPT+2.2%
Google AI Overviews−4.6%

The first two are statistically indistinguishable from zero. The AI Overviews dip was statistically significant, but the authors warn against reading it as harm caused by schema. In a separate test, five AI systems (ChatGPT, Claude, Perplexity, Gemini, and AI Mode) were checked for whether they use schema when fetching pages live. Results indicated they extract visible HTML content rather than JSON-LD.

Limitations of the case study:

  • Every page in the test already had 100+ AI Overview citations before schema was added. The test therefore can’t say whether schema helps a page get into the consideration set in the first place.
  • The study pooled all schema types together.
  • Other studies disagree: OtterlyAI reported a 1,500% jump in AI Overviews citations but declines on ChatGPT, and an AirOps study found a 6.5 percentage point advantage for JSON-LD pages on ChatGPT. Those are observational comparisons, not controlled before-and-after tests.

Google’s own position is that no special markup is needed for its AI features.

My opinion: Keep schema for rich results (Product, Review, Article). Don’t add it as an AEO tactic and expect a citation lift.

Myth 3: “Getting cited means getting clicks”

This is the strongest of the three claims, with a big caveat. Seer Interactive’s study covered 53 brands, 5.47 million queries and 2.43 billion organic impressions from January 2025 through February 2026. It found that being cited in an AI Overview delivers 120% more organic clicks per impression than not being cited. In concrete terms, that’s roughly 20,743 clicks per million informational impressions for cited brands versus 9,445 for uncited ones.

But Seer itself cautions that it can’t prove citation causes the higher CTR, because brands with stronger authority may simply be more likely to be cited. And the bigger picture is less comfortable:

  • In Seer’s earlier work, organic CTR fell from 1.41% to 0.64% on queries with AI Overviews.
  • Pew Research found that click rates fell from 15% to 8% on queries returning an AI Overview.
  • SparkToro measured that 68% of US Google searches ended without a click between January and April 2026.
  • There is a bright spot: informational queries without an AI Overview saw CTR climb from 2.9% to 4% by the end of 2025.

One more useful finding: through mid-2025, roughly 75% of sources cited in AI Overviews also ranked in Google’s top 10 for the same query. Classic rankings still feed citations.

My opinion: Being cited is worth pursuing, but you’re competing for a share of a smaller pie, and the benefit is correlation, not proven cause. I personally checked this with electronicsforu, and mostly informational queries are seeing the drop in organic clicks as they are getting shown in AI overviews. But, users who are coming from AI overviews or LLM platforms like chatGPT are spending more time on site and getting converted into leads.

What the evidence supports instead

  1. Rank well in organic search. The overlap with citations is large.
  2. Put clear answers in visible text. The retrieval tests suggest AI systems read what’s on the page, not hidden markup.
  3. Publish original data. A Princeton/Georgia Tech study reportedly found that adding statistics improved citation visibility by 41% on Perplexity.
  4. Earn third-party mentions. A Stacker/Scrunch controlled study reported a 239% median lift in AI search visibility from distributing the same content through third-party news outlets.

My Analysis and Opinion

TacticBest evidenceWhat to Do?
llms.txt137K-site study, 97% of files never requestedLow priority
Schema for AI citations1,885-page controlled test, no liftKeep for rich results only
Chasing AI citationsSeer: +120% clicks per impression, causation unprovenWorth it, with realistic expectations

Sources:

4 thoughts on “I Reviewed Tested AEO and GEO Experiments and Case Studies”

  1. Pingback: ChatGPT UTM Tracking Issue: Latest Update

  2. Pingback: Google Updates AI Search: More Inline Links and New 'Dive Deeper' Features Rolled Out

  3. Pingback: AEO Quiz: 25 Hard Questions on Answer Engine Optimization

  4. Pingback: GEO Quiz: Can You Get Your Content Cited by AI?

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top