TRAINYOURAGENT

SEO vs AEO vs GEO: what changed, and what the evidence actually shows

The mechanics changed and the fundamentals did not. Zero-click reached 68.01% of US Google searches in early 2026, and AI Overviews cut organic CTR by 34.5% to 61% depending on whose study you read. The single most useful finding: brand mentions correlate 0.664 with AI Overview visibility against 0.218 for backlinks.

The options, and the losing case for each

The evidence, with sample sizes and dates. Every row is traceable to a named study; none of these are our own figures.

Footnotes on that table

What actually changed, in numbers?

The share of searches that produce a click collapsed. SparkToro's June 2026 analysis of Similarweb panel data found that 68.01% of US Google searches in the first four months of 2026 ended without a click, against 60.45% in 2024 and roughly 49% in 2019. That is the single most consequential number in the field, and it is not about AI specifically — it is about Google answering more queries itself. The value of ranking first fell on the queries that get an AI Overview. Ahrefs studied 300,000 keywords and found position-one clickthrough dropped 34.5%, from 0.073 to 0.026, comparing March 2024 to March 2025. Seer Interactive, tracking 3,119 terms across 42 organisations and 25.1 million impressions, found a 61% decline in organic CTR on AI Overview queries between June 2024 and September 2025, from 1.76% to 0.61%. Those studies disagree and both are credible; the useful conclusion is that the decline is large, not that it is precisely one number. The visitors that remain are worth more. Semrush's June 2025 research reported that AI search visitors convert at 4.4 times the rate of traditional organic search visitors. Fewer, better-qualified visitors is a different business than more, worse-qualified ones, and it changes what you should optimise for. What did not change is that the mechanism is still retrieval from an index. Semrush's same research found that roughly 90% of the time, ChatGPT's search references pages ranking at position 21 or lower in classic search — which means classic indexability remains the precondition, even as ranking position stops being the payoff.

Is AEO different from SEO, or is it just SEO with better headings?

Mostly the latter, and that is not a criticism — it is the reason it works and the reason it is cheap. Answer engine optimisation is a set of formatting disciplines: every h2 is a question, every answer beneath it is self-contained and does not depend on the paragraph above, tables contain real values rather than checkmarks, units and dates are explicit, and the recommendation appears in the first forty to sixty words rather than after five paragraphs of context. The evidence that format matters is real. Wix Studio's AI Search Lab analysed 75,000 AI answers and over a million citations across ChatGPT, Google AI Mode and Perplexity in March 2026, finding listicles at 21.9% of citations, articles at 16.7% and product pages at 13.7% — 52% between them. It also found that query intent predicts format better than industry or model does: informational queries pulled 45.5% of citations from articles and 21.7% from listicles, while commercial queries pulled 40.9% from listicles. What AEO cannot do is manufacture authority. A perfectly structured page on a site nothing references does not get cited, because the retrieval step happens before the formatting step matters. AEO raises the conversion rate of pages that are already candidates; it does not create candidates.

Is AEO different from SEO, or is it just SEO with better headings? — specifics

What does the brand-mentions finding actually mean for a budget?

It is the most actionable number in this entire field. Ahrefs studied 75,000 brands in May 2025 and measured Spearman correlations against AI Overview brand visibility: brand web mentions at 0.664, referring domains at 0.295, and backlinks at 0.218. Read that carefully, because it is easy to over-claim. It is a correlation, not a causal result, and correlation between brand size and everything else is doing some of the work. But the ordering is the finding: unlinked mentions correlate roughly three times as strongly as backlinks with appearing in AI Overviews. That inverts twenty years of link-building priority. Muck Rack's analysis points the same direction from a different angle: across more than 25 million links in ChatGPT, Claude and Gemini responses spanning 17 industries, earned media accounted for 84% of citations, journalism alone for 25-27%, and paid or advertorial content for 0.3%. Across three editions from July 2025 to May 2026, earned media ranged from 82% to 89%. The budget implication is uncomfortable for anyone who sells content: the highest-leverage spend is probably not another twenty articles on your own domain. It is getting written about, quoted, listed and reviewed on domains you do not control — which is digital PR, and which most SEO budgets are not structured to buy.

Does llms.txt work?

The evidence says no, and this is the most useful negative result in the field. Ahrefs analysed 137,210 domains and reported in June 2026 that 97% of llms.txt files received zero requests during May 2026. Not zero AI requests — zero requests of any kind. Nothing fetched them at all. Of the 3% that were fetched, the composition is instructive: SEO audit tools accounted for 21.7% of requests and general web crawlers 13.1%, while AI retrieval bots — the ones actually used by ChatGPT and Perplexity search — accounted for 1.1%. In other words, the majority of traffic to llms.txt files comes from tools checking whether an llms.txt file exists, which is a closed loop. This matters beyond the specific file. llms.txt became popular because it is the kind of tactic our industry likes: technical, cheap, checkable, and satisfying to implement. It spread for two years on the strength of plausibility rather than measurement. If you have one, leave it — it costs nothing. If you are planning a quarter's work around it, do not. The general lesson is worth more than the specific one. In a field this new, the ratio of confident advice to measured evidence is extreme, and the correct posture is to ask for the study, the sample size and the date before reallocating a budget.

What should you actually do differently?

Five things, in descending order of evidence behind them. First, invest in being mentioned rather than only in being linked. The 0.664-versus-0.218 correlation and the 84% earned-media citation share both point the same way. That means digital PR, being included in other people's comparison and listicle content, and being the source someone quotes. Second, restructure your existing pages rather than writing new ones. Answer-first opening, question-shaped headings, self-contained blocks, real tables. The format study shows this affects which content gets cited, and it is a fraction of the cost of new production. Third, publish comparison and list content, because it is what commercial-intent queries pull from — 40.9% of commercial-query citations went to listicles in the Wix Studio data. Note that this is the listicle figure and not a separate comparison-article figure; we could not verify the widely quoted claim that comparison articles specifically account for 32.5% of AI citations, and we are not repeating it. Fourth, put a visible verification date on anything containing numbers, and actually re-verify on a cadence. A page with stale prices and no date is a page that will be cited wrongly and then stop being cited. Fifth, change what you measure. If 68% of searches produce no click, ranking reports measure something that no longer converts. Track brand mentions, citation appearances in AI answers, and conversion rate by traffic source — especially given that AI-referred visitors reportedly convert at 4.4 times organic.

Figures we deliberately did not publish

Price log — what was read, and where

What percentage of searches are zero-click in 2026?

68.01% of US Google searches in January to April 2026 ended without a click, according to SparkToro's analysis of Similarweb panel data published on 9 June 2026. That is up from 60.45% in 2024 and roughly 49% in 2019. The frequently quoted 58.5% figure is superseded.

How much do AI Overviews reduce clicks?

Two credible studies disagree. Ahrefs measured a 34.5% reduction in position-one CTR across 300,000 keywords, comparing March 2024 to March 2025. Seer Interactive measured a 61% decline in organic CTR on AI Overview queries between June 2024 and September 2025, from 1.76% to 0.61%. The honest answer is 'a large decline in that range'.

Do brand mentions matter more than backlinks now?

For AI Overview visibility, the correlations say yes. Ahrefs studied 75,000 brands and measured Spearman correlations of 0.664 for brand web mentions, 0.295 for referring domains and 0.218 for backlinks. That is correlation rather than causation, but the ordering — mentions roughly three times backlinks — is the finding.

Does llms.txt actually work?

The evidence says no. Ahrefs analysed 137,210 domains and found that 97% of llms.txt files received zero requests of any kind during May 2026. Of the 3% that were fetched, AI retrieval bots accounted for 1.1% of requests, behind SEO audit tools at 21.7%. Leave one in place if you have it; do not plan a quarter around it.

Book a 30-minute build call · Pricing · Call the live agent