Skip to content
Antoine Le LouëtIndependent senior developer (Home)
Back to the Journal

Technology5 min read

SEO, AEO, GEO: where is the decision made?

AI answers shift some attention to before the website visit. What this changes for content, citations and measuring results.

Three visibility goals: SEO leads to a website click, AEO to extracting a passage and GEO to a citation in an answer. These goals overlap.
Clicks, extraction and citations describe goals that can coexist within one search journey.

Part of the choice forms before the visit

A business owner asks an assistant such as ChatGPT or Perplexity how to choose a provider to take over their web application. The answer describes criteria, highlights risks and may suggest names. Before visiting any website, the reader already has a framework for comparing offers. In this hypothetical example, part of the decision is already being shaped in the answer.

This is how I suggest reading the SEO, AEO and GEO debate: as a change in the distribution of attention. Treating them as three independent disciplines hides their shared goal: making information accessible when someone is forming a choice.

The website retains a decisive role in checking a reference, understanding an offer or making contact. But measuring its visibility solely by visits leaves out what happens beforehand.

From clicks to citations: overlapping goals

  • SEO, Search Engine Optimization: make a page visible in search engines. Attracting a click to the website remains a frequent goal without summarising the entire practice.
  • AEO, Answer Engine Optimization: make an answer easy to identify and extract. A passage may directly answer a question in a snippet or voice response, with little surrounding context.
  • GEO, Generative Engine Optimization: seek citations or mentions in a generated answer, among sources or solutions presented to the reader.

These terms are used here as reference points. One page can attract a visit and supply a passage for a synthesis. AEO predates generative assistants; GEO extends longstanding questions of credibility and search visibility. No acronym guarantees that an engine will select a source.

What the 2024–2025 data shows

The SparkToro study published on 1 July 2024, based on the Datos panel, estimates that 58.5% of Google searches in the United States and 59.7% in the European Union end without a click. Its scope covers desktop and mobile web browsing, excluding Google apps and iPhones, among others. No click may mean a sufficient answer, abandonment or another search. It does not measure a purchase decision and cannot be attributed entirely to AI.

Pew Research Center studied 68,879 Google searches by 900 American adults in March 2025, then reconstructed results in April. Its analysis published on 22 July found an AI summary in 18% of searches. A traditional link was followed in 8% of visits with a summary, compared with 15% without one; a link within the summary itself was followed in 1% of relevant visits.

These observations associate summaries with fewer clicks. They do not isolate a causal effect, because questions triggering a summary differ from other questions. The 18% figure describes this American sample in 2025, not all French searches in 2026. Comparing different assistants’ monthly and weekly audiences would likewise not measure the purchases they influence.

For informational content, the economic risk is concrete: the answer may satisfy the need before the visit. I therefore consider traffic forecasts based solely on keyword volumes more fragile. Profitability also depends on the value of contacts obtained and production cost; these studies do not demonstrate a uniform decline in return on investment.

Give the engine a reason to retain the source

Google’s documentation states that SEO fundamentals remain valid for AI Overviews and AI Mode. No special markup or AI-specific file is needed to appear there. A page must, among other things, be indexed and eligible to display a snippet. Meeting these conditions does not guarantee selection.

Google also explains that its AI features may launch several related searches to construct an answer. Distinguish data used to train a model from pages consulted at search time. Being present in a training corpus guarantees neither a citation nor a recommendation. Inclusion in an answer may also come from a page found at that moment.

For a small or medium-sized business, I would start from information it can genuinely document: an authorised client case, a method explained with its limitations, precise answers to prospects’ questions. The author, date and sources should make claims verifiable. Explicit headings help locate a passage; original information provides a reason to reuse it.

Distribution then extends beyond the website. An up-to-date professional profile or a useful contribution to a specialist publication creates other opportunities to encounter the business. I would favour channels its clients use, with consistent, verifiable information. Publishing the same text everywhere to multiply mentions is not evidence of authority.

A citation still needs to produce a useful effect

A cited source may remain unnoticed, be misinterpreted or never be opened. A brand mention, a linked citation and a commercial recommendation describe different results. Their value should be examined separately.

I suggest choosing around ten questions actually asked by prospects. At regular intervals, record answers from several engines, their sources and any errors about the business. Preserve the date, wording and conditions of each observation: answers may vary by session or context. This monitoring provides evidence on a sample, not a comprehensive market share.

In parallel, track identifiable visits from these tools and qualified enquiries. Asking prospects how they built their shortlist may illuminate a journey website statistics do not fully capture. More citations without an observable effect on contacts remain a signal to examine, not proof of profitability.

SEO remains useful. For businesses whose prospects use these answers, a strategy limited to clicks now misses part of the journey. Before producing the next article, I would ask which question it helps resolve, what evidence it offers and how to verify its effect on enquiries received. Views or reactions alone do not fund the business.

Sources and useful links

Share on LinkedIn (new tab)

Keep reading
All articles in this category