The confusion has an understandable origin: the acronyms look alike, the tools overlap, and nobody enjoys admitting there is a new discipline that starts from zero. But one detail changes everything, and it is an asymmetry.
Optimising for GEO improves your SEO systematically. Optimising for SEO does not move your AI visibility. That alone reorders the investment priorities of most teams.
01Two different outputs#
Google indexes pages, ranks them and presents a list. The user decides where to click. A generative engine takes a question, consults its memory and its retrieval capability, synthesises several sources and generates an answer. No list, no position, and often no click.
| Variable | SEO | GEO |
|---|---|---|
| Output | Ordered list of links | Answer synthesised in natural language |
| Success metric | Position, CTR and organic traffic | Mention rate, SOV and Brand Impact Score |
| Main factor | Backlinks, keywords and domain authority | Semantic authority and entity clarity |
| Horizon | Stable, months or years | Volatile: 40% to 60% of sources change each month |
| Scope | Your domain | The whole ecosystem where you are mentioned |
| Relationship | Does not improve GEO | Improves SEO systematically |
The figure that bothers an SEO team most is the CTR one. When an AI Overviews block appears, the click on position one falls by 58%, and some analyses document drops of 79%. Staying first is worth less and less if the answer gets consumed without leaving the results page.
02Backlinks do not move GEO#
Backlinks have been the heart of SEO since the nineties: authority votes Google uses to rank. Language models do not work that way. They were trained on corpora where what counted was semantic coherence, consistency across sources, and how often a concept appeared associated with a brand.
A link from a major outlet helps on Google and does not necessarily raise the odds of a model citing you. What does move them is different: how clearly you define your category, how precise your claims are, and your presence in the sources models weight most. Wikipedia shows up in 47.9% of ChatGPT's citations, and a well-structured category page mentioning you as a reference gives you no traffic, no domain authority, and appears in no SEO report.
Technical detail · why prompts inferred from Google carry bias
Tools built on Google search data do something logical and wrong when they move into the generative world: they take the most-searched keywords and rephrase them as questions. The problem is that nobody talks to ChatGPT the way they type into Google. On Google you type "best CRM small business"; to a model you explain the whole situation, with team size, budget and the integration you need. A prompt inferred from a keyword measures a universe that does not exist, and the bias is systematic rather than random.
03What to measure, and with what#
In SEO you measure position, CTR and organic traffic: objective, historical, comparable metrics. In GEO the first one is mention rate, how often a model includes you when answering relevant questions in your sector. On top of that come Share of Voice and Share of Branded Voice, which filters out brand-free answers and measures your real share of the competitive space. And Brand Impact Score, which combines frequency, position within the answer and sentiment.
None of these exists in SEO tools, which is why measuring them is different work. GEORadar measures them with populations of personas, intents and funnel stages, 6,000 to 30,000 prompts per study against the seven surfaces, and with proprietary KPIs per vertical on top of the universal ones.
The rest divides up: the technical side, whether crawlers can collect your content, is audited by GEOdoctor; the alignment between what you say and what you get asked is S.A.M.; and both sit in the product stack.
If one idea survives this piece, let it be the asymmetry. Starting with GEO gives you SEO as a bonus. Starting with SEO leaves you exactly where you were in AI answers.
An earlier version of this text was published on zoopa.es.
- Aggarwal, P. et al. GEO: Generative Engine Optimization. KDD 2024, arXiv 2311.09735. https://arxiv.org/abs/2311.09735
- Ahrefs. The impact of AI Overviews on position-one CTR, December 2025. https://ahrefs.com/blog/ai-overviews-reduce-clicks/
- Semrush AI. Analysis of sources cited by ChatGPT, 2025. https://www.semrush.com/blog/
- BrightEdge. AI Visibility Report, 2026. https://www.brightedge.com/resources
Runs 498A, the applied AI lab of the Zoopa group, and GEORadar. Guest lecturer at the UAB, where he supervises AI master's projects with IIIA-CSIC.