GEO doesn’t replace SEO: it extends it. Generative engines (Google’s AI Overviews, AI Mode, ChatGPT, Perplexity or Gemini) don’t make up their answers out of thin air: they retrieve content from search indexes before writing them. That’s why being indexed and having authority are still the price of entry. What changes is the goal: ranking a URL in the top 10 is no longer enough; the challenge now is getting your brand cited or mentioned inside the answer the user reads. Here’s what each discipline is, what the 2025–2026 data says and what you should do if you want your business to stay visible.
GEO vs SEO
| Dimension | SEO | GEO |
|---|---|---|
| Goal | Rank URLs in the SERP and win the click | Be cited, mentioned or recommended inside the generated answer |
| Unit of competition | The page, against 10 results | The passage or snippet, against a set of sources the model synthesises |
| Metrics | Rankings, CTR, organic traffic, conversions | Citation share, brand mentions, sentiment, referral traffic from AI, branded searches |
| Key signals | Relevance, links, technical signals, E-E-A-T | All of the above + passage extractability, verifiable data, third-party brand mentions, freshness |
| Query type | Short, keyword-based | Conversational, long and broken down into sub-queries (query fan-out) |
The simplest way to see it: SEO gets people to click your link; GEO gets AI to cite and mention you. And they aren’t mutually exclusive.
“The mistake I see most often is treating GEO as a new kind of SEO that has to start from scratch. It isn’t. If your website isn’t crawlable, indexable and authoritative, no amount of GEO will help: the models draw on the same old indexes. GEO starts where SEO is already done.” – Rafa Villaplana, SEO consultant and CEO of Linking & Growing
What GEO (Generative Engine Optimization) is
The term comes from an academic paper, “GEO: Generative Engine Optimization” (Aggarwal et al.), published on arXiv and presented at the KDD ’24 conference. Its authors define GEO as the first framework to help content creators improve their visibility within the answers of generative engines: systems that combine a search engine with a language model, retrieve documents and write an answer backed by sources, with inline attribution.
The problem it addresses is new. As the paper itself points out, generative engines reduce the need for users to visit websites, which cuts traffic, and creators have little or no control over when and how their content is shown. Visibility stops being a position in a list and becomes how many of your ideas end up in the answer.
What works, according to the original study
The researchers tested nine techniques on a generative engine and measured their impact on visibility. The three most effective were:
- Adding quotations from relevant sources.
- Adding statistics: replacing vague claims with quantitative data.
- Citing reliable sources.
These techniques achieved visibility improvements of between 30% and 40%, and up to 37% on Perplexity. An honest nuance almost nobody mentions: that “up to 40%” is a relative improvement in visibility within the answer, measured in 2023–2024 with GPT-3.5. It doesn’t mean 40% more traffic or sales. Use the figure precisely.
And a damning finding for anyone looking for shortcuts: keyword stuffing doesn’t just fail to help, it hurts. It performed 10% below the baseline on Perplexity. The old keyword-stuffing tactics don’t carry over to generative engines.
How generative engines work
AI engines work as a retrieval-augmented generation (RAG) system: they break your query down into several sub-queries (what Google calls query fan-out), retrieve documents from the index, summarise them and write an answer with citations. This explains the most important figure in the whole 2026 landscape:
Only 38% of the pages cited in AI Overviews are in the organic top 10 for that search (Ahrefs, March 2026); a year earlier it was 76%. In other words, the model cites pages that rank for the sub-queries, not necessarily for the main query. Winning the top 10 no longer guarantees the citation.
Each engine also draws on different sources:
- Google’s AI Overviews and AI Mode run on Gemini over Google’s index. Google is explicit: to appear, a page only needs to be indexed and eligible to be shown with a snippet, with no additional technical requirements.
- ChatGPT Search retrieves mainly from Bing’s index and its own. It’s worth being indexed in Bing and allowing the OAI-SearchBot crawler.
- Perplexity runs a live search for each query using its own index and prioritises relevance, freshness and credibility.
- Gemini is growing fast: in one year it went from ~9% to over 25% of traffic to AI tools, while ChatGPT fell from 76% to around 55% (Similarweb, 2026). Optimising for ChatGPT alone is no longer enough.
What the impact data says (2025–2026)
Every figure with its source, because this topic is full of hot air:
- Position 1 CTR drops by 58% when an AI Overview appears (Ahrefs, December 2025 data on 300,000 keywords). A year earlier the measured drop was 34.5%.
- 8% of searches lead to a click with an AI summary vs 15% without one, and only 1% of clicks go to the links inside the summary itself (Pew Research Center, July 2025, the most solid independent source).
- Google sends 190 times more traffic to websites than ChatGPT (Ahrefs, February 2026). AI traffic is small, around 0.2–0.3% of the total, but it’s growing very fast and converts better: on Ahrefs’ own site it was 0.5% of visits but 12% of sign-ups.
- Brand mentions carry more weight than backlinks for AI visibility: a correlation of 0.664 vs 0.218 (Ahrefs, 75,000 brands). YouTube mentions had the highest correlation of all the factors studied (0.737).
- Fresh content gets cited more: URLs cited by AI are on average 25.7% newer than organic ones.
“Look at what this means for an SME. The traffic AI sends you today is tiny in volume, but people arrive ready to buy, they convert better, and much of AI’s influence you don’t even see: someone discovers you in ChatGPT and then searches for your brand on Google. If you only look at the last click source, you’ll miss half the story. I always measure two layers: rankings in Google and mentions in the models.” – Rafa Villaplana, SEO consultant and CEO of Linking & Growing
What works in GEO and what’s hype
This is the section that separates a serious strategy from a list of tricks. I’ve split it by evidence:
Backed by evidence
- Content with data, sources and quotes. It’s the only recommendation with direct academic backing. Replace “we greatly improved traffic” with “we increased organic traffic by X% in Y months, according to Search Console”.
- Extractable structure. A direct answer at the start of each section, headings that mirror users’ real questions, passages that make sense on their own, comparison tables. Also cover the fan-out sub-queries: comparisons, prices, alternatives, “how to”, “who it’s for”.
- Brand mentions and entity building. Digital PR, “best X” lists, industry media, YouTube, reviews. It’s the signal most correlated with AI visibility.
- Real freshness. Update the substance (data, examples, prices), not just the date.
- Technical accessibility. Be indexed in Google and Bing, allow OAI-SearchBot and PerplexityBot, don’t block snippets and serve key content in visible HTML.
- E-E-A-T. Identifiable authors, transparent methodology, first-party data.
No evidence (or evidence against)
- llms.txt. Across ~300,000 domains analysed, having one doesn’t predict citations; in fact, the models predicted better when that variable was ignored (SE Ranking, November 2025). Only 10% of domains have one, and Google has said it doesn’t support it.
- Schema as a citation lever. After tracking 1,885 pages that added structured data, no significant improvement in AI citations was found (Ahrefs, May 2026). Keep schema for rich results and the Knowledge Graph, but don’t sell it as a GEO trick.
- Keyword stuffing. It hurts, as we’ve seen.
“When someone promises to ‘rank you in ChatGPT’ with a magic file or a few schema tags, be wary: the data doesn’t back it up. What does work is the boring stuff, the usual things done well: content with real data, your brand showing up on sites that matter, and rigour. There’s no shortcut.” – Rafa Villaplana, SEO consultant and CEO of Linking & Growing
Are they opposed or complementary?
The industry consensus in 2026 is clear: they complement each other, and SEO is the foundation. Google says so (SEO best practices are still relevant), Ahrefs sums it up as needing both, with SEO as the foundation and GEO as the way to secure your future, and the technical architecture itself confirms it: if you can’t be crawled and indexed, you can’t be cited.
The nuance I stand by with my clients is this: SEO is necessary but no longer sufficient. GEO adds three disciplines that traditional SEO barely covered: PR and entity building, designing content so it can be extracted, and measuring visibility across prompts as well as keywords.
The state of GEO in Spain
Spain is going through this transition with very high AI adoption:
- 22% of Spaniards already use chatbots instead of a search engine most of the time, the highest rate in Europe (Bain & Company, 2025). Generative AI usage rose from 5% in 2023 to 61% in 2025.
- Organic search fell from 52% to 40.7% of traffic for the sites analysed in Spain, a drop of more than 11 points compared with 2.6 points globally (SE Ranking, 2026).
- In Spain, ChatGPT mainly cites El País, La Vanguardia and AS: appearing in the major national media is one of the most powerful GEO levers for a Spanish brand (Growwer, 2026).
The takeaway for any business targeting Spain: the window of opportunity in GEO is wider open there than in other markets, because adoption is running ahead of rigorous competition in Spanish.
Frequently asked questions
Will GEO replace SEO?
No. AI engines retrieve content from search indexes before answering, so technical SEO and authority are still essential. GEO is a layer added on top of SEO, not a replacement.
What matters most for appearing in ChatGPT or Google’s AI Overviews?
According to 2026 data, mentions of your brand on other sites (media, YouTube, lists, reviews) correlate with AI visibility far more than backlinks do, and content with verifiable data and sources gets cited more. The foundation is still being indexed and accessible to crawlers.
Is the llms.txt file any use?
For now there’s no evidence that it increases citations in AI models: across hundreds of thousands of domains analysed, having one doesn’t predict visibility. It only makes sense for technical documentation consumed by agents. It isn’t a GEO priority.
How much traffic does AI really send to a website?
Today it’s small, around 0.2–0.3% of the total compared with almost 40% from Google, but it’s growing very fast and tends to convert better, because users arrive further along in their decision. Much of its influence also shows up as branded traffic on Google.
How do I start working on GEO for my business?
In this order: make sure of technical accessibility (indexing, crawlers, visible HTML), rewrite your most valuable pages so they’re extractable (direct answer, sourced data, tables), build brand mentions in relevant media and platforms, and measure in two layers: SEO in Search Console, and GEO by tracking prompts and citations.