Every week another email lands announcing that SEO is dead and that you now need to buy a GEO consultancy package. The factual part of the warning is correct: search has changed. The commercial part deserves a closer reading.
We have gathered the public data available up to September 2026, cross-checked it against what we see in the accounts we manage in Brazil and Germany, and separated what has genuinely changed from what is simply a new name for old work.
What the numbers show
The first figure is not about artificial intelligence at all. It is about clicks. According to SparkToro’s study, using Similarweb panel data, 68.01% of Google searches in the United States ended without a single click between January and April 2026. In 2024 that figure was 60.45%.
The second figure explains part of it. The Pew Research Center tracked 68,879 real searches from 900 adults and found this: when an AI summary appears, users click a traditional result in 8% of visits. When no summary appears, they click in 15%. And the link cited inside the summary itself is clicked in only 1% of visits.
The third figure is the one that rarely makes it into the sales email: traffic coming from AI platforms is still small. Similarweb recorded 9.5 billion monthly visits across AI platforms between June 2025 and May 2026, up 70% year on year. That is a lot in absolute terms and still modest next to Google. Over the same period, the share of ChatGPT responses citing websites rose from 1.6% to 6.8%.
The honest reading of those three numbers is not “traffic has moved to AI”. It is something less convenient: organic traffic has shrunk, and only a small though growing slice reappears through another route. Businesses that address only one of the two movements lose at both ends.
What GEO actually is, without the marketing around it
GEO stands for Generative Engine Optimization. The term did not come from an agency. It came from an academic paper by researchers at Princeton and IIT Delhi, presented at KDD 2024, which tested systematically what makes content get cited by generative systems.
The paper’s central finding is that adjustments to the form of the content can raise visibility in generated answers by up to 40%. The adjustments that worked best are uncomfortably simple: cite your sources, include verifiable statistics, and quote named experts directly. There is no technical trick in there. It is editorial rigour, the same quality that separates a useful article from filler.
Where GEO and SEO are the same thing
Google’s own developer documentation is blunt about AI Overviews and AI Mode: there are no additional requirements and no special optimisations needed to appear in those features. In the documentation’s own words, you do not need to create new machine-readable files, AI text files, or markup.
What worked still works. The page has to be indexed, eligible for snippets, fast, correctly structured and genuinely answering the question. A company with solid SEO foundations already has half of its GEO work done. A company without them will not fix the problem by buying a new layer on top.
Where GEO genuinely differs
There are four concrete differences, and this is where the new work sits.
1. The metric changes. In SEO you track position, impressions and clicks. In GEO you track citations and brand mentions: how often your company appears in answers to the questions that matter to your business, and in what context. A brand can be cited hundreds of times without receiving a single click and still win the sale, because the decision was formed inside the conversation.
2. The source changes. Semrush analysed more than 230,000 prompts and over 100 million citations across ChatGPT, Google AI Mode and Perplexity. The most-cited domains are not corporate websites. They are Reddit, Wikipedia, LinkedIn, YouTube and news publishers. That carries a hard practical consequence for anyone who has only ever invested in their own site: much of your AI visibility is built off your domain. Editorial presence, forum participation, content signed by real people and press coverage have stopped being branding and become search infrastructure.
3. Technical access changes. The crawlers feeding those answers are not the familiar ones. OAI-SearchBot, GPTBot, PerplexityBot, ClaudeBot and Google’s Google-Extended token each need a deliberate decision in your robots.txt and your firewall. Plenty of companies blocked everything in 2024 as a precaution and are now invisible in those answers without knowing why. Check this before you buy any service.
4. The question changes. Nobody types “SME management software” into a chat. They type “which management system is worth it for a 12-person company in Germany that already uses Datev”. Long phrasing, embedded context, explicit comparison. Content that answers real questions with numbers and decision criteria gets cited. Generic content is ignored.
What is not working
A useful case for scepticism: the llms.txt file, sold through 2025 as the robots.txt of the AI era. Ahrefs analysed 137,210 domains in May 2026 and found that 97% of published llms.txt files received no requests at all during the period. Among the few that did, AI retrieval bots, precisely the ones that would matter for visibility, accounted for 1.1% of requests.
The rule we apply for clients is simple: until there is evidence that the platform reads it, it is a talking point, not a deliverable. And one reminder that saves money: nobody sells you a position in ChatGPT. There is no auction, no purchasable ranking, no guaranteed first answer. Anyone promising that is selling something else.
Europe and Brazil: two contexts, one strategy
In Europe there is an additional layer, and it is regulatory. The European Publishers Council filed a formal antitrust complaint against Google over AI Overviews and AI Mode in February 2026, and the European Commission had already opened an investigation in December 2025 into the use of publisher content. In practice, the rules on consent, opt-out and remuneration are still going to change. European strategies need to be built with that instability in mind rather than in spite of it.
In Brazil, adoption is mass market. OpenAI disclosed in August 2026 that Brazilians send around 215 million messages a day to ChatGPT, up from 140 million a year earlier, with the user base close to doubling over the period. Portuguese is the platform’s third most-used language. For an SME selling into Brazil, the conversation about your sector is already happening, with or without you in it.
What connects both markets is language. A company operating in Berlin and in Brazil needs to be cited in German, in English and in Portuguese, and answers generated in each language draw on different sources. Translating the website does not solve it. Publishing original content in each language does.
Checklist: what to do in the next 90 days
- Audit robots.txt and the firewall, and make a deliberate decision about which AI crawlers may access the site.
- Measure your baseline: how often your brand is cited today in answers to the 20 questions that matter to your business, in each language.
- Separate AI traffic in GA4, with dedicated sources for chatgpt.com, perplexity.ai, gemini.google.com and claude.ai.
- Review structured data: Organization, Product, FAQPage and Article, correct and validated.
- Rewrite your five most strategic pages to answer real questions, with dated figures, cited sources and explicit decision criteria.
- Publish at least one piece a month containing proprietary data, meaning something only your company can assert. That is what earns citations.
- Build presence off your domain: trade press, LinkedIn with named authors, forums and niche communities.
- Keep name, address, description and positioning consistent across every public source, because these systems cross-reference them.
- Agree the success criteria before you start: citation rate, share of voice in answers and attributed revenue, compared against the previous quarter’s baseline.
Want to know whether your company is being cited?
Agência Lira is a digital marketing and corporate communications consultancy working with small and medium-sized enterprises across Europe and Brazil. We cover strategy, paid media, content and digital structure, in English, German and Portuguese.
We run an initial AI visibility diagnostic: we map the questions that lead to your business, measure how often your brand appears in answers from ChatGPT, Gemini and Perplexity, benchmark that against your direct competitors and show you where the gap is. If the problem turns out to be SEO, we will tell you it is SEO. If there is nothing worth optimising yet, we will tell you that too.
Request your free audit and we will assess your case candidly.
If you would rather understand the scope of the work first, here is how Agência Lira’s marketing consultancy works.
Sources
- SparkToro, using Similarweb panel data, zero-click search study, June 2026
- Pew Research Center, “Google users are less likely to click on links when an AI summary appears in the results”, July 2025
- Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, “GEO: Generative Engine Optimization”, ACM SIGKDD 2024
- Google Search Central, official documentation on AI features in Search
- Semrush, “The Most-Cited Domains in AI”, 13-week study covering more than 230,000 prompts
- Ahrefs, llms.txt study across 137,210 domains, May 2026
- Similarweb, generative AI search statistics, June 2025 to May 2026
- OpenAI, Brazil usage figures released in August 2026
- European Publishers Council, antitrust complaint against Google, February 2026