O que é AEO ou GEO, a otimização para motores de resposta de IA?

É o trabalho de fazer com que uma marca seja nomeada nas respostas geradas por motores de IA como o ChatGPT, o Perplexity ou os resumos de IA do Google, e não apenas classificada numa lista de ligações. Ser nomeado depende mais da autoridade e da coerência da marca do que de truques técnicos.

Truques de SEO fazem uma empresa aparecer nas respostas do ChatGPT?

Ajudam um motor a indexar a página, mas o modelo não classifica uma página isolada. Ele reproduz a associação que o registo credível da web faz entre uma categoria e um nome. O que decide é a autoridade e a coerência da marca, não a otimização técnica da página.

É verdade que os resumos de IA reduzem as visitas ao site?

A Pew Research mostrou que, com um resumo de IA no topo dos resultados, os utilizadores clicaram num resultado tradicional em 8% das visitas, contra 15% sem resumo, e apenas 1% clicou na fonte citada no resumo. A descoberta passa a acontecer dentro da resposta.

Aparecer nos motores de resposta de IA pode ser garantido?

Não. A visibilidade num modelo é probabilística: varia por motor, pela forma como a pergunta é feita, e muda ao longo do tempo à medida que os modelos são retreinados. Quem promete primeiro lugar no ChatGPT vende certeza que não tem. Constrói-se o padrão, não a garantia.

A pesquisa tradicional acabou por causa da inteligência artificial?

Não. O Google continua a dominar o volume de pesquisa. O que mudou é que uma camada de respostas se colocou por cima: em cerca de 18% das pesquisas surge um resumo de IA, e motores como o ChatGPT respondem sem lista de ligações. Um novo intermediário passou a decidir quem é nomeado.

Brand in AI answers: authority decides who is named, not SEO

There is a question your potential clients already ask, and that your company does not hear. It is not asked in a meeting or on a form. It is typed into a text box, to an artificial-intelligence engine, and it sounds more or less like this: who does this, well done, in Portugal. On the other side, the machine answers with a list. On that list may be your name, your competitors may be there, or there may be a vague, and sometimes wrong, description of what your company does. What rarely happens is that the question reaches you. The decision of who appears in that answer has already been made, and it was not made by you. This article is about how it is made, about the evidence that the starting point of the search has moved, and about why who appears in the answer is decided with brand authority, not with SEO tricks.

The starting point of the decision has moved

For twenty years, anyone looking for a supplier began in the same place: a search box, ten blue links, and the short list the decision-maker built from them. That gesture is being replaced. Increasingly, the starting point is a machine that answers directly, with no list. The scale of the change is measurable. ChatGPT went from about 400 million weekly active users in early 2025, a figure reported by OpenAI, to more than 800 million in October, announced by Sam Altman at the company's DevDay event. This is not a curiosity of technological adoption. It is a change in where a decision-maker's first question lands.

And when the answer appears, it intercepts the visit. The Pew Research Center analysed the real browsing of 900 adults in the United States throughout March 2025. The conclusion is direct: when there was an AI summary at the top of the results, users clicked on a traditional result in 8% of visits; when there was not, they clicked almost twice as often, 15%. And within the summary itself, the link to the cited source was clicked in only 1% of visits. About 18% of all searches generated an AI summary. The click that once took the decision-maker to your site is being absorbed by the answer.

On the classic SEO side, the effect is confirmed. Ahrefs compared the click-through rate of 300,000 keywords with aggregated Google Search Console data and concluded, in April 2025, that the presence of an AI summary was associated with a drop of about 34.5% in the click-through rate of the first organic position. A reanalysis by the same authors, published in February 2026 with December 2025 data, raised that figure to 58%. The number to retain is not the exact percentage, which incidentally keeps rising. It is the sentence both support: being first no longer guarantees the visit. Discovery now happens inside the answer, and not on the page the answer points to.

This is not search dying

It is worth rejecting the exaggeration straight away, because it contaminates the analysis. The honest reading of this same data is not apocalyptic. Google has not collapsed and still dominates search volume. What happened is quieter, and more consequential: a layer of answers has placed itself between the question and the ten links. In about one in every five Google searches (the Pew estimate), an AI summary now sits above the results. In a separate universe, ChatGPT and Perplexity answer directly, with no list at all.

It is worth separating the two things, because the market tends to confuse them. Google's AI summaries (the AI Overviews) are Google synthesising its own index above the results, and it is these that the Pew and Ahrefs data speak of. Pure answer engines, such as ChatGPT and Perplexity, synthesise the answer from sources they retrieve from the web, often with citation, and even so the visit to your site may never happen. They are different systems. What they have in common is the essential: in both, a new intermediary decides who is named. The question a company has to ask is no longer just whether it is in first place. It has become whether it is even inside the answer. And that second question has a different owner.

Who is named is not decided with SEO

When a model answers who does this in Portugal, it is not ranking a page against a keyword. It is reproducing, in a sentence, the association that the written, credible record of the web has built between a category and a name. It names the brand that the web most consistently, and most credibly, links to that category. This is old brand theory in new clothes. Al Ries and Jack Trout described, in 1981, positioning as the place a brand occupies in the mind of the decision-maker: it is won by occupying a category association, not by shouting about the product. The machine's memory is now one more mind to occupy, and it is occupied in the same way, by being the name that credible sources keep attaching to the category.

David Aaker's framework, from 1991, applies with a new edge. Brand equity, he wrote, is the set of assets, and liabilities, linked to the brand's name and symbol, that add to or subtract from the value the company offers. The decisive word is the second. An indistinct brand is not worth zero, it subtracts. A company that looks like every other in its sector gives the model nothing to grasp, and the model grasps the association to another name. This is where the phrase what is alike has no authority ceases to be a stylistic stance and gains a consequence for distribution.

This is also why the SEO reflex fails on this terrain. You do not get into a model's answer by way of a well-written page title, because the model is not reading your page in isolation. It is synthesising a consensus about who is credible in a category, from everything it has seen. Schema markup, keyword density, a technically flawless page: all of that helps a crawler index you. None of it builds what the model actually rewards, which is a coherent, repeated and credible presence, saying the same about you everywhere. The technical trick optimises the page. What the machine names is the brand.

The part our own sector does not want to admit

The marketing market has already found a name for this (AEO, GEO) and is already selling it as another bag of tricks: appearing in ChatGPT in thirty days, AEO checklists, technical tweaks. It is the SEO playbook repeated, and it will disappoint for the same reason, because it treats a brand problem as if it were a technical one. The uncomfortable admission, for those who sell communication, is that here there is no lever to pull for a fixed price that produces the result the following week. You do not buy a place in the answer with ads, because the answer is not an advertising space. And you cannot reliably fool the model into letting you in either.

The only durable factor is authority: a credible, consistent and distinct presence, built over time. That is slow, it is difficult, and it is precisely for that reason that it is worthwhile, because what is slow and difficult to build is difficult for a competitor to replicate, and what is difficult to replicate is defensible. The companies that will be named two years from now are building that presence now, against the current of a market that would rather sell a checklist. Even measuring how a brand appears in this channel is not yet common practice. The discipline is young. The exposure is not.

The money: what being named, or omitted, does to the bottom line

The legitimate objection of whoever signs the cheque is the usual one: show me the money. The best way to see it is through the short list, the set of three or four names a decision-maker takes into a decision. That list used to be formed by the decision-maker's own research. It is now formed, or pre-filtered, by the machine's answer. Three consequences for the accounts follow, presented without promising what the evidence does not guarantee.

The first is the cost of acquisition. A brand named in the answer enters the short list with the credibility of having been recommended by a neutral-looking machine. It arrives half convinced, and that lowers the cost of turning it into a client. A brand absent from the answer has to buy its way back into a conversation from which it was excluded, and that raises the cost of acquisition. The second is the cost of omission, the only one that appears on no invoice. You receive no report of the deals in which you were never considered because the machine named someone else. It is the silent version of losing, the same form as the discount an indistinct brand pays without noticing: you never see the client who never called.

The third is defensibility. A position consistently associated with your category, in credible sources, is expensive to dislodge, precisely because it was not bought, it was earned across the record. Once the model attaches the category to your name, removing you from there requires a competitor to build more authority than yours, not to bid higher than your ad budget. None of these three things appears in Friday's campaign report. All of them appear in the answer the decision-maker's machine gives two years from now.

What this does not prove

None of this proves that working on authority will place your name in an AI answer, much less that it will keep it there. The caveats, because without them the numbers are worth nothing.

The evidence that clicks are intercepted (Pew and Ahrefs) is about Google's AI summaries and about click behaviour. It does not prove how a model like ChatGPT chooses who to name, because they are different systems. Pew's data is from adults in the United States, in March 2025, gathered by a panel that installs a tracking application; it is observational, does not establish cause, and is not Portugal. Ahrefs's figures are correlation with modelling, and come from an SEO-tools company. The 800 million is usage reported by OpenAI itself, without independent audit, and usage is not the search for suppliers.

More important: visibility within an answer is probabilistic. It varies by engine, by how the question is written, and it changes over time as the models are retrained. There is no first place in ChatGPT to guarantee, and whoever promises it is selling a certainty they do not have. What the evidence supports is a pattern: brand authority and coherence move visibility in this channel as a rule of the mechanism, not as a promise for a specific company. The result for a particular business is measured case by case. There are, in fact, figures circulating on this subject that we decided not to cite, because we could not confirm them in the primary source. That measurement, done seriously, is where a diagnosis begins, not where a sales argument ends.

Authority is no longer just reputation. It has become distribution.

What changes when you look at this as a director, and not as a campaigns manager, is the status of authority. It was treated as reputation, a soft asset, a decoration that could be deferred. The change this article describes gives it hard work: it comes to decide whether the company appears at the exact point where the decision-maker begins today. It is the same asset Aaker described, the one that adds or subtracts value, now paying in a new currency, distribution. Spreading ads does not buy a place in the answer. An SEO trick does not buy it durably. What buys it is being, in the credible record, the brand the category summons.

At NP we treat this as what it is, an asset that is measured and defended, not a matter of style. It is what the Method makes auditable, point by point, rather than a matter of opinion, and it is on this asset that we exist to work: converting perception into a financial asset, including the value of being the company the machine names when the client asks who does this.

Sources

Every number in this article was verified in the primary source. Where the source does not support the current reading, or has an interest in the outcome, we say so in the text. The conceptual frameworks are attributed by paraphrase, without verbatim citation, so as not to fabricate sentences the authors did not write.