SEO & AI Search

SEO is no longer just Google. And I think that's a good thing.

AI answers change the path to information, but they don't remove the need for a good website. How I look at visibility, sources and the promises of GEO tricks.

When someone is looking for an answer, they don't much care which category we file them under. They might open Google, ask ChatGPT, continue a conversation in Gemini or look at Perplexity. They want to understand an option, solve a problem or decide whom to trust.

That's why I don't find it particularly useful to debate whether SEO is dead. I'm more interested in what is changing in search and what I should do with a website because of it. My answer, for now, is not to abandon a solid technical foundation and good content. Quite the opposite: it seems to me we need them even more.

That doesn't mean it's enough to carry on as before and simply rename the changes. The way someone reaches a piece of information also affects the role our site plays. Sometimes it will be the destination of a visit. Other times it will be one of the sources in an answer that a person reads somewhere else.

With a classic search result I picture a fairly clear sequence: someone types a query, sees a list of results and picks a link. There are many different elements there too, but the position of an individual page is a reasonably understandable part of the story.

With an AI answer the path can be different. The system composes an answer, uses several sources, perhaps cites them, and allows follow-up questions. The user may make part of the decision before opening any website at all.

That pushes me towards a different question. Not only where my site sits in the list, but whether a clear and useful answer can be drawn from it. If a service description requires a lot of guessing, a person will struggle with it. I see no good reason to expect a system to understand unclear text any better.

At the same time, I don't assume every AI answer will replace a visit. If I'm choosing a contractor, checking terms or want to see the original figure, I still have a reason to open the source. The question is whether that source offers me anything beyond the summary I already got once I click through.

A mention, a citation and a visit are not the same thing

When talking about visibility, I'd like to separate at least three outcomes. A system can mention the company's name. It can cite the website as a source. Or a person can actually open the link and do something useful on the site.

To me these are not three names for the same metric. A mention without a link means something different from a direct visit. A citation is also not a recommendation of the product or the company. It may only mean that a particular page was used to answer a narrow question.

I can imagine a hypothetical case: a page with a good explanation of a procedure shows up among the sources, but the person only wants the basic information, not a contractor. Visibility like that can make sense, but it wouldn't be honest of me to automatically present it as a sales opportunity.

So I'd decide what interests me before I measure anything. Recognition? Visits to relevant content? Enquiries? Use of the product? If we don't separate these, we can quickly start celebrating numbers that tell us very little about the actual goal.

Individual systems don't play by the same rules

At Google, AI Overviews and AI Mode are part of Search, but they are not the same as the standalone Gemini app. I'd keep that distinction when reviewing a website. The names are related; the rules for access and display are not automatically the same because of it.

Google's documentation on AI features says that for a page to appear as a supporting link in AI Overviews or AI Mode it needs to be indexed and eligible to be shown with a snippet. No special extra optimisation is required, and inclusion is not guaranteed. I also wouldn't arbitrarily separate the traffic from these features out of the overall picture: Google includes it in the web search performance report in Search Console.

OpenAI, for ChatGPT search, describes the role of the OAI-SearchBot crawler and the need not to block it if we want our content to be eligible for summaries and snippets. I'd check that together with actual access through the server and the site's protection layers. An entry in robots.txt doesn't help me if the request is then stopped by another layer.

Perplexity describes PerplexityBot as a crawler for surfacing and linking websites in its results. It separates it from collecting data to train foundation models. So permission to visit a site is not something I'd equate, without thinking, with every possible use of the content.

With Gemini, sources and related links may be shown, but they are not part of every answer. A link shown when a claim is double-checked is also not necessarily the source the original answer came from. That alone is a good reason not to treat every green-highlighted sentence or list of links as proof that a site "ranks".

I'd check the technical foundation before the new acronyms

Before I did any special optimisation for an individual AI system, I'd want to know whether the site returns the right answers at all. Are the important pages accessible? Do links connect them sensibly? Does the URL stay stable? Does the important text exist in the rendered content as well?

Then I'd check the purpose of any restrictions. Some content is public, some isn't. I don't want to make drafts, private documents or user data more accessible just because we're talking about visibility. Those need proper protection, not optimisation for crawlers.

For public content, though, I'd look for unnecessary obstacles. A wrong noindex, a convoluted path to an important page or an unclear canonical can mean the system receives different signals from the ones I want to send. It isn't especially exciting work, but I'd rather start there than with a new add-on that comes with big promises.

I don't expect removing those obstacles to produce a mention in the next answer. Making something understandable and guaranteeing it gets selected are two different things. The first I can check and improve. The second I can't honestly promise. The half-hour checklist I run on my own site is in How to Check Whether Your Website Is Ready for AI Search.

It should be clear who is speaking and about what

The word "entity" quickly sounds more complicated than it needs to be for a website owner. When I hear it, I first think of recognisable things and the relationships between them: a person, a company, a service, a product, a place.

If the author is named in one place and appears only under a username in another, and the project page doesn't say who is behind it, the connections are harder to understand. The same goes for a term that means the company in one place, the product in another and a service in a third, without the difference being explained.

That doesn't mean I should repeat the same introductory paragraphs everywhere. It means sorting out the basic facts and the sensible links. Who wrote the article? What does that person do? Which project does the post refer to? Can the reader verify it?

That's why, on my own site, I keep the author page separate from the project pages. It's mostly a help to the visitor. Clear relationships seem to me a more sensible foundation than a list of keywords I hope someone will assemble into the right picture.

Structure should follow the question, not a template

Well-structured content, to me, isn't an article with as many subheadings as possible and a mandatory FAQ at the end. It's text in which I quickly understand the main claim, the conditions under which it holds, and the reasons for it.

If someone asks who the product is for, I'd answer that first. If they ask about limitations, I wouldn't bury them under a long introduction. If I compare two options, I'd say which criteria I'm comparing them on. With a procedure, I'd make sure the steps can actually be carried out in the order described.

That can be an ordinary paragraph, a list or a table. The format alone doesn't guarantee quality. Even a very tidy table can be empty if every cell says something generic.

I'd rather publish fewer claims that I can support. I'd add a source where I rely on documentation and mark my own view clearly. For a figure that can change, I'd check that it's still current. That makes it easier for the reader to judge, even if no AI system ever cites the page.

I can't add credibility with a single tag

I understand authority as something that has to be evident from the content and the work, not just written in a bio. If I claim to know something, it helps to show how I think about the limits and what I check in a solution.

That doesn't require revealing clients or confidential data. I can explain how I make decisions, cite a publicly verifiable source, or clearly separate an actual test from a hypothetical example. I don't need an invented story about a great success for the explanation to be useful.

That is exactly why I'm wary of advice that promises credibility through volume. More content can mean more useful answers, or it can just mean more repetition. If I don't understand a question well, ten versions of the same article don't help me much.

I find an author more convincing when they can also write down what they don't know yet. That is my criterion for trust, not a claim about a secret signal in the algorithm. I don't want a personal view of quality to become one more supposed ranking formula.

GEO, AEO, schema and llms.txt have their limits

We use the acronyms GEO and AEO for work related to generative answers and to answering questions. They can help name the field, but I wouldn't let them carry the main story. For a business owner it matters more to understand what will change on the site and how we'll check it.

Structured data can describe content more precisely. It cannot replace a service description that doesn't exist, or prove experience the author doesn't have. If the markup says one thing and the visible text another, I haven't tidied up the content; I've created a new inconsistency.

I take a similar view of llms.txt. It can be an additional description of what's on the site. I don't treat it as a universal standard for selecting sources, or as proof that the site is ready for every AI system. For its own AI features, Google explicitly says that no special AI text files or new schema are needed.

So I wouldn't buy the promise that one file or one tag will solve visibility. I'd want to know which concrete problem it solves, which system uses it, and how we'll separate its effect from other changes.

What I'd watch in practice

I'd look at several things together: search queries, relevant visits, behaviour on the site, and the actions that matter for the project. I'd follow AI mentions as an additional view, one where I have to know the limits of the sample.

One answer to one question is not a permanent position. If I checked how a site appears in answers, I'd note the question, the system, the date and the conditions of the check. I'd treat the result as an observation, not as proof that visibility as a whole has been solved.

It's precisely this wider view of search that seems like a good thing to me. It pushes me to understand a website as a clear presentation of knowledge, an offer and an identity, not just as a landing spot for a particular keyword. That still includes SEO. Only the goal becomes more useful: to be understandable and worth visiting, wherever the question started.

Photo of the author

I build web solutions and digital products. At GE-KO I work on web design, development and SEO; with Bizko I look for simpler ways through everyday business. I am interested in how technology saves people work, and in how AI is changing the web.

One step further.