AI search optimization — what it is and how it differs from SEO

Not long ago a customer typed “plastic surgeon Krakow” into Google and scrolled through a list. Today they increasingly ask ChatGPT: “which clinic do you recommend and how much does it cost?” — or read the AI summary above Google’s results without clicking any of them. Hence the new line in agency offers: AI search optimization, also known as generative engine optimization (GEO) or answer engine optimization (AEO). Here is what it actually means, how it differs from classic SEO, and what can realistically be done — and what nobody can honestly promise.

AI assistants have no ranking

That is the most important difference. Google shows a list of ten results and you compete for a place on it. ChatGPT, Gemini and Perplexity show no list — they assemble one answer from several sources they find or remember. There is no “third position in ChatGPT”. There is either presence in the answer or its absence, plus a list of companies the assistant named instead of you.

So the word “optimization” is loose here. It is more honest to talk about AI visibility: whether you are among the cited sources, whether the assistant knows your facts, and whether it recommends you for the questions you want to answer. GEO, AEO and “AI SEO” all describe the same work.

Where AI gets its answers

Modern assistants combine two things: knowledge built into the model and live web search at answer time. When you ask about a specific company, price or city, the assistant usually reaches for the web — and that is where your influence begins.

• Google AI Overviews and AI Mode build a summary from pages in Google’s index. Cited sources are shown below the answer. If your website does not answer the question directly, a competitor’s page that does will be cited.
• ChatGPT with web search also cites page addresses. In our measurements for service businesses the sources are often not only company websites but also industry directories, review sites and “best of…” articles — intermediaries.
• Google Business Profile, reviews and external profiles matter to assistants as much as the website. If your website says 25 years in business and a directory says 30, the assistant will pick one version — not necessarily the true one.

The practical conclusion: AI assembles the picture of your company from many places at once. Working on AI visibility means putting all of those places in order, not just the website.

Three layers of work

When we do this for clients we split the topic into three layers — and it is worth thinking about it the same way on your own:

1. Google: AI Overviews and AI Mode. For every important query: does an AI summary appear, whom does it cite, are you among the sources. This is the layer closest to classic SEO — indexing, content quality and clear answers still count.
2. Assistants: ChatGPT, Gemini, Perplexity. A set of questions written the way a person asks them: “which company…”, “how much does…”, “can you recommend someone in…”. Asked repeatedly, because answers change.
3. The fact base. Website, Google Business Profile, directories, reviews, structured data, the llms.txt file. This is where we most often find mismatches — and where they are easiest to fix.

How to measure without fooling yourself

The cardinal sin of “AI visibility” reports is drawing conclusions from a single query. Assistants’ answers change from day to day; the same company can be recommended in the morning and skipped in the afternoon. A few rules we follow — and that you should demand from anyone offering such measurement:

• A fixed question set. Fifteen to twenty-five questions, written once and asked the same way in every measurement. Only that gives an honest “before” and “after”.
• Repeated runs. Every question at least three times per engine. We report the result as “x out of N”, for example “the company appeared in 4 out of 9 answers”.
• No forced names. “Can you recommend someone in Krakow?” — not “list five specific companies”. If the assistant recommends nobody, that is a result too.
• Cited sources. Not only “are we there” but “how does AI know” — because that tells you where to work.
• No numbers without a source. If someone writes “X thousand questions a week in ChatGPT about your industry”, ask where it comes from. Usually they do not know.

What you can do yourself — five steps

1. Answer on your website the questions people ask AI. Starting price, scope, delivery time, location, “who it is for”. If a customer asks “how much does…” and your site only says “individual quote”, the competitor with a price list will be cited.
2. Align the facts. Name, address, phone, years in business, ratings, opening hours — the same on the website, in Google Business Profile and in every directory you are listed in. Mismatches are the most common finding in our measurements.
3. Take care of Google Business Profile and reviews. For local questions it is often a more important source than the website.
4. Add structured data and llms.txt. Structured data (organization, services, FAQ) helps machines understand the site; llms.txt is a simple file describing the company for language models — cheap and harmless, though no substitute for the points above.
5. Clean up the index. Theme demo pages, old language versions, duplicates — for AI that is noise blurring the picture of the company. At one law firm we measured, more than a hundred such pages were still in the index.

What cannot be promised

Nobody controls assistants’ answers — not us, not any agency. Anyone promising “guaranteed presence in ChatGPT” is selling something that does not exist. Two things can be promised: measurement that can be repeated and work on everything AI builds the picture of your company from. We usually see the first differences in the “after” measurement after four to eight weeks — which is why we treat this as continuous work, not a one-off action.

Where to start

By checking where you stand. We run a free AI visibility audit: a handful of questions your customers ask, put to ChatGPT three times each, plus Google AI Overviews for your key queries and a list of whom AI recommends instead of you. Within 3 business days you get a PDF with results and priorities — no website access, no obligations. Only on that basis does a conversation about next steps make sense.