AI Overviews or AI Assistants: What to Measure First

You have the budget and the attention to measure one thing properly. Not two. So the question arrives in its ugliest form: do you track whether your brand shows up in the AI summary above the search results, or whether it shows up when someone asks an assistant a question in a chat window?

There is no universal answer, and anyone who hands you one is guessing on your behalf. What there is instead is a decision you can make deliberately, based on where your buyer actually is when they ask the question that decides the deal.

Two surfaces, two different people

Strip away the branding and the difference is about intent, not technology.

The AI summary in a search engine intercepts someone who was already searching. They typed a query into a search box out of habit, and a machine wrote them a paragraph before showing the links. Their intent was to search. The summary happened to them on the way.

An assistant intercepts someone who has given up on the list of links entirely. They opened a chat, described their situation, and expected an answer, not a set of candidates to evaluate. Nobody types a paragraph into a chat window hoping for a page of blue links. They typed it because they wanted the machine to do the shortlisting.

Two people, two different points of a decision. That both interactions produce text written by a model is a coincidence of implementation, not a reason to treat them as one channel.

The question itself tells you where your buyer is

Short questions with a factual answer live in search. What is a heat pump. Melting point of stainless steel. Best running shoes. Someone would have typed these anyway, and the summary is answering something with a fairly stable, fairly public answer. Brands appear there roughly the way they always have in search: as a byproduct of pages that rank and say things plainly.

Long questions with constraints live in the assistant. We are a manufacturer with a few dozen employees, we export to Germany and Poland, we need a supplier who can handle small batches and certify origin, who should we be talking to. Nobody types that into a search box, because a search box has never been able to do anything useful with it. It goes into a chat, and what comes back is a shortlist of names with reasons attached.

So read your own sales calls. Write down the question a prospect asked in the first meeting, in their words. If it fits on one line and has a single right answer, you are in search territory. If it carries a situation, a constraint and an implied budget, it has moved into the assistant.

Where the deal is actually decided

A search summary tends to catch the early, informational moment: the person is still learning what the category is. Being named there is valuable the way awareness is valuable. Broad, cheap per impression, and rarely the last thing that happens before someone spends money.

An assistant conversation, when it is long and specific, tends to catch the moment the shortlist gets written. The person has stopped learning and started choosing. Being one of the handful of names in that answer, or not being one of them, is a commercial event with a short path to a meeting.

The honest version of the question, then, is not "which surface is bigger". It is: which surface is where your deal gets decided? A consumer brand selling something people buy without much thought may find broad presence in search summaries is the whole game. A company selling something complex and comparison-driven usually finds the assistant conversation is where it wins or vanishes, even if far fewer people are having it.

How verifiable each one is from outside

An assistant is straightforward to interrogate. Open a fresh session, ask your question, read the answer. Do it again, in another language, on another platform. The answer varies between runs because the models are probabilistic, which is why a single check is worthless and real measurement means many runs over time. But what you are measuring is right there, and it is what your buyer sees.

A search summary is harder. What appears above the results depends on the exact query, the country, the device, whether the engine generated a summary at all for that phrasing, and on signals attached to the person searching. Two colleagues in the same room can get different summaries, or one gets none. Measuring it from outside means reconstructing a context you neither control nor fully observe. That does not make it worthless. It makes it a different kind of measurement, carrying a different confidence, and you should know which kind you are buying.

A good result in one predicts nothing about the other

The tempting shortcut is to treat the two as proxies: measure the convenient one, infer the rest. It does not hold, and the reason is mechanical.

A search summary is assembled largely from pages the engine already ranks for that query. It leans on the same web your search work has been shaping for years. If you rank, you have a reasonable chance of being quoted.

An assistant answering a long, constrained question is doing something else. It draws on what it absorbed about your category during training and, sometimes, on pages it retrieves live, then compresses all of it into a recommendation. Whether it names you depends on how the wider web describes you: documentation, forums, review sites, comparison articles written by strangers, discussions in the language your buyer speaks. A brand can rank beautifully and still be absent from every assistant answer in its category. Ranking for a keyword and being a credible option for a situation are different achievements.

The corollary is uncomfortable. If you measure one surface, you know about one surface.

So choose, and choose on purpose

Start with the assistants if your buyers ask long, comparative, constrained questions; if your sale runs through a shortlist; if you sell across several languages and need to know which markets do not know you exist.

Start with the search summary if your category is dominated by short informational queries, if your buyers are consumers deciding quickly, or if your existing search traffic is the asset you are defending and you need to know whether the summary is absorbing the clicks it used to send you.

Then be explicit that you have chosen. The failure mode is not picking the wrong surface. It is measuring one, reporting it upward as "our AI visibility", and letting everyone in the room assume it covers both.

What we cover, and what we do not

So you do not spend a trial finding out the hard way: PSentry measures the assistants. Four of them: ChatGPT, Claude, Gemini and Perplexity. It runs your prompt set across all of them, in each language and market you sell in, and reports where you were named, where you were cited, which sources the models leaned on, and which competitors got recommended in your place.

It does not measure AI Overviews. It does not measure AI Mode. It does not measure Copilot. If the summary above the search results is the surface your business lives on, we are the wrong instrument, and the useful thing we can do is say so before you pay us anything.

None of this comes with a guarantee that your name starts appearing more: what gets reported is a reading, not a result manufactured to look good. That reading arrives twice a month rather than continuously, because an assistant's opinion of a brand shifts on the same unhurried clock as the web it was trained on, not on demand. And nothing in the process leans on a model's wording either way; it is watched, never steered. What you get is an account of where you stand in the conversations your buyers are having with the machines, in every language you sell in.

Frequently Asked Questions

If I can only measure one, is the assistant always the right call?

No. It is right when your buyers ask long, situational questions and your sale runs through a shortlist. If your category is short informational queries and quick consumer decisions, the search summary is closer to your money, and no vendor should argue you out of that because it happens to be the thing they sell.

Are the two surfaces not converging anyway?

The interfaces are borrowing from each other. The intent behind the question is not converging, and intent is what you are measuring. Someone who opens a chat to describe their situation is doing a different thing from someone who types three words into a search box, whatever the boxes end up looking like.

How do I find out which surface my buyers actually use?

Ask them, in the next sales call rather than in a survey: how did you first come across us, and what did you type. The answers are unscientific and few, and still more informative than any market study, because they are your buyers rather than an average of everybody's.

Does being cited in a search summary help me appear in an assistant?

Indirectly at most. Both read the same web, so content that is crawlable, clear and widely referenced tends to help on both. But there is no mechanism by which appearing in one causes appearance in the other, and no evidence you can point at to claim there is.

Why does PSentry not cover search summaries as well?

Because measuring them properly means reconstructing a personalized, location-dependent, intermittently generated result, and doing that badly is worse than not doing it. We would rather cover four assistants in a way you can check yourself than publish another number nobody can verify.