Gemini Is Not Google AI Mode, and Neither Is AI Overviews

Ask a marketing team whether the brand shows up "on Google's AI" and you will get an answer, delivered with confidence, that means almost nothing. Not because the team is careless. Because the question has no single referent. Google ships three different things with a generative layer inside, and they get talked about as if they were one product with three names.

They take different input. They serve different intent. They choose their sources by different mechanics. A brand can be present in one and entirely missing from the other two, and that is not a contradiction to be explained away. It is the normal outcome.

The confusion has a cost: it makes companies buy the wrong tool. So here is the separation, followed by the uncomfortable part about what our own product cannot see.

Three things wearing the same logo

Gemini, the assistant

Gemini is a chatbot. You go to it on purpose, open a conversation, type a question the way you would ask a colleague, and it writes prose back. No results page, no ranked list of links under the answer competing for the click. The user came for an answer and gets one, and either your name is inside it or it is not.

Structurally this is the same surface as ChatGPT, Claude or Perplexity: a conversation window where the text is the whole product. Hold onto that, because it is why a tool can treat those four as one family.

AI Overviews, the summary nobody asked for

AI Overviews is a block of generated text at the top of an ordinary Google results page. The defining feature, the one everybody skips past, is that the user did not request it. They typed a query into a search box, as they have for years, and Google decided this query deserved a synthesised paragraph above the results.

The classic list of blue links is still there underneath, and still gets clicked. The Overview coexists with that list rather than replacing it, which makes this a search page with a summary bolted on top, not a conversation. The pages feeding the summary tend to be pages already in the running for that query.

AI Mode, search that talks back

AI Mode is the middle case, which is why it causes the most confusion. It lives inside Google and you reach it from search, but it drops the results page and answers conversationally, with follow-ups that stay in context. The user opted in: they chose the mode, and arrived expecting a dialogue rather than a page.

What sits underneath is still search machinery. The question gets fanned out into related queries against Google's index, and the answer is composed from what comes back. Its instincts about which sources to trust are inherited from search, not from a chat model recalling what it absorbed about you.

Why the distinction is not pedantry

The input is different

Into a search box, people type fragments: keywords stripped of grammar, shaped by years of learning what search engines respond to. Into an assistant, the same people type paragraphs. They explain their company, list their constraints, ask for a recommendation, then argue with the one they get. Same buyer, same afternoon, entirely different text depending on which box is in front of them.

The intent is different

A search query usually wants a destination: the user expects to land somewhere and read. An assistant conversation usually wants a decision: the user expects to be told, and to push back. AI Mode sits between the two, which is what makes it new rather than a rebrand. Those are different moments in a buying process, and being named in one is worth something different from being named in another.

The source selection is different

An assistant answering from what it absorbed during training can describe your product at length and link to nothing, because there is nothing to link to: it is not reading a page, it is producing text from what it learned. A summary above a results page is assembled by the apparatus of ranked search, and leans on the pages that were already going to appear below it. Two selection mechanics, two different sets of names in the output.

Therefore the measurement is different

You cannot infer one from the others. A brand a chat assistant recommends warmly, because the wider web discusses it in the kind of text models absorb, can be absent from an Overview whose sources are pages ranking for a commercial keyword it never targeted. The reverse is just as easy: a site with strong ranked pages can appear in Overviews and never once be named in a conversation.

A blended "Google AI visibility" number that quietly averages the surfaces together is therefore worse than no number: when it moves, you cannot tell which surface moved.

What the question should have been

"Is my brand visible on Google AI" stays unanswerable until you say which of the three you mean. Rewrite it and it becomes tractable:

• Does the Gemini assistant name us when someone describes their problem and asks for a recommendation? A question about conversational visibility, sitting beside the same question asked of ChatGPT, Claude and Perplexity.
• Do we appear in the Overview above the results for queries we already care about? A question about search, adjacent to your rankings, answered with search tooling.
• Do we survive when a searcher switches into AI Mode instead of scanning the links? A question about a search surface behaving conversationally, and its own thing entirely.

Three questions, three answers, no obligation to agree.

The part we would rather say ourselves

PSentry monitors four assistants: ChatGPT, Claude, Gemini and Perplexity. Gemini there means the assistant, the chatbot, the conversation window described at the top of this page. That is the surface we run your prompts against, in each language and market you sell in.

We do not monitor AI Overviews. We do not monitor AI Mode. We do not monitor Copilot. If the thing keeping you awake is whether your pages feed the summary above the blue links, we are not the tool, and no amount of enthusiasm on our part will make us the tool. Buy something built for search surfaces, and learn that here rather than after paying us.

The boundary is about coherence, not modesty. The four assistants we cover are the same kind of object: a user in a conversation, asking for a recommendation, receiving prose. One prompt set across those four produces a comparison that means something. Bolting a search surface onto it would produce a score made of two incompatible measurements, the exact mistake this article is about.

How to decide which surface you actually need

Start from your buyer, not a vendor's feature grid. Picture the moment before they know you exist, and ask where they are standing when they ask the question that leads to you.

If they type a short commercial query and skim a page of results, your fight is on the search page, where the Overview is a new competitor for attention above your listing. That is a search problem, and it wants search tools.

If they describe their situation in a paragraph to an assistant and ask what to buy, your fight is inclusion in a sentence with room for a few names. That is where we operate, and it is a different problem: there is no position to climb toward. You are in the answer or you are not.

Most companies have buyers in both places, in proportions that shift by market. So the advice is unglamorous: decide which moment matters more to your revenue, buy for that one first, and refuse any single number claiming to cover both. Then ask every vendor, us included, one blunt question: which exact surfaces do you query, and can you show me a stored response from each. A tool that cannot separate a chat assistant from a search summary while answering that will not separate them in your dashboard either.

Frequently Asked Questions

Do Gemini the assistant and AI Mode use the same model?

They can share a model family without being the same product, and that is the trap. The model is one ingredient. What decides whether you get named is the surrounding system: what it retrieves, from where, how it weighs a search index against learned knowledge, and what the user typed. Same engine, different vehicle.

If I show up in AI Overviews, will I show up in Gemini?

Sometimes, and you cannot count on it. The two draw on different material and select it differently. Treat any correlation you observe as a coincidence worth watching, not a rule worth planning around.

Which of the three should I worry about first?

Whichever one your buyers actually use, which is a question about your market rather than about the technology. A team selling to developers and a team selling to procurement officers meet these surfaces in very different mixes.

Why does PSentry cover Gemini but not the Google search surfaces?

Because the four platforms we cover are the same kind of surface and can be measured with one consistent method. Adding a search summary would mean mixing two kinds of evidence and reporting the mixture as one score.

Does any of this change how often I should measure?

No. These systems are probabilistic: ask the same thing twice and the list of names can change. One check on any surface is an anecdote. What you want is the same prompt set, repeated over time, on a surface you have actually named.