Google now writes the first thing people read about you, and you cannot edit it

Every other element of your Business Profile is something you either wrote or can contest. The name, the categories, the description, the photos, the hours. Even the reviews have a flag button.

The AI-generated summary that now sits at the top of the profile is the exception. It is assembled from your review corpus, it updates on its own, and there is no field for it in your dashboard. For a lot of customers it is now the first sentence they read about your business, and it is the only one you have no pen for.

So the job changes. You cannot write the summary. You can only change what it is summarising, and the mechanics of how summaries get built mean that is a different task from the one most owners think they are doing.

Frequency beats severity, which inverts your triage

This is the part worth understanding properly, because it runs against every instinct you have about which reviews matter.

When you read your reviews, you triage by severity. The review alleging a safety problem gets your afternoon. The review complaining about parking gets an eye-roll.

A summariser does the opposite. It is looking for what is representative across a body of text, which means a theme appearing in fifteen reviews carries more weight than a single catastrophe, however bad that one was. One furious review about a botched job is an outlier and outliers are what summarisers discard. Fifteen mild mentions of a wait time are a pattern, and patterns are exactly what they are built to surface.

Which produces a genuinely uncomfortable conclusion: the complaint most likely to end up in the sentence customers read first is probably one you have been ignoring for two years because it never seemed serious.

Go and count rather than remember. Open your last sixty reviews, and tally every recurring noun: parking, phone, wait, price, cleanliness, the booking system. Not the worst thing anyone said. The thing the most people said. That tally is a better predictor of your summary than any individual review on the page.

Larkspur Veterinary found this the hard way. Their one-stars were about two genuinely bad outcomes, years apart, and the practice owner had spent a lot of emotional energy on both. Meanwhile nineteen otherwise positive reviews across eighteen months mentioned, in passing, that the phone goes unanswered at lunchtime. Nobody had ever treated that as a reputation issue. It was in nineteen reviews. The two bad outcomes were in two.

Fixing the phone cost them one scheduling change. It is the single highest-leverage reputation decision that practice made that year, and it did not involve responding to anything.

What this does not mean

Two bits of advice I would push back on hard, because both are already circulating.

You cannot keyword-stuff a summary. The idea that you should coach customers to use particular phrases so the summary picks them up fails on two counts. It is review gating adjacent at best and a policy violation at worst, which we covered in asking for reviews without violating policy. And it will not survive: a summariser reading twelve reviews that all use the same unusual phrase has been handed the signature of a coordinated campaign, which is the opposite of the outcome you wanted.

Volume is not the lever here either. Adding reviews changes your rating arithmetic, which is real and worth working through on its own. It does much less to a summary, because a summary is about proportions rather than counts. Thirty more reviews that mention the same lunchtime phone problem do not dilute it. They confirm it.

The open question that matters most

Here is what I genuinely do not know, and I would rather say so than guess: whether these summaries read your replies as part of the corpus, or only the reviews.

It matters enormously. If replies are in scope, then a consistent, specific response explaining that the lunchtime phone gap was fixed in March is editorial input into the summary, and responding becomes the one lever you have. If they are out of scope, replies do nothing for the summary and their value stays where it always was, with the human reading the page.

Google has not, as far as I can find, published an answer. Anyone telling you confidently either way is reasoning from vibes. What I would say is that writing replies as though they are being read by a machine costs you nothing, because the discipline it imposes, specific, factual, naming the thing that changed, is the same discipline that makes a reply work on a human. That is a rare case where hedging your bet has no price.

What that looks like in practice: The lunchtime phone gap was real and we fixed it in March by moving the second receptionist shift to cover 12 to 2. Specific, dated, falsifiable, names the mechanism. Compare it to We are sorry you had trouble reaching us and we are always working to improve. The second says nothing a machine or a person could use. If you want structures that default to the first shape, there is a set of templates for the positive reviews where these passing complaints usually hide.

Where the summary fits in the bigger shift

The summary on your profile is the visible edge of something larger, which is that machine-written synopses of your business are starting to appear in places you have no account with. AI Overviews were reported in October 2025 to have begun citing individual Business Profile reviews directly, which means review text is being quoted into answers well outside Maps.

We go into that properly in the piece on what the evidence actually says about review responses and local search, and the honest summary is that the mechanisms here are moving faster than anyone’s ability to measure them. What is not in doubt is the direction: more of what people read about you is being generated from your review corpus rather than from your marketing.

The practical version, in three lines

Count your recurring complaints instead of ranking your worst ones. Fix the cheapest frequent one, because frequency is what gets summarised. Write replies specific enough to be useful whether or not a machine is reading them.

The thing you cannot edit is downstream of the thing you can. That is the whole game now.

Sourcing note, and it is a real caveat. The existence and placement of these summaries is widely reported, but I could not confirm the details, the surfaces they appear on, or what is in the corpus against Google’s own documentation, and the usage statistics circulating about them disagree with each other badly enough that none are quoted here. The reasoning about frequency over severity follows from how summarisation works rather than from any statement by Google. Treat the mechanics as subject to change and the strategy as the durable part.