Review Management When AI Reads Your Reviews

A buyer used to scroll your reviews and form an impression. A growing share now ask an assistant who to hire, and meet a two-sentence characterisation of your business assembled from those reviews without reading any of them. Your star average is an input to that sentence. It is no longer the thing the customer sees.

Two audiences now

The human reader skims the most recent few, checks whether the bad ones sound like a pattern or a bad day, and looks at how you replied. That reader still exists and still matters.

The second reader consumes everything at once and outputs a characterisation. Ask an assistant about plumbers in your city and the useful ones come back with something like consistently praised for punctuality, several mentions of pricing surprises. That is a different output from a 4.6 average, and it is built from the words rather than the stars.

Which changes what is worth collecting. A rating with no text is nearly invisible to the second reader. A short specific sentence about what you did is worth several of them.

What actually moves the summary

The last row is the one businesses underestimate. A summary drawn from several sources notices when one of them disagrees with the others, and a profile that looks flawless in isolation while three other platforms say something different reads as less trustworthy, not more.

Ask for the detail, not the stars

Most review requests say some version of please leave us a review, which produces ratings with no text. A request naming the thing you want described produces sentences.

Asking a customer what they found most useful about how the job went, or how the process compared to what they expected, gets a paragraph about punctuality, communication or tidiness. Those paragraphs are what the summary is made of.

Timing does most of the work. A request sent within a day of finishing, from the person who did the job rather than from a marketing address, converts far better than one sent a fortnight later by a system. Steady collection beats a campaign, because the spike-then-silence pattern reads as stale within months.

Replies are written for the third party

The reviewer has already formed their view. The audience for your reply is the person reading it in four months, and the system summarising it next week.

That points at a specific shape. Name what went wrong without restating the complaint at length, say what changed as a result, and offer a route to sort it out. A reply that says the crew arrived late because of a scheduling error, that you have since moved to confirming windows the night before, and that the customer should call a named person, gives both audiences something concrete.

Two habits to drop. Arguing, which makes a single bad experience look like a pattern of defensiveness. And the identical templated thank-you under thirty reviews, which a human reader discounts instantly and which contributes nothing distinguishable to a summary.

Drafting replies with an assistant is sensible, as long as a person supplies the facts. The output to avoid is the one that reads as generated, which is the problem we described in the workslop quality problem. Never have it invent a remedy you did not offer.

Measure what the assistants say

Star average is easy to track and no longer tells you what a buyer encounters. Add a second measurement that takes fifteen minutes a month.

Write down the five questions a customer asks before hiring in your category. Ask each one in ChatGPT, Gemini, Perplexity and Copilot on the first of the month. Record whether you are mentioned and, when you are, exactly how you are described. Keep it in a spreadsheet so you can see the description change over six months.

That log is the only view you get of the thing buyers actually meet. The rest of the work that shapes it is in optimising for AI search and making your site legible to AI agents, and the paid side of the same surface is in ChatGPT ads.

A month of work

Change the wording of your review request to ask what was most useful. Reply to every review from the last six months with something specific. Check the three platforms you have neglected and make sure the basics match. Then run the five prompts and write down what comes back.

Do that again in three months and compare the descriptions. If the words have moved toward what you actually do well, the collection habit is working, which is a better signal than watching a decimal point on an average.

Frequently Asked Questions