← All articles
AEO · Reputation

Reviews are read, not counted: how AI answers use your reputation

13 October 2026·7 min read
Quick answer
  • AI assistants do not report your star rating. They summarise what people wrote.
  • Recurring specifics become the attributes attached to your name. Recurring complaints become the "however".
  • Replies, recency and platform spread are all inputs, and all within your control.
  • Nothing here requires buying anything, and buying reviews is the one way to make it worse.

Ask an assistant whether a particular nursery in Jumeirah Village Circle is any good, or what people say about a corporate training provider in Business Bay, and you will not get a number. You get a paragraph: "Parents frequently praise the small group sizes and the bilingual staff, though several recent reviews mention limited parking." Every clause in that sentence came from somewhere, and almost none of it came from the business's own website.

That paragraph is the review layer of answer engine optimisation, and most Dubai businesses are managing it with a tactic designed for a different era: get the average above 4.5 and keep the count climbing. The average still matters for the map pack. For the AI answer, it is close to irrelevant.

Star ratings were built for humans. Review text is what machines use.

A rating compresses hundreds of opinions into one figure so a person can scan a list. A language model does not need the compression. It reads the reviews themselves, looks for themes that recur, and writes a summary with the confidence those themes earn.

This changes what a "good" review profile looks like. Eighty reviews that say "great place, highly recommend" give the model a positive sentiment and nothing else to say about you, which is why such businesses get described vaguely or left out of "best for X" answers altogether. Forty reviews that mention the settling-in process, the Arabic-speaking teacher, the WhatsApp updates and the parking give it four attributes it can attach to your name and repeat. Specificity is what gets you named for a particular need rather than listed as one more option.

The four things an assistant pulls from your reviews

  • Attributes, in the customer's words. Not the phrases on your homepage, but what reviewers independently keep mentioning. If three parents write about outdoor space, you have an outdoor-space attribute whether or not you ever marketed it.
  • Caveats. Assistants are hedged by design and like to include a "however". The recurring negative, even a mild one, is the most likely candidate. One complaint about response times disappears into the noise. Five of them become a sentence in every answer about you.
  • Recency. Models distinguish between what older reviews said and what recent ones say. A profile whose last review is eighteen months old gets described in the past tense, if at all, and a cluster of recent reviews about a change in management will be summarised as exactly that.
  • How you respond. Owner replies are public text and are read in the same pass. A factual reply that says the parking arrangement changed in March and there are now twelve bays can be lifted as a correction. A defensive reply becomes part of the story, and usually the least flattering part.

Which platforms count in the UAE

Google reviews come first, for the same reason the Google Business Profile leads every other off-site list: it is the most structured, most authoritative public description of a local business. But the assistant is not limited to Google, and the second-tier sources vary by sector in ways that catch businesses out.

  • Schools and nurseries: parent reviews on the school-comparison sites, parent forums and community groups that are publicly indexed, and the regulator's published inspection reports, which read to a model very much like a long, authoritative review. If the inspection report and your parent reviews disagree, the inspection report tends to win.
  • Training centres: the course marketplaces and listing sites your sector uses, LinkedIn recommendations for your trainers, and any certification body that lists partner feedback. Buyers asking "is this PMP course worth it" are getting answers assembled largely from these.
  • B2B services: Google, LinkedIn, the B2B review platforms if you are on them, and, more often than owners expect, employee review sites. A buyer's assistant asked about a consultancy will happily mention that staff reviews describe high turnover. Reputation does not stay in the lane you put it in.

As with directory listings, the aim is not to be everywhere. It is to make sure the two or three platforms your buyers actually read are current and tell the same story.

Shaping the themes without gaming them

You cannot write your own reviews and you should not try. You can influence what gets written, honestly, in five ways.

  • Ask for specifics, not stars. "Would you mind leaving a review, and mentioning what you found most useful?" produces text an assistant can use. "Please give us five stars" produces a number it will ignore. One genuine ask a week, as the weekly routine suggests, is plenty.
  • Ask at the moment the specific thing happened. For a training centre, that is when the certificate result lands. For a nursery, it is after the settling-in period, not on the first day, and with the child-privacy care covered in the nursery profile checklist.
  • Reply to everything, factually, within the week. Positive reviews get a short reply that names the thing they praised, which reinforces the attribute. Complaints get a reply that states what is true and what has changed, and nothing about how the reviewer is mistaken.
  • Fix the recurring complaint rather than burying it. You cannot out-volume a pattern. As long as new reviews keep mentioning the same problem, the assistant will keep summarising it. Treat the caveat in the AI answer as the top of your operations to-do list, because that is exactly what it is.
  • Say the same things on your own site. A page that plainly states group sizes, languages spoken, parking, response times and whatever else reviewers keep raising lets the assistant corroborate the reviews against the source. Reviews and website agreeing is what turns a theme into a confident claim. This is a large part of the on-site work in our education engagements.

Two things to avoid, because the major platforms prohibit both and the penalties fall on your profile: offering anything in exchange for a review, and review gating, where only customers you expect to be happy are invited to post publicly while everyone else is steered to a private form. Ask everyone, at the right moment, and accept the occasional three stars as the price of a profile that reads as real.

What happens to negative reviews

Less than owners fear, provided the rest of the profile is specific and recent. A single angry review in a profile full of detailed positives barely registers in a summary. A pattern does. And a burst of generic five-star reviews that appears in the week after a bad one is the worst response available: platforms discount it, and a model reading the timeline tends to describe it for what it looks like.

Reviews that are genuinely false, or that name a child or a staff member, have removal routes on every major platform, and those are worth using. For the case where an assistant is repeating something about you that is simply untrue, the repair sequence in when AI gets your business wrong applies, and reviews are one of the first sources to check.

What to expect, honestly

Nobody can promise that an assistant will call you the best nursery in your area, and an agency that promises it is selling you something else. What steady reputation work does is change the description: from vague to specific, from "a nursery in JVC" to "known for small groups and bilingual staff", and from a summary with a "however" to one without. That movement happens before any change in how often you are named, and it is visible within a few months rather than a few weeks.

The place to start is to see the current summary. Our free AI visibility check shows what assistants say about your business today and which sources, reviews included, they are drawing on. If the answer surprises you, the wider picture of how assistants decide which business to recommend explains where reviews sit among the other signals.

Common questions

Do AI assistants actually read Google reviews?

Yes, along with reviews and recommendations on other public platforms. Asked what a business is like, an assistant summarises the recurring themes in the text, complaints included, rather than reporting an average. Specific, recent reviews give it more to say than a high score made of one-word praise.

Should I reply to every review, including the positive ones?

Yes. Replies are public text and are read alongside the reviews. A factual reply to a complaint can be picked up as a correction; a defensive one becomes part of the story. Short replies to positive reviews reinforce the attributes you want attached to you.

Is it allowed to ask customers for reviews in the UAE?

Asking is fine. What the platforms prohibit is offering incentives and review gating. Ask everyone at a natural moment, invite them to mention what specifically they found useful, and never buy or trade reviews.

Find out how AI summarises your reputation today

Get a free AI visibility check →