Does review count still matter?

Yes, as a threshold rather than a race. Enough reviews to establish a pattern matters; going from 90 to 140 generic reviews matters much less than getting twenty that name specific services. Volume without specificity plateaus quickly in AI answers.

How recent do reviews need to be?

Recent enough to signal an operating business. A profile whose newest review is two years old reads as possibly closed. A steady trickle beats a burst — five reviews a month for a year is more useful than sixty in one week, which also looks manipulated.

Do negative reviews hurt AI visibility?

Less than owners fear, and an unanswered one hurts more than the review itself. A professional, specific response is text the assistant reads too, and it demonstrates how you handle problems. A perfect 5.0 with no negatives at all can read as less credible than a 4.7 with visible, well-handled complaints.

Should I ask customers what to write?

You can ask them to mention what you did — that is a prompt, not a script. Do not write reviews for customers, offer incentives for positive reviews, or filter who you ask based on expected sentiment. All three violate platform policies and are detectable.

Guides/Technical fixesupdated 2026-08-04 · 5 min

How AI Assistants Actually Read Your Reviews

The short answer

AI assistants read review text, not just the star average. They extract which services you are praised for, how recently, and what specific qualities recur. Twelve reviews naming a specific job build a stronger recommendation than eighty generic five-star reviews saying "great service."

Most businesses think about reviews as a number: rating and count. Get the rating up, get the count up.

AI assistants read them as text. That changes what you should be asking for.

What actually gets extracted

When an assistant is deciding whether to recommend you and what to say about you, it pulls four things out of your reviews.

1. Which services you are praised for. Reviews mentioning "water heater" repeatedly build an association between your business and water heater work. That association is what makes you the answer to "who replaces water heaters in Tampa" rather than only to "best plumber in Tampa."

2. Recurring qualities. Speed, cleanliness, honesty about pricing, explaining the problem, showing up when they said. These become the justification the assistant writes when it names you — "known for fast emergency response" comes from somewhere.

3. Recency. How current the pattern is. A business whose reviews stopped eighteen months ago reads as possibly closed.

4. How you handle problems. Your responses are text too. A specific, non-defensive response to a complaint demonstrates something a five-star review cannot.

Notice what is not on that list: the star average by itself. It matters as a floor. It is not the interesting signal.

The specificity gap, illustrated

Two plumbing companies.

Company A: 84 reviews, 4.9 average. Representative sample: "Great job!" "Very professional." "Highly recommend." "Fast and friendly." "Would use again."

Company B: 31 reviews, 4.7 average. Representative sample: "Replaced our 50-gallon water heater the same day we called and caught that the expansion tank was shot." "Found a slab leak two other companies missed." "Quoted $340 for the drain line, charged $340."

Company A has a better rating and nearly three times the reviews. Company B gets recommended more often in AI answers, because Company B's reviews contain extractable facts: same-day water heater replacement, slab leak detection, quoted price held.

Company A's reviews contain nothing an assistant can use to justify a recommendation.

How to get the second kind

You cannot script reviews. You can absolutely ask people to mention what happened.

The text, sent same-day while it is fresh:

Thanks again for having us out today. If you have a minute for a Google review, it genuinely helps — and if you mention what we actually did (the water heater swap), it helps other people with the same problem find us.

That is a legitimate prompt. It asks for specificity, not sentiment. It does not offer anything in exchange and does not filter by expected rating.

Timing: same day, while the detail is remembered. A week later you get "great service" because they have forgotten the specifics.

Channel: text works far better than email for local service work. Send the direct Google review link, not a homepage link.

Who to ask: everyone. Asking only the customers you expect to be happy is review gating, and it violates Google's policies.

What to ask them to mention

Different for each trade. What you want is the phrase a future customer will type.

Plumbing: the specific job and the response time
HVAC: the equipment, the season, whether you fixed rather than replaced
Roofing: insurance handling and cleanup
Electrical: what you diagnosed, especially if others missed it
Cleaning: consistency of the same team, and move-out deposit outcomes
Auto repair: the vehicle and the problem
Movers: that the quoted price held and nothing was damaged
Locksmiths: that the phone quote was the final charge
Landscaping: the specific project and the problem solved

Each of these maps onto a real query. That is the point.

Where reviews need to be

Google first, by a wide margin — it feeds Gemini and Google AI Overviews directly through your Google Business Profile, and it is what most directory pages surface.

Then, roughly in order of usefulness for AI visibility: Yelp, Facebook, BBB, Angi, and your industry-specific platforms.

Do not neglect Google to spread thin across ten platforms. A strong Google profile plus two secondary ones beats a mediocre presence on eight.

Responding

Respond to everything. Two reasons: it is more text on your listing that assistants read, and it demonstrates how you handle problems.

On positive reviews: repeat the specific. "Glad we could get the water heater handled same-day, Jim." Now the service is named twice.

On negative reviews: answer specifically, without defensiveness, with what you did. "You are right that we were two hours past the window on Tuesday — our morning job ran long and we should have called you. We refunded the trip charge and I would like to make the rest right; my direct line is..."

That response does more good than the one-star does harm. It is also the single most-read text on your profile.

Review responses also feed your entity — they are dated, attributable text tied to your business.

Never: argue, mention private customer details, or dispute reviews reflexively. For a health or legal practice, responding at all requires care — discussing any patient or client detail publicly is a confidentiality problem regardless of who raised it first.

The 5.0 problem

A perfect 5.0 with a hundred reviews can read as less credible than a 4.7 with visible, well-handled complaints.

Assistants have absorbed a lot of coverage about review manipulation. An unbroken wall of five stars with generic text is the pattern that fake reviews produce. A realistic distribution with substantive text and thoughtful responses is the pattern real businesses produce.

Do not manufacture negatives. But do not panic about a 4.6 either — it is often a stronger signal than a 5.0.

What to avoid entirely

Buying reviews. Detected, penalized, and increasingly prosecuted.

Incentivizing positive reviews. Against Google's policies and FTC rules on endorsements. Asking everyone for a review is fine; offering a discount for a five-star one is not.

Review gating. Surveying customers first and only sending happy ones to Google is a policy violation.

Bursts. Forty reviews in a week after eleven months of silence is the pattern platforms flag. Steady is better.

What to do this week

Read your last twenty reviews. Count how many name a specific service.
Write a same-day review request text that asks people to mention the job.
Set up sending it to every customer, not selected ones.
Respond to every unanswered review, restating the specific service.
Answer your oldest unanswered negative review properly.

Then see the effect

Review specificity shows up in AI answers as better justifications and more mentions for specific services. Get a baseline now — our free scan runs twenty-five real customer questions across five assistants and shows you exactly how you get described, if you get named at all.

About fifteen seconds, free, no account.

Common questions

Does review count still matter?: Yes, as a threshold rather than a race. Enough reviews to establish a pattern matters; going from 90 to 140 generic reviews matters much less than getting twenty that name specific services. Volume without specificity plateaus quickly in AI answers.
How recent do reviews need to be?: Recent enough to signal an operating business. A profile whose newest review is two years old reads as possibly closed. A steady trickle beats a burst — five reviews a month for a year is more useful than sixty in one week, which also looks manipulated.
Do negative reviews hurt AI visibility?: Less than owners fear, and an unanswered one hurts more than the review itself. A professional, specific response is text the assistant reads too, and it demonstrates how you handle problems. A perfect 5.0 with no negatives at all can read as less credible than a 4.7 with visible, well-handled complaints.
Should I ask customers what to write?: You can ask them to mention what you did — that is a prompt, not a script. Do not write reviews for customers, offer incentives for positive reviews, or filter who you ask based on expected sentiment. All three violate platform policies and are detectable.

Stop guessing

See exactly what the five AIs say about your business.

Twenty-five real customer questions across Claude, ChatGPT, Gemini, Perplexity, and Google AI Overviews. Free, about fifteen seconds, no account.

Run the free scan