Reviews as an input to AI answers

Here is the operator version. Your reviews are the only body of text about your business written by people who are not you, in language customers actually use, describing things your website never mentions. Parking. The wait on a Friday. Whether the staff speak Spanish. Whether the vegetarian thing is real or a garnish.

That is not marketing copy. It is a description of your business by third parties, at volume, updated continuously. There is nothing else like it in your online presence, and systems that assemble answers about local businesses are pulling from a pool where it sits near the top.

How does a review actually reach an AI answer?

Through the front door, mostly. Google states that AI Overviews and AI Mode are built on core Search ranking, retrieving from the Search index. It also states that a Business Profile helps your services appear "in both AI responses and other Google Search results."

So the chain is not exotic. Reviews affect local ranking. Local ranking affects what the generative layer retrieves. The generative layer writes the answer.

The second path is the text itself. A retrieval system looking for a business that is good for large groups needs a source that says so, and often the only source that says so is a review. Your menu page does not mention group size. Fourteen customers did.

Whitespark's practitioner testing of AI Mode local answers in May 2026 lines up with that. Star ratings, review volumes, pricing and hours came through from Business Profiles or comparable sources, and the researchers' explicit finding was that reviews and unstructured citations, meaning blogs, news and community pages, are the heaviest inputs to those answers. Grade it honestly: that is vendor observation, not a controlled test. It points at the same asset Google names.

Nobody outside these companies can tell you how heavily review text is weighted, and anybody who states a percentage is inventing it. What you can say is that the text exists, it is indexed, it is specific, and specific beats generic in retrieval.

Does the number of reviews matter, or the timing?

Timing, more than most owners expect.

Whitespark's 2026 Local Search Ranking Factors report, in which 47 local search experts scored 187 factors, places Recency of Reviews at eleventh with a combined score of 164 and Sustained Influx of Reviews Over Time, rather than bursts, at fourteenth with 154. Quantity of native Google reviews with text sits at ninth with 170, and high numerical ratings at sixth with 181. Grade that source honestly too: it is a survey of expert opinion, and Darren Shaw, who runs it, says plainly that the participants do not have access to Google's algorithm and that correlation is not causation.

Shaw moved review recency from twentieth place in the 2023 edition into his personal top five, and put it bluntly: "The moment you stop getting new reviews, you're going to see your local rankings start to slip."

The corollary is the one people resist. Shaw again: "Getting a negative review is actually better than getting no new reviews at all. Review recency will increase your rankings regardless of whether the new review is positive or negative." Recency is the signal. Sentiment is a different conversation.

One thing you cannot do with those scores. They are survey points on a 0 to 5 scale summed across 47 respondents, not weights. Adding 164 and 154 and calling reviews some percentage of the algorithm produces a number that does not exist. Whitespark's own group weightings are disputed between secondary summaries, so cite the ranks and the scores, never a share.

Practically, a business with four hundred reviews and nothing new in eight months is in worse shape than a competitor with sixty and two a month. That competitor also has fresher text describing what the business is like now, which is what an answer engine ends up summarizing. Shaw's benchmarking rule is the simplest version: find out how often your top competitors get new reviews, then match it plus one. Review cadence is the whole mechanism, and it does not need an AI framing to be worth doing.

Should you ask customers to mention specific services?

No, and there is a controlled test on this that almost nobody in the AI search category cites.

Sterling Sky ran it. Joy Hawkins picked a Christmas tree farm with, in her words, "no SEO efforts, no map pack SEO, and their website was terrible," and ran the test in the middle of summer specifically because the business was not going to receive organic reviews that would muddy the result. Six people left reviews over several months containing terms like "Christmas trees" and "fresh trees," against a base of 18 total reviews.

The result, verbatim: "adding keywords to reviews did not help this business' local pack ranking on Maps... Actually, in the first instance, rankings got worse." A second independent test using an invented word found the same thing. Her recommendation is one sentence: "Don't coach your clients to include certain words in their reviews."

The confusion that keeps this myth alive is justifications, the little snippets Google shows under a listing quoting a matching review. Hawkins resolves it: "You have to rank first, then if they find a review referencing the service, they pull it in." The justification is a consequence of ranking, not a cause of it.

There is a second reason to leave the wording alone. Google's Rating Manipulation policy states that when soliciting reviews, merchants "should not... request that specific content be included," and names both staff review quotas and requests that reviews identify a staff member as prohibited. Every "ask them to mention the hygienist by name" script is on the wrong side of that sentence.

Say the gating rule precisely, because most people get it wrong. Routing happy customers to a public review link while diverting unhappy ones to a private form is explicitly banned by Google, in the clause prohibiting merchants from discouraging negative reviews or selectively soliciting positive ones. The FTC's 2024 rule on consumer reviews does not name gating by that word. It reaches the adjacent conduct: incentives conditioned on sentiment, and misrepresenting a filtered set of testimonials as representative. Both matter. Saying "the FTC bans gating" is the kind of imprecision that gets corrected in front of a client. Gating is prohibited and it also makes you rank worse, which is the version worth remembering.

The rule's own safe harbour tells you the compliant shape: reviews resulting from generalized solicitations to purchasers are outside the prohibitions. Ask everyone. Do not condition on sentiment. Do not script content.

What is a review keyword program actually worth?

Zero, and you can show the arithmetic.

The quote. An agency offers "review optimization for AI visibility" at $450 a month. The deliverable is a request script asking customers to name the service and the city. Year one: $5,400.

What it can be shown to do. One controlled test, six coached reviews, no ranking improvement, and worse rankings in the first instance. The demonstrated effect is zero or negative, and the tactic sits against a written platform policy that carries review posting restrictions and public consumer alert banners as enforcement.

So price the line item at zero. Not because it is unproven. Because it was tested and it did not work.

Then reprice the money. Suppose your current flow is one new review a month, and a request habit tied to the moment of service gets you six. That is 12 reviews a year against 72 derived. The same $5,400 across 72 reviews is $75 per review (derived), against $450 per review (derived) at your current rate. Substitute your real numbers: count the reviews you received in the last sixty days, double it for an annual estimate, and divide the quote by both.

The comparison that decides it. One of these purchases buys a tested-and-failed tactic. The other buys the recency and sustained-influx signals that a survey of 47 practitioners ranks in the top fifteen factors, and the fresh third-party text that an answer engine summarizes. Same money.

What about your responses to reviews?

They are indexed text too, written by you, attached to the most-read page about your business.

A response that says "thanks for the kind words" adds nothing, and Google's own review guidance agrees: instead of sending the same thank you to everyone, it tells businesses to focus on reviews where they can share a helpful update or answer a question. A response that says "glad the gluten-free menu worked out, we keep a separate fryer for it" adds a fact about your operation to a page that gets crawled. That is not the main reason to write good responses, since a review response is customer service in public first, but it is a real secondary benefit.

This connects to how assistants decompose questions. A request like "somewhere in Kaimuki with high chairs that is not loud" gets broken into parts, and query fan-out means each part gets searched separately. Reviews are frequently the only source answering the parts.

Worth noting what you do not control here. On your own site you decide crawler access, and whether to block AI crawlers is a decision worth making deliberately. On review platforms, the platform decides for you. Your reviews are being read regardless of what your robots.txt says.

What to do this week

1. Count how many reviews you received in the last sixty days. If the answer is zero, that is the entire finding. 2. Check your top three competitors' cadence over the same window and set your target at theirs plus one. 3. Set an asking cadence tied to a moment in the service, not a monthly campaign. Bursts produce a spike and then a flat line, and unusual volume patterns are an enumerated policy violation. 4. Stop any script that suggests wording, names a staff member, or sets a staff quota. It does not work and all three are named in Google's policy. 5. Answer the last ten reviews with one concrete operational fact in each response. 6. Read your last thirty reviews as a stranger would. The themes you see are roughly the themes an AI answer will repeat about you.

If you have moved locations recently, check whether older reviews and older citations still assert the old address, because moving without losing local rankings depends on that cleanup.

Want to know what your reviews say about you?

Send your Google review link to eric@seod.com and I will read your most recent reviews and send back the three themes an AI system would most likely repeat about your business, good and bad, plus your review cadence next to your two closest competitors'. No charge, no pitch, and it is often uncomfortable reading.

I offer this because owners rarely read their own reviews in bulk, and the summary is usually a sharper description of the business than anything on the website.

Otherwise, keep going through the AI search library.