Do keywords in reviews affect rankings? A controlled test says no

This is one of the most durable pieces of local SEO folklore. It sounds plausible. Google reads the review text, the review text mentions "emergency plumber San Jose," therefore the profile ranks better for emergency plumber San Jose. Clean logic, no evidence behind it.

The reason it matters is that acting on it costs you twice. You spend effort on something that does not work, and you degrade the reviews you do collect.

What did the test actually find?

Joy Hawkins and the team at Sterling Sky published the test in January 2023 and last updated it in December 2024. The design is the reason it is worth more than a hundred correlation charts.

They picked a Christmas tree farm that, in their words, had no SEO efforts, no map pack SEO, and a website that was terrible. They ran it in the middle of summer, so the business was not going to be getting any reviews on its own. That combination isolates the variable about as cleanly as anyone can isolate anything in local search.

Six people left reviews over several months, containing keywords like "Christmas trees" and "fresh trees," against a base of 18 total reviews. On a profile that small, six new reviews is a large intervention.

The result, in Sterling Sky's own words: adding keywords to reviews did not help this business' local pack ranking on Maps, and actually, in the first instance, rankings got worse.

Their recommendation is equally direct: do not coach your clients to include certain words in their reviews, because this case study shows that it does not improve rankings.

A second independent test using an invented word found the same thing.

That is a controlled test rather than a correlation study, which is the distinction worth caring about. Correlation studies routinely find that high ranking businesses have keywords in their reviews. They also have more reviews, better categories, more established profiles, and more links. Untangling which of those is doing the work is exactly what a correlation study cannot do.

Reviews are a ranking factor. The words inside them are not the mechanism.

Then why do keyword reviews show up under high ranking businesses?

Because of justifications, and the causal arrow points the opposite way from the one people assume.

A justification is the snippet Google displays under a map pack result, often lifted from a review that matches the query. Owners see it, conclude that the review text produced the ranking, and go write a training script.

Hawkins addresses this directly. Justifications are often pulled in when Google finds a review that matches the query, and Google does the same thing with posts. But she adds the sentence that settles it: you have to rank first, and then if Google finds a review referencing the service, it pulls it in.

The justification is a consequence of ranking, not a cause of it. It is a real feature, it is nice to have, and it arrives after you have already won the position on category, proximity, and review signals that have nothing to do with wording.

What about keywords in my review replies?

Same answer, and there is a survey number for this one.

Whitespark's 2026 Local Search Ranking Factors report asked 47 local search experts to score 187 factors. Among the lowest scoring factors in the entire report, in the same neighbourhood as geo tagged photos and social follower counts, sits "Keywords in Owner Responses to Reviews" with a score of 28. Keywords in the Google Business Profile description scores 41, also near the bottom.

Contrast that with what does score. Quantity of native Google reviews with text ranks ninth among local pack factors. Recency of reviews ranks eleventh. Sustained influx of reviews rather than bursts ranks fourteenth.

The same survey that ranks review volume and recency in the top fifteen puts keywords in your replies near the very bottom. Reviews matter enormously and the words you stuff into them matter almost not at all, which is a consistent picture rather than a contradiction.

Grade that survey honestly, because it is opinion rather than measurement. Darren Shaw, who runs it, says so himself: it is a survey of opinions from local search experts, they do not have access to the actual algorithm, much of their opinion is based on observation, and correlation is not causation. He said that in the comment thread of the Sterling Sky keyword test, which is exactly the case where the survey's respondents were wrong and a controlled test proved it. Use the survey for direction and the controlled test for the verdict.

Why does everyone believe the opposite?

Because the belief is useful to the people selling review software.

"Prompt your customers to mention your services" is a feature that is easy to build and easy to sell. It gives a tool something to do beyond sending a link. And because businesses that use review tools generally do collect more reviews, the tool appears to work. It does. Just not for the reason on the box.

There is also a genuine confusion between ranking and reading. Keywords in reviews can matter to a human scanning your profile. Someone searching for gluten free options who sees three reviews mentioning the gluten free menu is more likely to choose you. That is a conversion effect happening in a person's head, not a ranking effect happening in an index.

Keep the distinction clean. Real customers describing what they actually came in for is good and useful. Staff coaching customers to insert a target phrase is a different thing, and it is the thing that does not work.

What is the actual cost of coaching keywords?

Three costs, and the third is the one that should stop you. The first two have arithmetic attached.

Opportunity, in hours. Say the keyword coaching adds twenty seconds to each ask, because the server now has to explain what to mention rather than just handing over the card.

A restaurant runs 14 server shifts a week with a target of 10 asks per shift. That is 140 asks a week, and 140 times twenty seconds is 47 minutes a week, or about 3.4 hours a month of paid floor time.

Price it properly. The Bureau of Labor Statistics Employer Costs for Employee Compensation release for March 2026 puts accommodation and food services at $19.92 in total compensation per hour worked against $16.12 in wages, a multiplier of 1.24 times base wage. At an $18 base that is $22.32 an hour, so 3.4 hours is about $76 a month, or $912 a year.

That is not a catastrophic number. It is a real number, spent on an intervention a controlled test found produced nothing, and it comes out of the same budget as the thing that does work.

Quality. A review reading "Great service from this emergency plumber in San Jose" is transparently coached. Readers discount it, and a profile full of them looks purchased even when every review is genuine.

Compliance. This is the one that should end the conversation. Google's Maps policy states that when soliciting reviews, merchants should not request that specific content be included. Coaching a phrase is requesting specific content, in the policy's own words. The same clause bans asking reviewers to name a staff member, and the keyword script usually shows up in the same training document as the staff quota, which Google bans separately.

Where does this finding stop applying?

Two places, and both are worth knowing so you can answer the person who argues with you.

Third party review sites are indexable pages. A Yelp or industry directory listing is a web page that Google crawls, and text on that page can feed ordinary organic relevance the same way any page's content does. That is a different mechanism from the local pack, which is what Sterling Sky tested. So review text can matter on the open web while doing nothing in Maps, and both statements are true at once.

One test is one test. It ran on a single small profile with 18 existing reviews, in a category with no competition during the off season, and Sterling Sky is an agency rather than a lab. It is still the strongest evidence available on the question, because the alternative is correlation studies and vendor blog posts, and it is corroborated by a second test with an invented word. Treat it as settled until somebody publishes a better test rather than a louder opinion.

Neither exception rescues the practice, because the policy problem survives both. Requesting specific content is prohibited regardless of whether it would have worked.

What should I do instead?

Ask more people, more often, closer to the visit.

Volume and recency are where the return is. Get the request in front of every customer at the moment the experience is fresh, because timing shifts response rates more than any wording change will.

Let customers say whatever they want. The unprompted specifics are the valuable part. When someone mentions the thing they came in for in their own words, that is worth more to the next reader than any phrase you could have suggested.

Put your keywords where they belong, which is your own website and your profile fields. Your services list, your business description, your category selection, and your page content are all things you control and all things that genuinely influence what you rank for.

Answer reviews the way a person would. Do not work your target phrase into every reply. Responses are customer service that happens to be public, and treating them as a ranking lever makes them worse at both jobs.

What to do this week

Open whatever training material your team uses for review requests. Find the line telling staff to prompt for services or the city name. Delete it.

Replace it with a shorter ask that specifies nothing about content.

Then look at your last twenty reviews. If more than a couple contain your city and service in the same sentence, someone has been coaching, and you should expect those reviews to read as low credibility to anyone paying attention.

Do the hours arithmetic above with your own shift count and wage. Knowing the coaching costs you real money makes the deletion permanent rather than temporary.

Finally, take the effort you just freed up and put it into the plumbing that actually drives discovery. If your citations are inconsistent, or if you never sorted out whether citation building is worth doing at all or is mostly theatre, that is a better use of the same hour.

Questions about advice you were given?

Email me at eric@seod.com with the review tactic someone told you to run, whether it came from an agency, a software vendor, or a conference talk. I will tell you whether there is real evidence behind it, what the evidence actually says, and what I would do instead. Include the source if you have it.

This one is genuinely useful to me too. Half of what circulates in local SEO traces back to a single unsourced blog post, and I like finding out which half.

Otherwise keep reading the reviews and reputation library, including how Yelp's solicitation rules differ from Google's.