How many reviews before AI recommends you

The number

Across the audits we run, businesses that get named by AI engines tend to clear a rough floor: 30 or more reviews at 4.3 stars or better. In competitive metros that floor rises past 100.

Below it, a business is rarely named regardless of how good its website is. Above it, other factors start deciding.

This is an observed pattern, not a published rule. No engine documents a review threshold. But it shows up consistently enough across markets to be worth planning against.

Why engines lean on reviews at all

Because reviews are the closest available proxy for the question actually being asked.

Someone asking an assistant for a good med spa is not asking which one has the best website. They are asking which one is good. An engine has no direct way to assess clinical quality, so it uses the strongest available signal of aggregate human judgement — which is reviews.

That is also why review signals cut across all four engines despite their different indexes. ChatGPT reaches them through Bing and Yelp, Gemini through Google, Perplexity through community discussion and aggregators. Different paths, same underlying signal.

Recency is doing more work than volume

This is the part most operators get wrong.

A practice with 200 reviews and nothing new in six months consistently loses to one with 60 updated weekly. The engine is not only asking is this place good — it is asking is this place currently operating and currently well regarded.

A dormant review profile reads like a business that may have declined, changed hands, or closed. Volume without recency is a historical record, not a current signal.

The practical implication: a steady trickle beats a burst. Ten reviews a month for six months is worth more than sixty in one campaign followed by silence.

Where the returns stop

Past roughly double the local median, additional reviews stop meaningfully changing whether you are named.

Getting from 20 to 60 matters enormously. Getting from 300 to 340 does not. If you are already the most-reviewed practice in your metro, the constraint has moved somewhere else — usually third-party editorial coverage or community presence — and more review effort is wasted effort.

Check your local median before setting a target. Chasing a number without a benchmark is how practices over-invest in the one lever they know how to pull.

Multi-unit operators have an extra problem here: reviews accrue to a location rather than to a brand, so a strong flagship does nothing for a new site across town. That is covered in why AI names the treatment, not your studio.

The rating floor is separate

Volume does not compensate for rating.

Below about 4.3 stars, more reviews will not fix the problem — they may make it worse by increasing confidence in a mediocre score. If you are sitting at 4.0, the work is operational rather than marketing, and no amount of review solicitation substitutes for it.

How to actually get them

The only sustainable answer is to build the request into the visit rather than bolting it on afterwards.

• Ask at the point of maximum satisfaction — usually immediately post-treatment, not three days later by email
• Make it one tap. A QR code at checkout outperforms an emailed link substantially
• Spread the ask across staff so it does not depend on one person remembering
• Set a per-location monthly target and report against it, because what gets measured gets asked for

Work out your own required pace with the review velocity calculator — it takes your current count, the local threshold and your patient volume and tells you the request rate you actually need.

What not to do

Do not buy reviews. Every platform detects it, penalties are severe and often permanent, and for a healthcare operator the regulatory exposure is worse than the marketing loss.

Do not gate the ask. Screening for happy patients before requesting a review breaches Google’s policies and several platforms’, and the risk is not worth the average-rating gain.

Do not incentivise. Discounts or entry into a draw in exchange for a review violates most platform terms and, in healthcare, may create additional regulatory problems in some jurisdictions.

The shortcut in this category is genuinely worse than the slow route. A suppressed or penalised review profile takes far longer to recover than it would have taken to build honestly.

The caveat that applies to all of this

No AI platform publishes how it weights reviews, or whether it weights them at all in a given query. The thresholds above come from measured output across many audits and from patterns documented across the industry — they are a planning floor, not a specification.

They also move. Treat any specific number as something to re-check rather than something settled.