You have been told to chase reviews for about fifteen years. The reason given was always Google, and the payoff was always vague.
The reason has changed, and the new one is much more direct.
Why reviews reach further than they used to
Google treated your reviews as one signal among hundreds, and then handed the customer ten links to choose from. An assistant does something different. It picks two or three products, names them, and that is the whole answer.
That is a far riskier thing for a machine to do. Recommending the wrong product reflects on the assistant, not just on the company that makes it. So before it stakes an answer on you, it looks for corroboration — some sign, from somewhere that is not you, that you are real and that people were happy.
Reviews are the cheapest and most abundant form of that corroboration on the internet.
How much difference do reviews make?
A large study suggests the difference is the difference between existing and not.
Seer Interactive, commissioned by Trustpilot, looked at 804,491 AI answers covering 1,926 brands across ChatGPT, Gemini, Perplexity and Google's AI Mode in March 2026. What they found, sorted by review profile:
- No profile at all: named in about 1% of relevant answers.
- A claimed profile, median 13 reviews: named in about 54%.
- An active profile, median 81 reviews: named in about 75%.
Two honest caveats, because they matter. The research was paid for by a company that sells review profiles, and an association is not proof of cause — businesses with 81 reviews tend to be doing several other things right at the same time.
Even allowing for both, the first step in that list is the one to look at. Going from nothing to a claimed profile with a dozen reviews is not a marketing campaign. It is an afternoon, and it sits next to the largest jump in the data.
Does this hold for software and product brands?
Yes, and the evidence is stronger here than almost anywhere else.
Shero Commerce collected every source cited across Google AI Mode, ChatGPT and Perplexity for purchasing questions in 60 product categories. Of the 1,851 sources, 2.8% were the brand's own pages. Third-party sites made up 59% — comparison articles, roundups, review coverage.
One number in that study is worth sitting with. When an assistant did name a specific brand, it cited that brand's own page 31% of the time. So your site does get read. It mostly gets read after something else has already put you in the running.
That is the argument for reviews in one sentence. Third-party coverage decides whether you are a candidate. Your own website is what gets checked once you already are.
How many reviews is enough?
Fewer than you would guess, and the number is less important than the pattern.
The jump from nothing to a median of 13 reviews carried most of the distance in the data above. Nobody needs four hundred. What matters more is that they are recent, and that they keep arriving. Sixty reviews spread over two years read like a going concern. Ninety that stop dead eighteen months ago read like a product nobody uses any more.
If you are starting from zero, ask the next ten customers who told you they liked the product. That is the whole plan.
Does one bad review hurt?
Much less than having nothing at all.
A 4.6 average over sixty reviews is a stronger position than a flawless 5.0 over three, and assistants treat it that way, because a handful of perfect reviews is exactly what a fake profile looks like. An occasional complaint with a calm, specific reply underneath it reads as a real company that handles problems.
What does hurt is a pattern — the same issue named over and over. An assistant summarizing your reviews will reproduce that pattern in its answer, and you will never see it happen.
Where to put your effort first
In this order, because the order is roughly the size of the payoff:
- Claim your profile on the platform your buyers actually read. G2 or Capterra for B2B software, Trustpilot for DTC and e-commerce, the app stores for a consumer app, Amazon for anything on a shelf. Fill it in completely.
- Make your details identical everywhere. One product name, one category, one description of who it is for, across every listing and profile. Conflicting details are how assistants end up confidently telling buyers the wrong thing.
- Get to a dozen reviews on your main platform. Then keep them arriving at a trickle rather than in bursts.
- Reply to all of them. Short, specific, naming the use case.
- Say plainly on your own site what you sell and who it is for. Written out as text, not sitting inside an image, a video or a product carousel.
The one you cannot control is Reddit, and it matters more than any single listing — community threads are among the largest sources ChatGPT draws on for product questions. You cannot plant those, and you should not try. What you can do is make sure that when someone does name you, the details they find everywhere else are right.
None of that requires a new website, and none of it is new marketing advice. What is new is the size of the consequence for skipping it. Under the old rules, a thin review profile cost you a few places in a list of ten. Under the new ones, there is no list of ten.
If you want to know where you currently stand before you start, that is what checking what AI already says about you is for — or we will run it across every assistant for you, free.