Explainer
Do Google reviews still matter when customers ask AI instead?
The short answer
Yes, and arguably more than before. When an assistant recommends a business it is looking for reasons, and reviews are the densest source of them. 45% of US consumers now use AI tools to find local businesses, up from 6% a year earlier, and only 1.2% of businesses get recommended by ChatGPT.
What actually changed
The channel changed and the evidence did not. BrightLocal's 2026 survey found 45% of US adults had used an AI tool to find a local business, against 6% the year before, with ChatGPT alone at 31%.
But an assistant asked to recommend a plumber has the same problem a person does: it needs a reason to name one business over another. Reviews are the densest available source of reasons, which is why they survived the shift intact.
What did change is how few businesses are eligible. SOCi found only 1.2% of local businesses get recommended by ChatGPT against 35.9% appearing in Google's local three-pack. The pool is small, and being in it is cheaper than most people assume.
What an assistant reads in a review
Not the star rating alone. A rating is one number and an assistant is assembling an argument, so the useful signal is in the text: what the business is good at, what it was hired to do, whether the same praise recurs.
This is why a business with forty specific reviews often outperforms one with two hundred that say great service. Specificity gives an assistant something to quote.
What actually helps
- Reviews that name the job: a boiler replacement, a crown, a wedding shoot
- Recent ones, because a stale profile reads as a business that may not still be operating
- A spread across platforms, since assistants draw on Yelp and Foursquare data as well as Google
- Your replies, which show a business that is present and are themselves readable text
Why recency now matters as much as volume
Two hundred reviews with the most recent from three years ago describes a business that was good once. Forty from this year describes one that is busy now, and an assistant weighing which to recommend has no trouble telling those apart.
There is a supporting signal in how AI systems treat content generally: HubSpot's research found a visible last-updated date predicts citation more strongly than the original publish date. Freshness is doing real work across the board.
The practical version is a steady trickle rather than a campaign. Five reviews a month indefinitely beats fifty in one push, and it beats it on exactly the axis assistants care about.
Beyond Google
OpenAI licensed Yelp data in July 2026 to power ChatGPT's local answers, on top of an existing licence for Foursquare's Places dataset. So Yelp is no longer an optional platform you ignored a decade ago; it is feeding an assistant directly.
Reddit matters more than most owners expect too. It is 12% of ChatGPT's US citations, and Perplexity cites it in 46.7% of responses to commercial queries. You cannot manufacture that, and you should not try, but being the business people mention favourably in a local thread is worth more than it looks.
How to actually get them
The only method that survives a busy week is an automatic one. Manual asking fails for the same reason every good intention fails: the moment to ask is when the job is finished, and that is exactly when you are packing the van.
Trigger the request off job completion in whatever software you already use. Keep it to one sentence and one link. Send it by text, since SMS open rates run around 98% and an email lands in a folder nobody opens.
And reply to everything, including the bad ones. A calm reply to a three-star review does more good than the review does harm, because it is read by everyone who comes after, and increasingly by something summarising you.
Five reviews a month indefinitely beats fifty in one push.
What not to do
Do not buy reviews. Detection has improved, platforms remove them in batches, and a profile that loses thirty reviews overnight looks worse than one that never had them.
Do not gate the ask so only happy customers get the link. Several platforms treat that as a violation, and it produces a suspiciously perfect profile that reads as bought.
Do not chase a five-star average. A business with a handful of three-star reviews and thoughtful replies reads as real. A flawless profile reads as managed, and an assistant summarising you has plenty of both to compare.
Common questions
- Do reviews affect whether AI recommends my business?
- Yes. An assistant recommending a business needs reasons, and review text is the densest source of them. Specific reviews that name the job are more useful than a high star rating with generic text.
- How many reviews do I need?
- Fewer than you think, and more recently than you think. Forty specific reviews from this year generally outperform two hundred generic ones from three years ago, because recency signals a business still operating.
- Does Yelp still matter?
- More than it did. OpenAI licensed Yelp data in July 2026 to power ChatGPT's local answers, alongside an existing Foursquare licence, so your Yelp entry now feeds an assistant directly.
- Should I reply to bad reviews?
- Always, calmly, and without arguing the facts in public. The reply is read by everyone who comes after and by anything summarising your business, so it does more good than the original review does harm.
- Is it worth buying reviews?
- No. Detection has improved and platforms remove them in batches, so a profile that loses thirty overnight ends up looking worse than one that never had them.