Someone in Norwich asks an assistant for "the best dentist near me that's good with nervous kids" and gets three practices back, each with a sentence of reasons: calm with children, easy parking, Saturday appointments. Nobody at those practices wrote that sentence; it was assembled from their patients' reviews. When an assistant names a local business, reviews are usually its evidence, and volume, recency and content each do a different job in that answer.
How an assistant answers "best nearby"
An assistant with web search doesn't rank a list the way the map pack does: it breaks the question into smaller ones, fetches evidence for each, and writes a summary. Google describes its AI Mode as using a query fan-out technique, where one question becomes many searches. Other assistants work the same way, and the sub-questions for a local "best" query are predictable: who is nearby, who is well rated, who handles the specific situation, who is open when it matters, and what have people said recently.
Then it needs evidence, preferably third-party, specific and current. For a small business, the largest body of text that fits all three is its reviews. Your own website says what you'd like to be true; a directory says you exist; a review says a named person had a specific experience last month. That is why the reasons an assistant gives read like review excerpts.
Two honest limits. No assistant publishes how it weighs reviews against other sources, so what follows is observed mechanism, not specification. And not every local query gets an AI answer: plenty of "plumber near me" searches on Google still return only the map pack, which runs on the rules in How local search works.
Volume is confidence
An assistant recommending a business is staking its own credibility, and it hedges by preferring businesses with more independent reports. The count and the star rating are also the two facts an assistant can read straight from a listing without opening a review, so they act as a first filter: a business with a thin record rarely makes the shortlist, however good it is.
There is no published threshold, and chasing one leads to the policy breaches covered in Getting reviews consistently. That lesson's routine is the whole answer: a steady flow, month after month, compounds into volume.
Recency is proof you're still there
An assistant is also wary of recommending a business that has closed or gone downhill, and its only cheap check is the date on your latest reviews. A profile whose newest review is a year old looks, to a model summarising the web, like a business nobody has visited in a year. Recent reviews say you are trading, and that the praise describes the business as it is now.
This is the same steadiness argument as the map pack, with a sharper edge: an assistant writes sentences, and "recent reviews mention" is a sentence it likes to write. Give it something recent to mention.
Content answers the sub-questions
Volume and recency get you on the shortlist. Content is what gets you named for the specific question. Look again at the fan-out: "good with nervous kids", "Saturday", "emergency", "wheelchair access", "Headingley". A review that says "fitted our new boiler in Headingley on a Saturday and talked my nervous mum through the controls" answers four sub-questions in one sentence. A review that says "great service, would recommend" answers none.
You can't write reviews for customers, because Google's content policy requires a review to reflect the reviewer's own genuine experience, but you can shape what arrives.
- Ask in a way that invites specifics. Your ask message can say "if you can, mention what we did and where", a request for detail rather than for words or stars.
- Put the specifics in your replies. Your response is text too, and can name the service and area naturally when the review didn't: "Glad the Saturday boiler install in Headingley went smoothly." One natural mention, as Responding to reviews describes.
- Make your site agree with your reviews. An assistant cross-checks, so if reviews praise your emergency callouts and your website never mentions them, the claim weakens; a service page that says the same thing makes both more credible.
- Keep the facts consistent. Hours, address, phone and services should match across your profile, site and directories, because an assistant that finds disagreement tends to leave you out rather than guess.
One warning if you show reviews on your site. Google's review snippet documentation says marking up your own rating on your own page is self-serving and ineligible for rich results, so quote reviews as text and leave the star markup off.
Anyone promising placement in AI recommendations is selling the same thing as a guaranteed map-pack ranking, which Google's own help says cannot be requested or bought. The later chapter on how AI assistants pick local businesses covers the other sources alongside reviews.
What to take away
- An assistant answering "best nearby" breaks the question into sub-questions and answers them from third-party evidence, which for a small business is mostly reviews.
- Review volume gets you on the shortlist, recency proves you're still trading, and review content decides whether you're named for the specific question.
- You can't script reviews, but you can invite specifics, add them naturally in your replies, and make your site say what your reviews say.
- No assistant publishes its weighting and not every local query gets an AI answer, so build the evidence and check the answers rather than chasing a promise.
Next
Assistants and Google both cross-check your details across the web; the next chapter starts with NAP consistency and citations, which makes your name, address and phone agree everywhere.