Key takeaways:
- Assistants name two or three businesses, not ten — being merely competitive stops paying
- Reviews feed the data these answers are built from; a thin profile is barely represented
- Specific review wording gives a model something concrete to say about you
- Stale profiles read as possibly-closed to anything weighing recency
- Nobody can optimise for this directly — be wary of anyone claiming otherwise
A customer with a burst pipe used to open a search engine, glance at the map results, and call one of the first few. Increasingly they ask an assistant instead, and get back a sentence or two naming one or two plumbers with a reason attached. The underlying question is identical. The shape of the answer is not, and that shape changes what it is worth being good at.
The compression is the whole story
A results page is generous. Ten listings, a map, several more below — being fourth still puts you in front of someone, and a share of people scroll and compare. An assistant is not generous. It names two, perhaps three, and moves on. There is no fourth position to occupy.
That is the single most important thing to understand about this shift, and it does not require any theory about how these systems work. Whatever the mechanism, an answer with three slots is far less forgiving than a page with ten. Being roughly as good as your competitors used to earn you a share of the traffic. In a three-slot answer it frequently earns nothing.
What these answers are actually built from
It is worth being careful here, because there is a lot of confident nonsense written about it. No provider publishes how a local recommendation is assembled, and the systems change. What can be said with reasonable confidence is where the underlying information comes from: map and directory data, business listings, and pages that discuss or compare local options. Reviews are a heavy component of nearly all of it — they are the main thing distinguishing one plumber's listing from another.
The practical consequence is indirect but reliable. A business with nine reviews is thinly represented everywhere that data is drawn from. It is not being filtered out by an algorithm with an opinion about it; there simply is not much there to say. A business with two hundred recent reviews has a rich, consistent presence in exactly the sources these answers are built on.
Why the wording started to matter more
A ranked list does not need to describe anything. It can put you fourth without forming a sentence about you. An assistant recommending you has to say something — and what it can say is limited to what exists in writing about your business.
This gives specific reviews a value they did not previously have. "They replaced our water heater the same day we called and cleaned up after themselves" contains a service, a timeframe and a behaviour. "Great service, highly recommend" contains nothing that could be turned into a reason. Both are five stars and both help your average; only one gives a model something to work with when a customer asks who is good for emergency water heater work.
You cannot script this, and you should not try. What you can do is nudge: asking customers to mention what you actually did for them is a small change to a request template that meaningfully changes the material you accumulate over a year.
Recency reads as still-in-business
Freshness has always been a local ranking factor, and it plausibly carries more weight in an answer that recommends rather than lists. Consider it from the system's position: a business whose most recent review is from last week is evidently trading. One whose most recent is from fourteen months ago might have closed, changed hands, or stopped caring. Recommending the second carries a risk the first does not.
This is the strongest argument for treating review collection as an ongoing habit instead of a campaign. A business that gathered eighty reviews two years ago and nothing since looks worse on this dimension than one with thirty spread evenly across the last twelve months, despite the larger total.
Consistency across sources
If your name, address and phone number differ between your website, your map listing and the directories that carry you, you are harder to identify confidently as one business. That has always been unhelpful for local SEO. When something is assembling an answer from several sources at once, ambiguity is more costly — a system that cannot reconcile two records may reasonably favour a business it can.
None of this is new advice. It is the same hygiene local SEO has recommended for a decade. What has changed is that the penalty for ignoring it has grown.
What not to buy
Expect to be sold "AI search optimisation" as a distinct service, with a monthly fee and a dashboard. Be sceptical. There is no submission endpoint, no ranking panel, no verified technique for placing a business inside a generated answer, and anyone claiming a reliable method is describing something they cannot demonstrate.
The honest version is duller and cheaper: accumulate genuine recent reviews, keep your listing details accurate and consistent, and make it easy to establish what you do and where. That is what feeds every one of these systems, and unlike a proprietary technique it also works if the systems change again next year.
What to actually do about it
Nothing here calls for a new strategy, which is the reassuring part. It raises the stakes on the one you should already have.
Collect reviews continuously rather than in bursts, so recency holds. Ask customers to mention the specific job, so the language accumulating about you is concrete rather than generic. Keep your details consistent everywhere they appear. And accept that being slightly better than a competitor no longer earns a slightly smaller share — in an answer with three names, it increasingly earns all of it or none.
SnappyRatings keeps review collection running continuously — QR, email and SMS requests with automatic follow-ups. Start a 21-day free trial →