How review signals feed AI recommendations
What a language model weighs when it decides which local businesses to name
The four signals a model reads
Volume
"Is this a real, established business with enough history to name?"
Threshold signal
Works like a floor, not a scoreboard. 9 reviews vs 300 matters. 400 vs 600 mostly doesn't. Compare yourself to the companies AI already names in your city.
Rating
"Is this business credible, and does the profile look organic?"
Band signal
Below ~4.0 hurts. 4.3 to 4.8 is the credible band. A 5.0 on 22 reviews reads as a campaign, not a business. Chasing 4.6 → 4.9 is wasted effort.
Recency
"Is this business still operating and still doing the work well?"
Freshness signal
The one most businesses lose quietly. A few new reviews every month beats 60 in one quarter, then silence. A 14-month-old newest review answers nothing.
Wording
"What exactly does this business do, and where does it do it?"
Highest leverage
Models read the sentences, not just the score. "Same-day water heater swap in Collierville" earns a specific recommendation. "Great service!" earns none.
The model builds a short list — there is no page two
User asks an AI assistant: "Who should I call for same-day water heater replacement near Germantown?"
Business A
280 reviews · 4.7 · newest 3 days ago · reviews name water heaters and the suburb
Business B
190 reviews · 4.6 · newest this week · several mention same-day service
Business C
340 reviews · 4.5 · steady flow · strong on plumbing repair generally
Not named: the 5.0-rated company with 22 reviews, and the 4.8 with 400 reviews whose newest one is from last year.
Review gatingScreening out unhappy customers breaks Google policy and creates the too-perfect pattern.
Bought reviewsShort, generic, five-star clusters are legible as fake to a system built to read language.
Chasing a perfect scoreMoving 4.6 to 4.9 is the most common wasted project in local reputation work.
Replying to every reviewWorth doing for the humans reading your profile. Don't expect it to change an AI answer.