How AI Engines Decide Which Reviews to Quote

Ask ChatGPT or Perplexity “is [a local business] good?” and the answer often includes a quoted review. Sometimes flattering, sometimes not. The choice of which review to quote is not random. There is a mechanism, and most local operators do not understand it.

This is the working explanation.

What models look at when they choose a review to quote

Three factors determine which review gets pulled into an AI answer.

The first is platform credibility. AI engines weight reviews from Google, Yelp, BBB, and industry-specific platforms (Healthgrades for medical, Avvo for legal) higher than reviews from low-trust platforms. A review from your GBP is weighted higher than a review on a directory aggregator.

The second is recency. A 2024 review carries more weight than a 2019 review. Models bias toward recent because user intent in local search is overwhelmingly current (“is this place good now?” not “was it good five years ago?”). Reviews older than 18 months are deprioritized.

The third is specificity. A review that says “great service” is generic and uninteresting. A review that says “they fixed our heat pump in 90 minutes after the home inspection found a refrigerant leak” is specific, named, and concrete. Models prefer specific reviews because they let the model cite verifiable details that match the buyer’s question.

The implication for local businesses

Three concrete shifts.

First, a steady drip of recent reviews matters more than a large old pool. A business with 80 reviews from the last 12 months will get cited more often than a business with 500 reviews mostly from 2020-2022. See review velocity for why.

Second, the quality of the prompt you give customers when asking for reviews shapes the quality of the reviews you get. “Please leave a review” produces “great service, would recommend.” “What was the specific problem we solved for you, and how long did it take?” produces the kind of detail AI engines pull.

Third, multi-platform reviews matter. A business with 200 Google reviews and zero on BBB or industry-specific platforms looks weaker to an AI engine than a business with 100 Google + 30 BBB + 20 Yelp + 10 Healthgrades. The cross-platform pattern is read as “verified across multiple sources” instead of “all eggs in one basket.”

What about negative reviews

Negative reviews get quoted too, especially in queries with skeptical intent (“are there complaints about [business]”). The defense is not suppressing negative reviews. The defense is:

Responding to every negative review professionally and specifically. The model sees both the original review and the response. A response that addresses the specific complaint and offers a resolution path makes the negative review look like a one-off the business handled well, rather than a pattern.

Outweighting negative reviews with positive specificity. If your last 20 reviews have specific positive details and the one negative review is generic, the model is much more likely to pull from the positives.

What to actually do

Four operational changes.

Update your review request flow. After every job, send a request that asks for specifics: “What problem did we solve? How long did it take? Who on our team helped you?” Reviews that answer those questions are AI gold.

Build cross-platform reviews. Most local businesses concentrate on Google. Spread to two or three other platforms relevant to the industry: BBB, Yelp, and one industry-specific platform. The cross-platform footprint helps both Google ranking and AI citation.

Respond to every review within 7 days. Both positive and negative. Specifically, by name when possible. The response is part of what the model reads.

Track review velocity per month, not total review count. Velocity is the signal. Use a simple spreadsheet or any review-tracking tool. The metric to watch is “reviews per month over the last 12 months.”

The broader pattern

This is part of a bigger shift in how AI engines understand businesses. Old SEO logic: more is better. New AI-citation logic: recent and specific is better.

A business with 80 carefully cultivated, specific, recent reviews across three platforms will be cited more often than a business with 500 generic Google reviews from 2021. The cost of building 80 strong reviews is much lower than the cost of building 500 weak ones, and the AI-search payoff is higher.

What the audit looks at

The Google Business Profile Scorecard component of the audit measures review velocity, review specificity, response cadence, cross-platform footprint, and the gap between your review profile and the three strongest competitors in your city. The deliverable names exactly which fixes in this list will move the most surface area first. Score your own GBP first with the GBP scorecard tool or book the full audit.

The short version

AI engines pick reviews to quote based on platform credibility, recency, and specificity. Operators who train their review request flow to elicit specific feedback, build cross-platform footprints, and respond consistently to both positive and negative reviews get cited more often than operators who chase total review count. Recent and specific beats large and generic. Update the request prompt; the rest follows.