A homeowner opens Perplexity at 10 p.m. and types “best plumber in Portland for old houses.” Three company names come back, each with a one-sentence summary, each quoting something a real customer said. The homeowner picks one and makes the call.
That whole transaction happened without Google, without a search results page, and without any paid advertising. The reason one plumber got recommended and the other three didn’t is almost entirely about reviews, and the way AI search engines interpret them is different from how Google’s blue-link search used to.
This is a practical look at how reviews actually flow into AI recommendations, which platforms matter, and what a business should do this quarter to improve the odds.
How AI models actually read reviews
There are two distinct paths reviews take into an AI answer. Understanding the difference is worth the next few paragraphs.
The first path is retrieval-augmented generation, which is what Perplexity, Gemini, Google AI Overviews, and ChatGPT with web access use. When a user asks a recommendation question, the system performs live web searches, pulls snippets from review sites and business pages, and synthesizes an answer. The reviews that get quoted or paraphrased in the response were pulled fresh during that query.
The second path is training data. ChatGPT, Claude, and other chat models have consumed massive amounts of publicly available review data during training. When a user asks about a well-known business, the model can answer from memory, often with summarized sentiment from reviews it saw months or years ago. This is why a business with a strong review history in 2023 might still get recommended in 2026 even if their current reviews have slipped.
For local and small business queries, retrieval dominates. For national brand queries, training data dominates. Most practical AEO work targets the retrieval path because it responds faster to effort and because it is where most buyer queries happen.
Which review signals matter most
AI engines don’t just read review text. They score a business on a cluster of review signals, and the mix matters.
Average star rating is the first filter. Businesses below 4.0 stars rarely get recommended by AI. The 4.3-4.7 range is the sweet spot. Above 4.8, AI models sometimes get suspicious of review manipulation and start weighting rating velocity and review text variety more heavily.
Review count is the second filter. Under 20 reviews, most AI engines treat a business as too new or too small for confident recommendation. Between 50 and 200 reviews, the business becomes a candidate. Over 200, it becomes a default option in the category.
Recency is a major factor AI engines weight heavily. A business with 500 reviews averaging 4.6 stars where the most recent review is 14 months old loses to a business with 120 reviews at 4.5 stars where the last 30 came in this quarter. Freshness signals the business is still operating and still delivering.
Review text content gets parsed for specific services, pain points, and outcomes. A roofer with reviews mentioning “fixed our leak,” “reshingled the whole garage,” and “helped with insurance paperwork” will rank in AI answers for all three of those queries even if they never wrote a page about insurance paperwork on their site. The reviews are doing the SEO work.
Response presence signals an active business. AI engines see businesses with owner responses on 80%+ of reviews and treat them as more credible than silent ones. The response text itself also becomes indexable context.
Which platforms AI engines trust
Not all review sites weigh equally. Here is the rough hierarchy as of mid-2026, based on what shows up in AI answers.
Google reviews lead. They feed Google’s AI Overviews directly and are indexed heavily by third-party AI models. Every local business should treat Google as the primary review platform.
For software and B2B, G2 and Capterra dominate. ChatGPT and Gemini cite both platforms extensively when answering “best software for X” queries. A software company with under 50 G2 reviews is typically invisible to AI shopping queries.
For service businesses, the second tier varies by category. Yelp for restaurants and consumer services. BBB for contractors and financial services. HomeAdvisor and Angi for home services. Trustpilot for online commerce and fintech. Houzz for interior design and home improvement.
Industry-specific platforms matter more than generic ones for their niche. A wedding photographer on The Knot and WeddingWire beats the same photographer with triple the Yelp reviews when the query is wedding-specific.
Glassdoor, Indeed, and employee review sites affect AI answers about company culture and employer reputation, even if the query is about products or services. A company with aggressively negative employee reviews can see AI models reflect that in responses about the brand’s reliability.
The review velocity problem
Most businesses with stale AI visibility suffer from the same issue: review velocity has slowed. They crossed a milestone 18 months ago and stopped pushing for new reviews. AI engines see the gap and stop recommending them.
Review velocity is the rate at which new reviews come in. A business adding 4-8 Google reviews per month signals active operations. A business that added 80 reviews in a burst two years ago and none since signals a business that might not exist anymore.
Target a steady cadence. For a small local business, 4-8 reviews per month across Google plus the category-specific platform is enough to maintain AI visibility. For a medium business (5-25 employees), 15-30 reviews per month keeps the engines confident. For larger operations, 50+ per month is reasonable.
The easiest way to hit velocity is to build review requests into the service workflow. A restaurant prints a QR code on every check. A contractor sends a one-click review link 48 hours after job completion. A SaaS company emails power users a review ask 30 days after activation. Make requesting a review part of the operation, not a marketing campaign.
Writing to the AI, not just the customer
A subtle lever most businesses miss: customers often don’t know what to write in a review. They type “great service, thanks” and move on. This produces a review that does almost nothing for AI search context.
Prompt the customer. Your review request email or thank-you page can include suggestions like: “If you have a moment, we would love it if your review mentioned the specific service we provided, the person who helped you, and what problem we solved. Future customers searching for help find these details useful.”
This small prompt changes the aggregate review corpus. Instead of 200 reviews saying “great service,” you get 200 reviews naming specific services, team members, and problems. AI engines consume this richer corpus and recommend you for more queries.
Don’t write reviews for customers, don’t suggest specific language, and don’t incentivize reviews beyond what platform rules allow. Just guide the customer toward detail.
Responding to reviews the right way
Responses matter both for AI signals and for the humans who read them on review pages. Most business responses are wasted effort because they are templated.
For positive reviews, a response should acknowledge a specific detail from the review, use the customer’s first name, and add a short human line that isn’t corporate. “Glad we got the tub drain cleared before your in-laws arrived, Paul. Tell Trisha the kids can stop taking showers at their place.” That response adds indexable text, shows the business pays attention, and reads like a human on both sides.
For negative reviews, acknowledge the problem, take responsibility where warranted, offer a specific path to resolution, and move the conversation offline. “We dropped the ball on the Tuesday appointment, and I owe you that apology in person. I’m calling you at 2 p.m. tomorrow unless you tell me another time works better.” Don’t argue in the review response. Don’t attack the customer. AI engines pick up defensive or combative responses and downgrade the business.
Respond within 48 hours when possible. AI engines and humans both read response latency as a signal of how the business operates.
How reviews interact with schema and structured data
AI engines combine review data with the structured data on your website to build a complete picture of the business. A business with strong reviews and weak website structure often loses to a business with fewer reviews and clean schema.
Add Review schema and AggregateRating schema to your homepage and service pages. This tells search engines and AI models the review data is yours and connects it to your business entity. Most businesses never do this and lose review credit in the process.
Add Organization schema with your full business details. Name, address, phone, founded date, service categories, service areas. AI engines use this to disambiguate your business from others with similar names and to confirm the reviews belong to you.
If you have a knowledge graph presence or a Wikipedia-style entity, link your reviews to that entity through sameAs properties in your schema. This is advanced territory, but it’s how enterprise brands capture AI recommendations more reliably.
The feedback loop
Reviews don’t just influence what AI engines say about you. They influence what customers know to ask for. A customer who reads AI answers citing specific services will ask about those services. A customer who reads reviews mentioning specific team members will ask for them. The review corpus shapes the sales conversation.
Pay attention to what your reviews say. If the theme is “fast response time,” double down on that positioning. If reviews consistently praise a specific person or service line, feature them on the website. The reviews are telling you what your market values. Most businesses ignore the signal and make marketing decisions based on what the leadership team thinks matters.
The 90-day plan
If AI search visibility is the goal, here is what actually moves the needle.
Month one: audit every review platform where your business has a presence. Claim any unclaimed listings. Ensure name, address, and phone match across all platforms. Add 10-20 review requests per week into your service workflow.
Month two: respond to every unanswered review from the past 12 months. Add Review and AggregateRating schema to your website. Start prompting customers to mention specific services in their reviews.
Month three: track AI search mentions for your business. Use tools like Profound, AthenaHQ, or manually check by asking ChatGPT, Perplexity, and Gemini the top 20 queries for your category and location. Document where you appear and where competitors beat you.
By month three, most businesses see measurable lift in AI mentions. By month six, the cumulative effect of consistent velocity and responses produces real traffic from AI referrals. The businesses that commit to the system in 2026 will be the default recommendations in their categories for the rest of the decade.
Reviews are not a marketing side-project anymore. They are the dominant signal AI engines use to decide who gets recommended. Treat them like the growth channel they are.