Do reviews and ratings affect whether AI recommends you?

Yes, reviews and ratings are part of what AI assistants read. Google says Gemini answers questions about places from Google Maps data, including ratings and reviews. ChatGPT has used licensed Yelp reviews and ratings since 23 July 2026, and Perplexity licenses Yelp and Tripadvisor data. No assistant publishes how much weight a rating carries. What is well established is the effect of ratings on people, and people are the ones who read the AI answer and decide.

What AI assistants read about your reviews

The clearest statement comes from Google. When it opened Google Maps grounding to developers in October 2025, Google described Gemini answering questions about places using Maps data on more than 250 million places, including addresses, hours and user ratings, with reviews used as a source for the answer (Google, 17 Oct 2025). In other words, the rating and reviews Gemini sees are your Google rating and your Google reviews. BrightLocal's tests found that Google's AI shows Business Profile data even when it cites a different source (BrightLocal, 22 Jul 2025).

Other assistants get review data through licensing deals:

• ChatGPT: since 23 July 2026, OpenAI has licensed Yelp's reviews, ratings, photos and business details, covering 330 million reviews; since 10 August 2026, users can book a table or join a waitlist through Yelp inside ChatGPT (PYMNTS, 10 Aug 2026). The deal is not exclusive.
• Perplexity: shows Yelp maps, reviews, photos and business details for local questions since March 2024 (Maginative, Mar 2024), and since January 2025 receives Tripadvisor's reviews, ratings, review summaries and 11 million business listings in 43 countries (Tripadvisor, 9 Jan 2025). In BrightLocal's 2025 tests, Perplexity cited Yelp in every industry checked (BrightLocal, Jul 2025).
• Siri and Apple Maps: ratings in the US and Canada come mainly from Yelp, with Apple's own ratings added on top (BrightLocal).
• Copilot: searches with Bing; industry overviews report that Bing's listings show Tripadvisor, Yelp and Facebook ratings rather than reviews of its own (Microsoft Bing, Oct 2025).

Licensing deals change. Everything in this list is dated for that reason, and any of it can be out of date by the time you read it.

Which rating each assistant sees

These ratings are not the same number. Across 21,841 US restaurants, the Google rating was on average 0.7 stars higher than the Yelp rating, 93% of restaurants rated higher on Google, and the top ten lists on the two platforms barely overlapped (Li and Hecht, 2020). A restaurant that looks like a 4.6 to Gemini can look like a 3.9 to ChatGPT. Checking only your Google rating tells you about one assistant.

What is shown for AI, and what is not

Here is the honest state of the evidence on the AI side:

• Shown: assistants have access to ratings and reviews and use them as sources (the Google, Yelp and Tripadvisor announcements above).
• Observed, not proven: places that ChatGPT recommended had an average rating of 4.3 in SOCi's study of more than 350,000 locations of 2,751 chain brands (SOCi, 28 Jan 2026). SOCi sells local marketing software, the data covers chains rather than independents, and an average among recommended places does not show that the rating caused the recommendation.
• Expert opinion: in Whitespark's 2026 survey of 47 local search specialists, the top factor for visibility in AI answers was being included in expert "best of" lists (Whitespark, 6 Nov 2025). That is what practitioners believe, not a measurement.
• Not shown: no assistant publishes a minimum rating, a review count it needs, or how reviews are weighted against other signals. Anyone quoting a precise threshold for AI is guessing.

What is shown for people

The human side has decades of research, much of it causal. It matters for AI because the person reading an AI answer usually checks the rating before they call or book.

Ratings and revenue

• Yelp, Seattle: a one-star increase in Yelp rating led to a 5 to 9% increase in revenue for independent restaurants, with no effect for chains. The study uses rating rounding to separate cause from correlation (Luca, Harvard Business School).
• Yelp, San Francisco: moving from 3 to 3.5 stars raised the chance a restaurant was fully booked at 7 pm from 13% to 34%; moving from 3.5 to 4 added another 19 percentage points (Anderson and Magruder, Economic Journal, 2012).
• Tripadvisor, Italy: when cheaper EU mobile roaming made reviews easier to read on the go, revenue and employment at mid- and high-rated restaurants rose 3 to 10%, and the probability that low-rated restaurants closed doubled (Donati, Marketing Science, 2025).

These studies use Yelp and Tripadvisor data. There is no comparable causal study on Google ratings, and because Google ratings run higher than Yelp's, the Yelp thresholds above should not be read as Google thresholds.

What people say they do

BrightLocal's 2026 survey of 1,002 US adults is a survey of stated intentions, not observed behavior (BrightLocal, 11 Feb 2026):

• 68% say they would only use a business rated 4 stars or more; 31% say 4.5 or more.
• 74% say they look at reviews written in the last three months.
• 89% say they expect the business to reply to reviews, and 80% say they are more likely to use a business that replies to all of them.

The recency figure is worth noticing. A high rating built years ago, with nothing recent, may not satisfy the reader who goes from the AI answer to the reviews.

Owner replies

Replying to reviews is one of the few actions with research behind it, though most of that research uses hotel data:

• Hotels that began replying on Tripadvisor saw ratings rise by 0.12 stars and review volume rise by 12%, with fewer short negative reviews (Proserpio and Zervas, Marketing Science, 2017; hotel data).
• Public replies to negative reviews raised later ratings when the reply was visible to the next reviewer, and personalized replies worked better; replies to positive reviews slightly lowered later ratings (Wang and Chaudhry, Journal of Marketing Research, 2018; hotel data).
• A later study of more than 150,000 reviews found the opposite on positive reviews, where replies raised later ratings and revenue, and found replies to negative reviews lowered them in the short run but improved them in the long run (Wu, Ye and Chu, Journal of the Academy of Marketing Science, 2024).

The studies disagree on details, so the practical reading is simple: reply, make replies specific, and judge the effect over months rather than weeks. Google's own guidance says replies show customers you value their feedback (Google Business Profile Help).

What not to do: fake and bought reviews

If ratings matter, buying them is tempting. The evidence says it does not last, and the platforms the assistants read are actively removing them.

• It is illegal in the US. The FTC's final rule, announced 14 August 2024, bans buying or selling fake reviews and paying for reviews that express a particular sentiment, and allows civil penalties against knowing violators (FTC, 14 Aug 2024).
• Platforms remove them at scale. Google blocked or removed more than 240 million policy-violating reviews in 2024, along with 12 million fake business profiles (Google, 7 Apr 2025). Of roughly 22 million reviews Yelp received in 2025, about 70% were recommended, 17% not recommended, 11% removed by moderators and 2% removed by their authors (Yelp Trust and Safety Report 2025).
• The boost is short. On Amazon product data, buying reviews produced a noticeable but short rise in ratings and sales; after purchases stopped, average ratings fell and the share of one-star reviews rose sharply (He, Hollenbeck and Proserpio, Marketing Science, 2022; e-commerce data).
• It is a sign of weakness, not strength. In Boston, about 16% of Yelp restaurant reviews were filtered as suspicious, and restaurants were more likely to commit review fraud when their reputation was weak or competition increased (Luca and Zervas, Management Science, 2016).

The same FTC rule also covers rewarding customers for reviews of a particular sentiment, such as a discount for five stars, and reviews from staff or insiders that do not disclose the connection. When a filtered or removed review disappears, so does any rating it lifted, in every assistant that reads that platform.

What to do instead

1. Find out which ratings matter for you: your Google, Yelp and Tripadvisor numbers, side by side, since each assistant reads a different one.
2. Ask every real customer for a review, steadily rather than in bursts, on the platforms your customers and the assistants use.
3. Reply to reviews, positive and negative, specifically and promptly.
4. Fix what the reviews complain about. On hotel data, properties that acted on frequent complaints improved their ratings more than others (Ananthakrishnan, Proserpio and Sharma, 2023).
5. Check what the assistants actually say, with the same questions every week, and record which rating and which review site they cite.

None of this guarantees an assistant will name you. It makes sure that when one does, or when a customer checks, the reviews behind your name hold up. That is the part AIReputa measures and works on with clients.