Which Review Ratings Does AI Quote About Your Business?

Between March and October 2026, the Cheers AI visibility grader asked ChatGPT, Gemini and Perplexity what customers say about 455 local businesses by name. Only 1 in 20 answers stated the business's Google star rating or review count. Most answers cited other directories or review sites. Your reputation across those pages matters when a customer asks AI about you. This study used the engines' developer APIs with web search enabled; the rating comparison below covers the figures those answers actually quoted.

Last verified: November 1, 2026.

The Google numbers that did show up were mostly right, though review counts missed more often than star ratings. The ratings they hear instead can come from a page with a small fraction of your Google reviews behind it, so start with the BBB and Angi pages those answers cite.

Google is testing AI answers about businesses in Search and Maps too. On October 29, 2026, Search Engine Roundtable reported that Google is letting more businesses into its AI-powered Business Profile messaging pilot, where customers message the business from Search or Maps and an AI agent answers routine questions before handing off to the owner (only on phones that support RCS messaging). A practitioner quoted in that post said the agent can only answer as well as the profile does. On September 17, the same site reported that Google has been testing an AI Overview version of the local knowledge panel, with a "Show more" button into an AI Mode chat.

We've measured the ratings on the ChatGPT app's map cards before, in Why ChatGPT Shows a Lower Star Rating Than Google. For this study, we checked the ratings and review counts AI states when a customer asks about one business by name, along with the sources it attributes them to. Whether AI recommends you to a homeowner who's never heard of you is a separate test.

Most answers quoted no Google number at all

Just 67 of the 1,346 answers quoted a Google rating or count, covering 59 businesses, and most of those gave both a rating and a count.

Stars themselves weren't rare: about a quarter of the answers (355) stated a star rating of some kind, far more than stated a Google one. Part of the reason is the question itself, which asks what customers say rather than what the rating is, so every Google number in this sample is one an engine chose to volunteer.

What AI reads when a customer asks about you

Half the answers cited the business's own website (52.3%). Three in four cited at least one directory or review platform other than Google (75.4%), and the BBB or Angi turned up in 55.6%.

The BBB alone appeared in 45.9% of by-name answers across all trades, with Perplexity citing it in 63.7% of its answers and ChatGPT in 46.9%. Angi appeared in 26.5% and HomeAdvisor in 14.8%. Home-service companies get their own BBB breakdown in our piece on whether BBB accreditation is worth it for AI search.

ChatGPT cited one review-aggregator site that republishes businesses' Google reviews in 26.0% of its answers, and a figure it labeled "Google" usually came through another page. In 35 of the 37 ChatGPT answers that stated a Google rating or count, the answer named a review-aggregator page, a directory listing or the business's own website as its source. That makes the number secondhand, only as current as the page that repeated it.

Perplexity cited Yelp in 21.8% of its answers and, more oddly, the jobs site Indeed in 26.2%, across all trades. Company pages on Indeed carry reviews from employees, so when Perplexity cites one in an answer about what customers say, an employee's view of the company can sit right next to the customer reviews. Our article on why AI cites Indeed reviews for home-service companies has the home-service figure.

In this clip I walk through where AI looks before it recommends a company.

On camera I sort that work into the places where a business needs to be visible. In these by-name answers, your own site and the directory and review pages led by the BBB and Angi carried much of the evidence. Build those records so an assistant can find a clear, consistent account of the business you run.

Those are the sources behind questions about one company; for the sources behind "who should I hire" questions, see Where Does ChatGPT Get Its Information?

BBB and Angi ratings in AI answers often ran below Google's

Same business, same run. When an answer quoted a business's BBB or Angi star rating, the figure ran 0.2 stars or more below that business's Google rating 44% of the time (27 of 61 ratings, 53 US businesses) and matched it within a tenth of a star just as often.

The two sites didn't behave alike. BBB figures ran low in 9 of 17 ratings across 16 businesses (median 0.28 lower), and none ran 0.2 stars or more above Google's. Angi's 44 ratings across 39 businesses came closer to an even split: 18 ran low (median 0.10 lower) and 21 matched.

HomeAdvisor's 12 ratings showed no consistent gap. Review-aggregator pages that republish a business's Google reviews matched its Google rating within a tenth in 17 of 18 (17 businesses), which is what you'd expect from a page built on the same reviews.

The BBB figures sit on far fewer reviews: every one came with a review count, the median was 25, and 14 of the 17 rested on fewer than one review for every ten the business had on Google. Some of the low BBB and Angi figures rested on a handful, such as a 4.0 average from 4 reviews. At that size one unhappy customer moves the number a long way. A rating built on so few reviews describes a smaller, different set of customers than a Google history many times its size, so the gap on its own doesn't make either figure wrong.

BBB's overview of its ratings says customer reviews don't go into its letter grade, which rests on complaint history, time in business, licensing and similar information in the business's file. So the stars an AI answer quotes from a BBB page and the letter grade on that same page are two separate figures.

Most Google figures met our comparison rules

When an answer did state a Google rating, it met our comparison rule in 59 of 65 answers: within a tenth of a star, or a stated range containing the captured Google rating.

Review counts were shakier. Of the 61 answers that stated a Google review count, 45 met our comparison rules and 16 missed, split evenly above and below. The rules allow a 10% or two-review tolerance for exact counts and wider allowances for ranges and approximate language, as described in the method. Counting both kinds of figure, 21 of the answers that quoted a Google number carried at least one that didn't match.

Among the misses, some answers read a Google count off another page that showed a lower number; one answer, working from a third-party directory listing, gave a company with about 2,300 Google reviews "roughly 1,700 Google-sourced reviews." Others rounded up past the real count, like "over 800" for a roofer with about 600.

Google says a business's rating is the average of all ratings published on Google for the place, and that a new review can take up to two weeks to show in the score (Google Business Profile Help). One caveat of our own: a page that counts only reviews with written text will look low next to Google even when it's current, because Google's count includes star-only ratings.

Gemini flagged a similarly named business in 9% of answers

Gemini pointed out a different business with a similar name in 39 of its 450 by-name answers, while ChatGPT and Perplexity rarely did it.

Naming the other company can keep the two apart. In 13 of those answers, though, Gemini went on to describe the other company's complaints, poor ratings or legal trouble, inside an answer about yours. Those are only the cases you can see, since we can't measure how often an engine blended two businesses without saying so.

Street addresses were rarely wrong: 4 of the 90 answers that gave the business's own street address didn't match the Google profile.

One fix covers both problems, and it starts with making your identity easy to tell apart. Use your real business name, exactly as it appears on your signage, on every profile, with nothing added. Where a profile shows an address, make it the same address everywhere, and if you hide your address on Google because you serve customers at their homes, keep it hidden there. Then link your site to your own profiles with sameAs, which gives an engine one more way to tell you apart from a business with a similar name; sameAs Schema: JSON-LD Examples for Local Businesses has the markup.

Check what AI says about your business

It's five steps, and all you need is three chat apps and a spreadsheet.

This week, start with your three busiest locations, or your only one. In ChatGPT, Gemini and Perplexity, ask the same question: "What do customers say about [your company name] in [city]?" Log every star rating and review count in each answer, one row per number, with the site the answer ties it to and your live Google rating and count beside it. The apps also show ratings on map cards and business blocks that our developer-API sample didn't have, so log those on their own rows with the site each one comes from. Give the log columns for the date, location, engine, the number as stated, the site the answer tied it to (with its URL), your live Google figure, and the fix you made with its re-check date.

Now open the BBB profile those answers cite and treat it as a page customers read. Claim it, correct the name, address, phone, hours and category, and reply to the customer reviews on it. Paying for BBB accreditation is a separate question.

Do the same on Angi and HomeAdvisor. When a BBB or Angi rating rests on a handful of reviews, one bad job can pull it down, and it still lands in the same answer as a Google count many times larger.

Next, find the pages that repeat your Google numbers. Search your company name with "Google reviews" and check the directory listings, manufacturer dealer pages and review-aggregator pages that display your Google rating or count. Where a page lets you refresh or correct it, do; where your own website prints a rating and count, keep it current or put a date next to it.

Then clean up your name, address and sameAs links as described above. After 30 days, ask the same question in the same three engines and compare the new answers with your log.

For multi-location brands and PE portfolios, add a by-name check per location to your quarterly audit, with one row per number quoted and the site it came from. Flag every location where an answer quotes a BBB or Angi rating 0.2 stars or more below its Google rating, and re-run the check 30 days after any fix.

Once you know which page an answer is reading, the step-by-step correction workflow is in How Do I Fix Wrong Business Information in ChatGPT?

Study scope

These answers came from developer APIs with web search turned on, not from the apps people use, so the sample has no map cards and its rates shouldn't sit beside numbers from the Cheers Market Baseline, our own research panel. The ChatGPT app's map cards have their own rating gap, mostly from Yelp-linked cards, which we measured separately in Why ChatGPT Shows a Lower Star Rating Than Google.

The Google figures are too few to rank the engines against each other, or to split by the two changes inside the window: new models in July, and the business's city going to ChatGPT and Perplexity as an approximate location from September 23. Gemini's grounding links can undercount some sites, so its citation shares may run low.

The businesses picked themselves by running our public grader, Cheers' free AI visibility check, and they lean established, with a median of 285 Google reviews and a median rating of 4.9. About 7 in 10 are home-service companies.

Too few answers stated a phone number, opening hours or a closure for us to report a rate; phone accuracy gets its own study. Google's AI Overviews and its AI review summaries weren't part of this check either, and what an owner can and can't change in that summary is in Can You Change a Google AI Review Summary for Your Business?

Method

The Cheers AI visibility grader sends its questions to ChatGPT, Gemini and Perplexity through each company's developer API, not through the consumer apps. Web search is on for every question: ChatGPT runs its web search tool, Gemini grounds its answer in Google Search, and Perplexity searches the web by design. Each engine gets the same short instruction to answer the way a helpful assistant would for a real buyer, to name only real businesses and to say when the evidence is thin. For each business we ask four buyer questions built from its category and city (for example, "Who are the best [trade] companies in [city]?") and one question that names the business and asks what customers say about it; until early October, that question asked what customers praise or complain about. Every question goes out as a fresh request with no signed-in user, chat history or personalization, and we store the answer text and the links it cites exactly as returned. Since September 23, ChatGPT and Perplexity also receive the business's city as the asker's approximate location, and in late September we changed the wording of the buyer questions. Between March 18 and June 8, ChatGPT answered the four buyer questions for about a third of businesses in one combined request; we retired that method. Answers from developer APIs can differ from what a signed-in person sees in the apps, which add map cards, phone numbers and business listings that the APIs don't return. Engine models change over time; this sample spans March 17 to October 30, 2026, Pacific time.

This study uses only the by-name question, counted November 1, 2026, from the latest complete grader run for each Google place, minus Cheers' internal runs: 455 businesses in 361 cities (427 US, 17 Canadian, 11 elsewhere or no address). Of 1,364 by-name questions, 171 used today's wording and 18 returned an error, leaving 1,346 answers; citation shares count all 1,364, so an error cites nothing.

We compared each figure with Google profile data the grader captured in the same run. A rating matched within 0.1 of the live rating, or if a stated range held it, and a count within 10% or two reviews. "Over N" matched when the live count was at least N and N at least half of it; "nearly N" when the live count was at most N and at least three-quarters of N; "hundreds" and "thousands" by order of magnitude. We left out figures that blended Google with other sites, and five-star-only counts.

A wider first pass caught 127 answers (9.4%) with "Google" near a rating, an "N-star" phrase or a review count; reading all 127 left 67.

We read every line that tied a rating or count to Google, and every rating tied to the BBB, Angi, HomeAdvisor or a review-aggregator page. A BBB or Angi figure ran low at 0.2 stars or more below the same business's Google rating from the same run. Those comparisons cover what the answers attributed to each site, not the sites' own pages. We read every candidate answer to confirm a flagged similarly named business, and compared street addresses with the Google profile.

We report shares and medians rather than averages, every published rate has at least 10 businesses behind it, and no business is named.

Sources

• Google Business Profiles AI-Powered Messaging Expands, Search Engine Roundtable, October 29, 2026. More businesses in Google's AI messaging pilot, RCS phones only.
• Google testing the local knowledge panel in AI Overview form, Search Engine Roundtable, September 17, 2026. The AI Overview panel test and its "Show more" button.
• Overview of BBB ratings, BBB, September 11, 2026. Customer reviews don't feed the letter grade.
• Understand review scores for local places & businesses, Google Business Profile Help, undated. The rating is the average of all ratings; updates take up to two weeks.

Sources read November 1, 2026.

Dylan Allen-Arnegård is the CEO & Co-Founder of Cheers, the done-for-you local visibility partner for established home-service companies and PE-backed portfolios.