Can Bad Reviews Stop ChatGPT From Recommending Your Business?

There is no published bad-review cutoff

OpenAI's description of ChatGPT Search says the product can search the web, use location for local results and show citations. It also warns that results can be incomplete, incorrect or outdated. The guidance does not name Google rating, review count, review wording or review recency as a guaranteed local recommendation factor.

That leaves room for reviews to be visible without turning them into a secret formula. Axios reported that Yelp licensed reviews, ratings, photos and business information to OpenAI for local responses. That describes one source relationship. It does not establish a score at which ChatGPT accepts or rejects a business.

A current preprint on local AI recommendations helps show why the answer is not as simple as "bad rating equals exclusion."

In a preprint paper published in 2026, researchers compared 2,208 search answers using ChatGPT, Claude, Gemini, and Perplexity to 4,776 food-and-drink venues in Bali. Review volume was associated with a venue's entering an answer, but star rating wasn't associated with entry. Among venues that were recommended, rating was associated with appearing first on the list. The study was observational. Its limited scope, just one region and one category, and combining answers from four AI systems, prevents it from establishing a universal threshold for local services in North America.

The result does not mean ratings are unimportant. It means the study found different relationships for being included and being placed first, in a narrow market that is not interchangeable with an electrician, dentist or restoration company.

Read the complaint before you read the average

Google says a mixture of positive and negative feedback can feel more trustworthy to customers, and that a negative review does not necessarily mean poor business practice. The text and circumstances matter. A one-star complaint about a closed parking lot does not carry the same buying consequence as several recent customers saying a locksmith added an undisclosed charge after arriving.

Google says more reviews and positive ratings can help a business in its own local results. That is Google's published system, and it should not be carried over to ChatGPT as though the same formula applies. Google Maps can also create Place Topics and review snippets from recurring words and themes in reviews.

A Place Topic may show the prevalent sentiment. That is why repeated language can become easier for customers to encounter than one unusual complaint. The source for this behavior is Google's current explanation of business summaries on Maps, not a ChatGPT ranking document.

This means that owners who panic over a single star rating should read the review text instead. They should be asking themselves what the reviews consistently say.

The practical distinction between the two things is an isolated complaint or one review of average or low rating versus recent and repeated complaints about issues that make the vendor a bad fit for the purpose at hand. Missed appointments, surprise charges, questions about safety, failure to return calls, or anything else that makes a vendor an inappropriate choice are decision-critical issues.

If several recent customers independently describe the same problem, the first job is operational. Asking for more reviews before fixing it only puts fresh customers into the same process.

One low rating changes a small profile more

Finally, it helps to think in terms of review-score arithmetic. For example, Google averages all the published ratings. A new rating might take up to two weeks before it appears on the public page. Let's say your business starts with all five-star reviews and receives one one-star. The average rating (computed as the sum of all the ratings divided by the number of ratings) would be:

These are just sample calculations based on assumed starting conditions. They aren't any threshold or safety level, nor are they predictions.

The calculation is AnswerPrism's illustration of the arithmetic, not a claim about how ChatGPT ranks businesses. It explains why the same one-star rating creates a larger visible change on a profile with five existing reviews than on one with one hundred.

• Five existing reviews: (5×5 + 1) ÷ 6 = 4.33
• Twenty existing: (20×5 + 1) ÷ 21 ≈ 4.81
• One hundred existing: (100×5 + 1) ÷ 101 ≈ 4.96

Four review problems need four responses

An isolated, credible complaint. Check the job record before answering. If the customer found a real failure, correct it and reply briefly enough to protect their privacy. A useful response tells the next customer what the business understood and what will happen next. It does not need a campaign to push the review down the page.

Several recent reviews describing the same failure. Assign the problem to someone who can change the work, give that person a deadline and check whether new jobs improve. A repeated complaint about the same decision-critical issue is public evidence of a real buying concern. A polished owner reply cannot substitute for the repair.

An old complaint about a process that has changed. If a reply would help, name the relevant change without making promises the business cannot support. Keep asking every eligible customer for an honest review through the usual follow-up. Do not offer an incentive or ask only the customers expected to leave five stars.

A fake, abusive, off-topic or extortion review. Use Google's reporting process. Google removes reviews that violate policy, not reviews merely because the owner disputes them. For negative-review extortion, Google tells owners not to pay or engage with the extortionist, to preserve messages and review links and to submit the evidence.

Google's advice for ordinary negative reviews is practical: avoid personal attacks, acknowledge a real mistake, explain relevant limits, apologize when appropriate, personalize the reply and move account details to phone or email. The response should make the situation clearer for a future customer without exposing the person who complained.

Check whether reviews caused the recommendation problem

A competitor's higher rating does not prove that reviews caused your business to be absent from a ChatGPT answer. The competitor may be closer, open at the required time, clearer about the service, better documented by independent sources or simply present in a variable answer.

Test a question that a real customer could ask. Include the service, the city and one constraint that would change the choice. An emergency electrician available tonight is a different question from an electrician for a planned panel replacement next month. Run the question in two fresh conversations, then save the businesses named, the reasons given and the visible sources.

If review sites appear among the sources, inspect the exact material used. If the answer points to a service page, directory, news report or other evidence instead, follow that trail. Reviews are one part of the public record, not a published ChatGPT score.

Start with the newest twenty reviews. Group the complaints that affect whether a customer would hire the business. One odd complaint needs a fair response. A current pattern needs an operational owner and a deadline.