How Do I Dispute a Negative Review Summary in ChatGPT?

Is the summary false or just unflattering?

Start by checking the summary against its sources:

Be honest in this step. Disputing an accurate summary wastes time, and fixing the underlying problem is the only thing that will change it.

How do you report a false summary?

1. Document it: the question, the full answer, the date and every cited source.
2. Use the thumbs-down feedback on the answer and explain precisely what is false, with links to the correct information.
3. For serious false claims, such as allegations of crime, fraud or safety failures, use OpenAI's content reporting route. The process is covered in how to report a false allegation in ChatGPT through OpenAI content reporting.
4. Fix the source if one is cited: a misattributed complaint, a scraped copy of someone else's reviews, or an article that confused you with another company.
5. Retest a week later, and again after a month.

How do you change an accurate but outdated summary?

• Respond to complaints publicly. A short, factual reply on the review platform shows what happened next: "We refunded this order in full and changed our delivery partner in May." Replies give assistants the other half of the story, if they can read them.
• Close the loop in your public reply. When you resolve a complaint, say so in the reply with what you did; reviewers can update their own reviews if they choose, and readers and assistants see the resolution either way.
• Publish what changed. A page on your site describing the changes you made after recurring complaints, such as a new refund policy, extra staff or a new supplier, with dates.
• Earn recent, detailed reviews. Ask every customer, not only happy ones, to describe their experience. Short reviews such as "Great!" give a summary little to work with; a sentence about delivery or service gives it something to quote.
• Stay within the rules. The FTC's rule on consumer reviews, in effect since October 2024, prohibits fake reviews and incentives conditioned on positive sentiment. It also prohibits review suppression, such as using groundless legal threats or intimidation to stop people posting.

How review content shapes recommendations is covered in how reviews, ratings and review text affect whether AI recommends your business.

Why does Claude mention my old complaints but not my responses or newer reviews?

Because the old complaints are easier for it to find and read. Several things combine:

• Access rules on review platforms. Some review sites keep AI search crawlers out of their business pages. Yelp's robots.txt, for example, keeps Anthropic's and OpenAI's search crawlers off its business pages. Crawling is not the only route, though: Yelp licenses its reviews to OpenAI for ChatGPT and supplies review data to Perplexity, so those assistants can still use Yelp reviews. Claude has no such arrangement we know of, and Yelp blocks Anthropic's crawlers, so Yelp reviews are much less likely to reach Claude's answers, though its search provider's index can still surface them. Check each assistant separately before deciding where recent reviews will be seen.
• Complaint pages are rich text. A complaint on a consumer complaint site or a forum is long, specific and dated. Complaints on the Better Business Bureau stay on a business's profile for three years from filing. A detailed three-year-old complaint is a better source for a summary than ten recent one-line reviews.
• Responses may be hidden. Owner replies that sit behind a "show response" button, or load by script, may not be in the text a crawler receives.
• Different search index. When Claude searches the web, Anthropic lists Brave Search among its subprocessors for web search. The pages that rank there for "[business] reviews" or "[business] complaints" may differ from what you see on Google.
• Model knowledge lags. Without a search, Claude answers from training data, which predates your newer reviews.
• The question invites it. "What do people complain about with [business]?" retrieves complaint pages by design.

The fix is to put your best recent evidence where assistants can read it. Ask every customer for an honest review on platforms that allow requests and that AI search crawlers can read. Make your replies visible as text, and publish a dated "what we changed" page on your own site. Check the robots.txt of the review sites that matter to you. What you find there tells you which reviews assistants can see.

Can you use your own site?

Yes, carefully. A page of customer feedback on your own site is readable by assistants, but it is self-selected, and assistants and readers know it. Google's review snippet documentation also says self-serving reviews of a business on its own site are not eligible for review rich results. Use your site for what only you can say: what changed after complaints, your guarantees and policies, and links to your profiles on independent review platforms.

How do you track the summary?

Ask ChatGPT, Claude, Gemini and Perplexity, several times each:

• "What do customers say about [business]?"
• "[Business] reviews"
• "Common complaints about [business]"
• "Is [business] reliable?"
• "[Business] vs [competitor] customer service"

For each answer, record the sentiment, the complaints mentioned, whether they are dated, and whether any response or improvement is mentioned. Note which sources are cited. Track the share of answers that mention outdated complaints as current, and aim to push it down month by month. How to ask an assistant directly what it relies on is covered in how to ask ChatGPT why it does not recommend your brand.

What might an illustrative case look like?

Consider an illustrative furniture retailer whose ChatGPT and Claude summaries said "customers frequently report delivery delays and poor communication". The complaints were real but dated from a warehouse move three years earlier. Recent reviews were positive but short, and most sat on a platform that kept AI search crawlers out. The retailer replied to the old complaints with what it had changed, and published a dated page on its new delivery process. It began asking every customer for an honest review, in their own words, on a platform that allows AI search crawlers. After three months, most summaries said delivery problems had been reported "in the past" and described recent reviews as positive.

What do people ask about review summaries?

Can OpenAI remove a summary because it is negative?

No. Reports are for false or harmful content, not unfavourable but accurate summaries.

Should we ask customers to post on a specific platform?

You can suggest platforms that allow review requests, such as Google, but ask all customers, not only happy ones, and offer nothing in return for a review. Do not ask for Yelp reviews: Yelp does not allow businesses to request them.

How long do old complaints keep appearing?

As long as they remain the most detailed, findable sources. Newer, detailed reviews and visible responses shorten that time.