What to Do When an AI Engine Says Something Negative About You

The instinct is to treat this like a bad review: respond, dispute, escalate. That is mostly wrong. An AI answer is a summary of what the engine found, weighted by what it trusts. The useful question is not "how do we make it stop" but "where did it get that."

Step one: find the source

Most engines that search the web will show citations if you ask, and Perplexity shows them by default. Run the exact prompt again and read the cited pages. In our testing the negative line almost always maps to one of these:

• A Reddit thread, often older than you would expect
• A G2 or Capterra review cluster with a recurring complaint
• A comparison article written by a competitor or an affiliate
• Your own documentation or changelog describing a known limitation
• A news story about an outage, a price change, or a layoff

If the engine does not show sources, search the key phrase in quotes with your brand name. That usually surfaces the thread that taught the model the sentence.

Step two: decide if it is true

This is the step people skip. If the complaint was accurate eighteen months ago and you fixed it, you have a freshness problem. If it is still accurate, you have a product problem, and content work will not outrun it. If the complaint keeps showing up in new reviews, it keeps showing up in answers. A negative line that is true and current is market research you got for free.

Step three: fix the record, not the answer

You cannot edit an AI answer. You can change what the engine finds next time. The tactics depend on the source.

Old Reddit threads. You cannot delete them, and flooding a subreddit with positive posts gets noticed. What works is a newer thread with a real, specific update. A founder or support lead answering a current question with actual response-time numbers is the kind of content engines quote.

Review sites. If the complaint was fixed, the fastest change comes from recent reviews that mention the fix. Ask customers who experienced the improvement to review honestly. Do not write them yourself or pay for them. Fake reviews get detected, and the detection becomes its own negative story.

Competitor comparison pages. You cannot change their page. You can publish your own comparison that is more specific, with real limits stated on both sides. Engines tend to favor the page with numbers over the one with adjectives.

Your own docs. If your changelog still says "known issue: exports may time out" from two years ago, update it. Engines treat your documentation as authoritative on your limitations, so keep it current.

News coverage. A story about an outage will stay indexed. What you can add is a public postmortem on your own site. That page tends to get cited alongside the original story, and sometimes instead of it.

Step four: publish the counter-fact

For each negative claim, make sure one page on your site states the current truth with a number. "Median first-response time for support is under four hours on all plans, measured monthly" is quotable. "We care deeply about our customers" is not.

Put the fact where an engine would expect it: pricing complaints on the pricing page, support complaints on an SLA page, reliability complaints on a status page.

Step five: measure the change

Track the prompts that produced the negative line. Run them weekly across the engines that matter and record whether the line still appears, whether it has softened, and which sources are cited. At AEO GEO Labs we fold this into the weekly share-of-answer report with a separate row for sentiment, because a brand mentioned often but negatively is not winning.

Retrieval-based engines tend to shift within weeks of new sources appearing. Answers drawn from training data take longer.

What not to do

Do not send legal letters to the engine. Do not try to get the Reddit thread removed. Do not create accounts to argue with the reviewer. Each produces a new, fresher negative source, and fresher sources tend to win.

The only durable fix is making the true, current fact easier for an engine to find than the old complaint. For help tracing a negative line to its source, AEO GEO Labs includes source mapping in every visibility audit.