How Do I Audit a Reputation Agency's Promise to Suppress Negative AI Answers?

What can an agency actually change?

Engines summarise what sources say. If many independent sources report a genuine problem, no legitimate campaign will make assistants stop mentioning it quickly.

What should you ask the agency?

1. Which specific answers will you change? Ask for the exact questions, engines and wording they consider negative.
2. Which sources drive those answers? A competent agency will show the cited pages and explain which ones matter.
3. What will you do to each source? Correction, removal, a response, new content or new coverage.
4. What is the legal or policy basis for each removal? "We will get it taken down" needs a reason a platform or court would accept.
5. How will we measure progress? Ask for the question set, the number of runs and the reporting format.
6. What will you not do? Get written confirmation that they will not use fake reviews, fake accounts, undisclosed paid posts, or threats.
7. What happens if nothing changes? Pay for defined work and evidence, not for a promised outcome.

Which promises are red flags?

• Guaranteed removal from AI answers by a date.
• Claimed partnerships with OpenAI, Google or other AI providers that let them edit answers.
• Mass publishing of near-identical articles or AI-written pages. Google's spam policies explicitly apply to its generative AI responses, so this can backfire.
• Fake reviews or fake community posts to drown out complaints. The US Federal Trade Commission's rule on reviews bans fake and AI-generated reviews and allows civil penalties.
• Intimidating reviewers. The same rule bans review suppression through unfounded legal threats, physical threats, intimidation or certain false public accusations.
• Hidden text or instructions aimed at AI crawlers.
• No measurement plan, or screenshots of a single favourable answer as proof.

How do you measure whether it worked?

1. Agree the question set before work starts: 15–40 questions where the negative answer appears, plus a few control questions about unrelated topics.
2. Define "negative" precisely, such as "the answer mentions the 2024 data breach" or "the answer says the company is being sued". Vague sentiment scores are hard to audit.
3. Take a baseline: run each question three to five times per engine, and record how often the negative claim appears and which sources are cited.
4. Repeat monthly with the same method.
5. Link changes to actions. If the claim disappears after a specific source was corrected, that is evidence the work helped. If it disappears across all questions at once, including unrelated ones, a model update may be the cause.

Documenting false statements properly is covered in how to document a false statement generated by an AI assistant. Telling apart errors in sources from errors the model introduced is covered in how reputation teams can separate source errors from model synthesis errors.

What does an audited engagement look like?

Suppose a regional builder finds that assistants describe it as "facing multiple lawsuits over defective roofs". The agency's baseline shows the claim in about two thirds of runs across four engines. The cited sources are a local news article about one lawsuit that was later dismissed, and a complaints site repeating it. The agency's plan lists three actions. It will ask the newspaper to add the dismissal to the article, with court records attached. It will request a correction from the complaints site. And it will publish a page on the builder's site explaining the case and its outcome. Two months later, the newspaper has added an update and the complaints site has not responded. The claim now appears in about a quarter of runs, mostly in one engine that still cites the complaints site. The report shows exactly which source changed and which did not, so the builder can judge the work rather than take it on trust.

How long should it take?

Corrections to a specific wrong source can show in answers within weeks for engines that search the web live. Changes to what a model has absorbed during training take longer, often until a later model version. Genuine, widely reported problems fade slowly, as newer coverage builds up. An agency that sets expectations along these lines is more credible than one quoting a fixed date. Typical timescales are covered in how long it takes to reverse negative brand sentiment in AI answers.

What should an agency ask before promising to remove a competitor from AI answers?

Sometimes the request runs the other way: a client asks an agency to push a competitor out of AI answers. Before agreeing to anything, the agency should ask:

• Is the concern that answers are false, or just that the competitor appears? Correcting false claims about the client is legitimate. Removing a competitor's genuine presence is not a service anyone can honestly sell.
• What exactly do the answers say about the competitor? If they contain false claims about the client in comparison, the client can pursue corrections to those sources.
• Would any proposed tactic involve fake reviews, false reports, spam links or misleading content? Those can expose both agency and client to platform penalties and legal claims.
• Is there a legitimate alternative? Improving the client's own content, reviews and coverage so it appears alongside, or ahead of, the competitor.
• Can the result be measured honestly? The same baseline and repeated runs apply.

Why sabotage tactics fail is covered in whether you can sabotage a competitor's visibility in AI search.

What do people ask about suppressing negative AI answers?

Can we ask OpenAI or Google to remove a negative answer?

You can report inaccurate answers through the feedback tools, and some providers accept requests about personal data. They do not remove accurate criticism of a business on request.

Does publishing lots of positive content push negative answers out?

It can dilute old material if the content is useful and gets cited. Low-quality volume is more likely to be ignored or treated as spam.

Should we respond publicly to the negative source?

Often yes, if the response is factual and calm. Assistants sometimes include a company's response when summarising a complaint.

How much should suppression work cost?

Price it as defined work: corrections, content and outreach, with monthly measurement. Be wary of success fees tied to removing all negative mentions.