AI Says My Company Is a Scam: How to Investigate the Claim

Is the AI accusing your company, or describing misuse of its name?

Read the whole answer, including the sentences around the word “scam.” Identify who is alleged to have done what. A warning about someone pretending to represent a business is different from an accusation that the business itself committed the act.

The FTC describes business impersonation scams in which messages appear to come from a familiar business but actually come from a scammer. A business name appearing in such a warning does not, by itself, identify that business as the perpetrator. Source: FTC, Business Impersonator Scams.

These categories are starting points, not verdicts. A complaint deserves examination even if the chatbot has exaggerated it. A business's own denial also needs evidence; it does not automatically settle the dispute.

Anthropic acknowledges that language models can produce incorrect facts or content inconsistent with the supplied context. Its guidance recommends checking claims against sources and warns that mitigation does not eliminate errors. That makes verification necessary, but it does not make every adverse answer a hallucination. Source: Anthropic, Reduce hallucinations.

Save the original answer and the evidence behind it

Before challenging the chatbot, save the complete exchange. If a customer sent a cropped screenshot, request the surrounding wording and links where they can share them appropriately. A cropped image may omit a qualification or the identity of the organization being discussed.

• The test: exact prompt, date, AI product and model label if shown.
• The answer: full text, screenshots and any accessible conversation link.
• The sources: cited URLs, titles, relevant passages and the date you checked them.
• The context: whether search was shown, and which files or company facts were supplied.
• The incident: messages, sender details and transaction records relevant to the allegation, if available.

Store originals securely and use redacted copies for routine discussion. Avoid pasting customer identifiers, account details or private correspondence into a general-purpose chatbot merely to test an accusation.

For suspected internet crime, the FBI's Internet Crime Complaint Center advises keeping original documents in a secure location. Its guidance identifies material such as email headers and transaction records as potentially relevant evidence. Source: IC3, Related Evidence.

If messages or links look suspicious, do not interact with them just to investigate. The FTC advises against clicking links or calling numbers supplied in unexpected messages that appear to come from businesses. Use an independently known contact route instead of the one in the disputed message. Source: FTC, Business Impersonator Scams.

Check whether the cited page supports the accusation

Inspect the original source through a safe route. A page title, search snippet or link beside an answer is not enough. Locate the passage that supposedly supports the allegation and record what it actually says.

Check the subject, action and scope

Ask three questions: which organization is identified, what conduct is described, and what evidence is offered? A report that someone received a suspicious job offer does not establish that the named employer sent it. Equally, a genuine transaction complaint should not be relabeled impersonation without checking who was involved.

If the answer cites an official notice, read that notice directly and preserve its wording and status. Do not broaden a claim about a particular product, transaction or individual into a conclusion about every activity of the company.

Distinguish a source from a related link

Google's Gemini documentation explains that links may point to sources or related content, including public websites and connected material. Not every response includes links. Check how a link was presented and whether it supports the precise claim, rather than assuming it was evidence for the entire response. Source: Gemini Apps, View related sources.

If no supporting passage can be found, write “not supported by the material reviewed.” An unavailable page or an unsuccessful search is not proof that the source never existed. Do not invent an explanation of the model's training data or internal reasoning.

Do not treat repeated wording as independent corroboration

Several pages may repeat the same original statement. Trace the claim back as far as you can and distinguish independent accounts from copies, summaries and references to the same incident. Keep a separate note of what you verified directly.

A search for your name alongside “scam” can help locate relevant material. The word was part of your query, however, so its appearance in the results is not a finding about the company. Evaluate the contents of each result.

Separate your business from namesakes and impersonators

Compare the domain, location, contact details and activity in the source with your own records. A matching name or copied logo is not enough to establish affiliation. If identifiers point to another organization, use our guide to AI confusing your company with another company.

For a suspicious recruitment message

Ask your recruitment team whether the sender and vacancy are authorized. Use contact details already known to the business. Check the full email address and any destination domain rather than the display name alone. Record the evidence for your answer, including the possibility that an external recruiter is authorized.

The FTC warns that job scams can appear on job sites and social media, and advises people not to pay for the promise of a job. Such a request deserves investigation; the use of your company name does not establish who made it. Source: FTC, Job Scams.

For a disputed payment or customer transaction

Check whether the transaction involved your business, an authorized intermediary or an unrelated party using its name. Compare relevant order records, communications and the destination of the request. Handle sensitive evidence through an appropriate private channel.

If the complaint concerns your actual service, investigate it through your normal complaint process. Correcting an AI summary does not resolve an underlying customer problem. If the facts remain disputed, preserve that uncertainty.

For an older incident or warning

Compare dates and what happened afterward. An accurate historical warning is not necessarily a description of present conduct. Our guide to outdated company information in AI answers explains how to preserve the timeline without rewriting history.

Choose the right correction or reporting route

If the AI answer misstates the source

Use the product's available response-feedback mechanism. Supply the exact sentence, the source passage and the specific mismatch. For example, explain that the page describes impersonation of the business while the answer attributes the impersonator's actions to the business itself.

For Gemini, Google provides feedback controls for inaccurate responses. Its documentation says the associated conversation is included with feedback and explains how included material is handled. Review what will be shared before submitting sensitive evidence; available options can differ by account. Source: Gemini Apps, Send feedback or report a problem.

For a ChatGPT response, our guide to requesting correction or removal of incorrect company information explains the reporting routes and the separate process for personal data. Match the request to the specific claim and the evidence you have preserved.

If a publisher has the wrong identity or facts

Send a focused correction request through the publisher's process. Identify the page and sentence, explain the error, and provide evidence for the proposed change. Ask for a correction to what is inaccurate, rather than removal of accurate criticism simply because it is unfavorable.

Subject: Please review the identification of [company name]

Page or response: [URL or reference]
Statement to review: "[exact sentence]"
Specific issue: [wrong entity, unsupported attribution, or other error]

Supporting evidence: [public URL and relevant passage]
Proposed correction: [precise wording supported by that evidence]
Still unresolved: [any material uncertainty]

Please review this point and let us know the outcome.

If there is evidence of an impersonating site or account

Report the specific activity to the platform or relevant service provider, preserving the report reference and response. A fake social account, a phishing site and an inaccurate article may require different processes. Describe the evidence rather than submitting only the chatbot's conclusion.

ICANN's DNS abuse guidance covers harms including phishing and malware. For a generic top-level domain, it describes reporting abuse to the registrar first and a contractual-compliance route for concerns about the registrar's response. This is not a general process for removing unfavorable website content. Source: ICANN, DNS Abuse Mitigation Program.

Where suspected crime is involved, use the relevant reporting authority for your circumstances. In the United States, IC3 receives internet-crime complaints and shares information with appropriate agencies. Its FAQ also allows reports from people outside the United States. Filing a complaint does not guarantee an investigation or a website takedown. Source: IC3, Filing and handling complaints.

If your customers need a public warning

When the evidence justifies a notice, make it specific and dated. Describe the conduct you have verified, identify your genuine contact route and explain how customers can check a message. Distinguish an unauthorized sender from your company without claiming that every complaint is fraudulent.

Exclude private customer details and avoid turning suspected malicious URLs into clickable invitations. If facts change, update the notice clearly. A warning published by the business is useful context, but it is not independent proof that all disputed conduct came from someone else.

Ask the AI to identify evidence, not endorse your preferred conclusion

After preserving the original answer, ask for a claim-by-claim review. Avoid replacing the investigation with “prove my company is legitimate” or asking the model to assume misconduct. Both approaches choose the conclusion before evaluating the evidence.

Review this statement about [company name]:
"[exact statement]"

For each factual allegation, identify:
- the organization or person the source describes;
- the conduct alleged and the date or period;
- the source URL and the passage supporting the allegation;
- whether it concerns the company, a namesake, or an alleged impersonator.

Separate source statements from your own inferences.
If you cannot access a source or verify a claim, say so.
Do not treat the presence of the word "scam" as evidence.
Do not assume either wrongdoing or innocence.

This is an editorial template, not a validated detector of fraud. Check every quoted passage yourself. If you supply excerpts because the model cannot access the sources, label the result as based on supplied text.

A revised answer may be more precise, but the revision does not establish that the underlying incident was resolved. Keep the model's interpretation separate from the documents and records used to assess it.

Retest the claim and record what actually changed

Once you have a meaningful update, repeat the original question under comparable conditions. Record the product, model label, date, retrieval activity and any added context. A new conversation alone does not establish that personalization or other context was absent.

Distinguish these outcomes in your record:

• Source corrected: the publisher changed the inaccurate statement.
• Abuse action observed: the platform reported an action, or the specific content is no longer available when checked.
• Answer corrected with context: the AI revised its wording after you supplied evidence.
• Comparable answer improved: the original question produced a better-supported response without you supplying the correction.

Record unsuccessful checks and unresolved claims as well. A missing search result is not proof that the incident did not happen. One successful response does not establish a permanent correction across models or users.

What ReputationScan can show about name misuse

ReputationScan examines how its tested Claude and Gemini models describe a company by name, with the country when supplied. Those model queries have no web tools and do not receive the company's website. Website information and other public signals are collected separately. The scanner does not currently test ChatGPT.

Separately, its name-misuse check searches for the company name alongside “scam” and may include an apparent warning linked from the company's website. The analysis is instructed to distinguish third-party misuse of the name from wrongdoing by the business. This is a limited review of the material returned, not an independent determination of who committed an act.

The search term itself is not a finding. A warning can merit human review, while no usable result is not a certificate that a business or offer is safe. Automated analysis can make mistakes, and a company-authored warning is not independently verified merely because the scanner surfaced it.

Read the evidence, the tested models and the report's original scan time. Repeated requests can return a cached report. See what each scan check measures and the scan methodology before drawing broader conclusions.

ReputationScan refers interested businesses to SecondSideMedia, a partner that publishes corrective public records, and may earn a commission from resulting business. A scan does not change AI answers, remove content or establish whether a business is legitimate.

Sources and review notes

Sources checked on 15 September 2026. Templates and review categories are editorial guidance, not case findings or experimental results. Scanner descriptions were checked against its implementation.

1. FTC — Business Impersonator ScamsMessages impersonating familiar businesses and precautions with unexpected contact.
2. Anthropic — Reduce hallucinationsFactual errors, evidence-based verification and the limits of mitigation.
3. FBI Internet Crime Complaint Center — Frequently Asked QuestionsPreserving evidence, eligibility to report and complaint handling.
4. Gemini Apps — View related sourcesThe distinction between sources, related content and responses without links.
5. FTC — Job ScamsRecruitment-scam patterns and checking job offers.
6. Gemini Apps — Send feedback or report a problemResponse feedback and the associated conversation material.
7. ICANN — DNS Abuse Mitigation ProgramDNS abuse, registrar reporting and the limits of ICANN's role.