AI gave wrong information about my business. now what?

ChatGPT just told a potential customer that your business is in a different city. Or that you offer a service you discontinued two years ago. Or that you were acquired by another company when you are fully independent. Or that you’re pricing is half of what it actually is. The AI said it with complete confidence. No disclaimer. No "we are not sure about this." Just a clean, authoritative statement that happens to be wrong, delivered to a person who was about to become your customer.

This is not a rare occurrence. For general knowledge questions, which includes brand facts, pricing, and product details, the average hallucination rate across AI models is approximately 9.2% (ZipTie/AllAboutAI, 2025). That is roughly 1 in 11 AI-generated responses about your company containing fabricated or incorrect information. And the damage extends beyond inaccuracy. When AI states wrong pricing, it creates mismatched expectations your sales team has to correct. When AI fabricates a founding date or employee count, it undermines trust with buyers who fact-check. When AI confuses you with a competitor, the customer may never contact you at all.

The instinct is to contact the AI Company and demand a correction. That instinct is understandable but largely ineffective. ChatGPT does not maintain a database of company facts that you can update. It synthesizes information from training data and web retrieval. Fixing the AI's output requires fixing the sources the AI learns from. The community consensus from practitioners who have successfully corrected AI misinformation is clear: fix the sources, and AI eventually follows (Am I Cited, 2026).

How do you find every error AI is making about your business?

Before you can fix anything, you need a complete inventory of what is wrong. Block out two hours for a thorough audit across every major platform.

Create a test prompt list. Write 15 to 20 prompts that cover every aspect of your business a consumer might ask about. Include basic queries: "What is [Your Business Name]?" and "What services does [Your Business Name] offer?" Add specific questions about pricing, location, hours, team members, service areas, specialties, and comparisons to competitors. These should be the exact questions your real customers would ask.

Test across all major platforms. Run your full prompt list across ChatGPT, Perplexity, Gemini, Claude, and Microsoft Copilot. Each platform uses different training data and retrieval methods, so errors will vary. Metricus found that error types cluster by platform: ChatGPT tends toward feature conflation and fabricated details because it relies heavily on training data, Perplexity tends toward outdated pricing because it pulls from stale web sources, and Gemini shows higher rates of competitive misattribution (Metricus, 2026). You need the full picture from every platform, not just one.

Document everything in a spreadsheet. Create columns for the platform, the prompt, the incorrect claim, the correct information, and a severity rating. Severity matters for prioritization. An AI saying you close at 5 PM when you close at 6 PM is an annoyance. An AI telling customers you went out of business when you are thriving is an emergency.

Trace each error to its source. This is the most important step. Search the web for the exact incorrect phrasing the AI used. Often you will find the source: an old blog post, a stale directory listing, a competitor comparison article, a press release from before your rebrand, or an outdated third-party profile. When you find the source, you have found the fix. If the incorrect information only exists in outdated training data with no current web source, the fix is different: you need to flood the web with current, correct information that future model training and web retrieval will find.

How do you fix the sources AI is learning from?

The core principle: you cannot directly change what AI says, but you can change what AI learns from. Here is the practical framework.

Fix your own properties first. Update your website, Google Business Profile, and every platform you directly control with accurate, current information. Create a "Company Facts" page on your website with clear, structured data: business name, location, services, founding date, team size, service area, and any other facts the AI got wrong. Format this page with tables and clear headings. Add Organization schema markup that explicitly states every fact you want AI to get right. This single authoritative page becomes a reference that AI systems will cite over piecing together information from multiple conflicting articles.

Fix third-party listings. Audit every directory listing, review platform, and third-party profile where your business appears. Correct every piece of outdated or inaccurate information. If Yelp says you are at an old address, update it. If an industry directory lists services you no longer offer, remove them. If an old press release on a wire service still shows pre-rebrand information, issue a new one with current facts. Every conflicting source you leave uncorrected gives AI a reason to state the wrong information.

Request corrections on third-party content you do not control. If a comparison article or industry publication has wrong information about your business, contact the author or editor with a polite correction request. Provide a clear factsheet with the accurate details and a link to your Company Facts page as a source. Many publications will update their content when contacted professionally.

Publish new, authoritative content with correct information. Press releases distributed to credible publications create recent, authoritative sources that AI retrieval systems find and cite. One practitioner documented that distributing press releases with correct company information, combined with pitching local business publications for coverage, led to measurable improvements in AI accuracy within six weeks (Am I Cited, 2026). Each new article containing correct facts pushes the incorrect information further down in the AI's confidence calculation.

Consider implementing an llms.txt file. This is a relatively new standard, a Markdown file placed at your website's root directory that provides AI models with structured, authoritative brand content. OpenAI and Perplexity support it as a content access protocol. It does not guarantee AI will use it, but it reduces ambiguity about what your brand considers official information (ZipTie, 2026).

How do you report errors directly to AI platforms?

Direct reporting has mixed effectiveness, but it is worth doing alongside source-level fixes.

ChatGPT: Use the thumbs down button on any response containing incorrect information. Select "This is harmful or unsafe" and explain the specific error with correct details. If you are a ChatGPT Enterprise customer, use your dedicated support channel to escalate.

Perplexity: Perplexity has a feedback option on every response. Because Perplexity uses real-time retrieval rather than static training data, updating your sources often produces faster corrections on Perplexity than on other platforms. If your website and listings are correct, Perplexity should reflect the changes relatively quickly.

Gemini and Claude: Both have feedback mechanisms in their interfaces. Submit corrections with specific details about what is wrong and what the correct information is, with links to authoritative sources.

Direct reporting tells AI companies that their models are spreading inaccurate information about a specific business. It helps, but it does not replace the more effective strategy of fixing the underlying web sources that feed these platforms.

How long does it take to correct AI misinformation?

Meaningful corrections typically take two to six months. The timeline depends on the platform and the type of error.

Perplexity corrections happen fastest because Perplexity uses real-time web retrieval. Once your sources are correct, Perplexity can reflect the changes within days to weeks.

ChatGPT corrections take longest because ChatGPT relies more on training data that is updated on a model refresh schedule. Even with ChatGPT's web browsing capabilities, deeply embedded training data errors may persist until the next model update. Source-level fixes improve ChatGPT's browsing-based responses faster than its training-data-based responses.

Google AI Overview and Gemini corrections fall in between. Google has direct access to its own index, so updating your Google Business Profile and website can influence Gemini and AI Overviews relatively quickly.

One critical warning from practitioners: AI misinformation correction is not a one-time project. It is ongoing reputation management. One monitoring specialist reported that after successfully correcting misinformation for a client, ChatGPT started stating wrong information again four months later because it ingested a new article that contained old data (Am I Cited, 2026). You need to monitor AI responses about your business on an ongoing basis and catch new errors before they become entrenched.

What is the real business cost of AI misinformation?

The cost is direct and measurable. One Reddit user shared losing a sales opportunity because a prospect encountered false product information on ChatGPT: the prospect's expectations were already misaligned before the sales call, and the correction damaged trust that could not be rebuilt (ZipTie, 2026). Multiply that across every prospect who encounters incorrect information and either calls a competitor or arrives with wrong expectations that your team has to spend time correcting.

The Air Canada case established legal precedent that companies can be held liable for information their AI systems state, even when fabricated (Yoast/ZipTie, 2026). While this specific case involved a company's own chatbot, it signals the legal environment around AI accuracy is tightening. U.S. courts issued 37 AI hallucination rulings in 2024, 73 in the first five months of 2025, and 50 or more in July 2025 alone (ZipTie, 2026). The trajectory is clear.

Fixing AI misinformation is not optional reputation management. It is a revenue protection strategy and increasingly a legal prudence one.

Frequently Asked Questions