When ChatGPT and Gemini Get Your Business Wrong (And What to Do)
Key takeaways
- ChatGPT, Gemini, and Perplexity confidently tell customers wrong things about your business — wrong hours, wrong prices, wrong services, wrong locations.
- Hallucinations happen most when your online data is thin, contradictory, or years out of date.
- The fix isn’t “prompt engineering.” It’s cleaning up the sources the AI reads: GBP, your website, industry directories.
- You can’t complain your way out of this. AI providers don’t have customer service for local businesses; they update from public sources.
- Regular AI-audit: monthly, ask each major model about your business. Fix the source of any wrong answer.
Two weeks ago, an East Texas HVAC client called me because ChatGPT was telling potential customers he charged $89 for a service call. He hasn’t charged $89 since 2021. It’s now $135. He’d been losing quotes because callers were showing up expecting the old price and getting sticker shock.
He didn’t have a bug on his website. He had a hallucination in the AI. And it was costing him money on every single call.
Here’s how AI models get your business wrong, why they do it confidently, and what small businesses can actually do about it.
Why AI models hallucinate about small businesses
Large language models don’t have real-time knowledge. They have a training data snapshot plus, in the case of ChatGPT and Gemini, some ability to browse the live web on demand. When you ask “how much does East Texas HVAC Guys charge?” the model tries to answer from one of three sources:
- Memorized training data (which could be years old)
- Live web search (which pulls from indexed pages)
- Structured data from partners (Google Business Profile, Yelp, industry directories)
If your business has a thin online footprint, the model has less to work with. It fills gaps with pattern-matching — “HVAC companies in East Texas typically charge $75-$100 for service calls” — and states it as if it knew your price specifically.
The four common hallucination patterns
1. Stale pricing
Prices from your 2021 blog post live forever in AI training data. If your rates changed but nothing on the current web reflects the new number, the AI answers with the old number. Confidently.
2. Merged businesses
When two similarly-named businesses exist in the same region, the AI sometimes merges them into a single “answer” that combines both. Your address with their reviews. Their phone with your services. It’s a mess and it happens more than you’d think.
3. Fabricated services
“Yes, Smith Auto Shop in Longview does transmission rebuilds.” Except they don’t. The AI is filling in what an auto shop usually does. If your site doesn’t clearly list your actual services, the AI guesses — and guesses too generously.
4. Wrong hours or closure claims
“Bob’s Diner is closed on Mondays.” Except Bob changed to open on Mondays six months ago. Or worse: “Bob’s Diner has permanently closed.” Except Bob is right there, open, wondering why traffic is down 30%.
The AI doesn’t know it’s wrong. It doesn’t know it doesn’t know. And it will not apologize when it costs you a sale.
How to check what AI models say about you
Every month, spend 15 minutes asking the major models about your business:
- ChatGPT: “What can you tell me about [business name] in [city]?”
- ChatGPT: “How much does [business name] charge for [service]?”
- ChatGPT: “What are [business name]’s hours?”
- Gemini: same three questions
- Perplexity: same three questions
- Google AI Overview: search “[business name]” and “[services] near me”
Note every wrong answer. That’s your fix list.
Why “reporting” doesn’t work
None of the AI providers have a “correct my business info” form for small businesses. You cannot email OpenAI and expect a response. You cannot flag Gemini’s answer. The AI will not remember your correction if you tell it in a chat.
What you can do is fix the sources.
The source-cleanup playbook
Step 1: Google Business Profile
This is the most influential single source. Fix:
- Current pricing (as a range or “starting at”) in your description and services
- Complete services list — every service you offer, no filler
- Accurate hours, updated for holidays
- Category matches what you actually do (not “restaurant” if you’re specifically a pizzeria)
Step 2: Your website
Every claim the AI got wrong should be corrected on your website in a place that gets crawled:
- Pricing on a dedicated pricing page (or per-service page)
- Services page with every service listed clearly
- An About page with the founding year, ownership, and business location
- A blog post if there’s a recent change (“We now offer weekend appointments”)
Step 3: Third-party directories
Yelp, Facebook, industry directories (Angi, Thumbtack for services; TripAdvisor for hospitality; Healthgrades for medical). Make sure all major listings agree on:
- Business name (exactly the same, no variations)
- Address
- Phone number
- Hours
- Category
This is called NAP consistency (Name-Address-Phone) and AI models use consensus across directories as a confidence signal.
Step 4: Wikipedia / knowledge graph
Most small businesses aren’t on Wikipedia and shouldn’t try to be. But if your business is notable (long-established, unique, or newsworthy), a knowledge-graph entry substantially reduces hallucinations. This is a project for a specialist.
The time-to-fix reality
Cleaning up your sources doesn’t instantly fix AI hallucinations. Training data snapshots are updated in cycles, and web-search-enhanced answers depend on when the AI last crawled your pages. Typical timelines:
| AI provider | Typical time to reflect a correction |
|---|---|
| Google AI Overview (uses live search) | 1-2 weeks after your site is recrawled |
| ChatGPT with browsing | 2-4 weeks |
| Perplexity | 1-2 weeks (uses live web) |
| Gemini | 2-4 weeks |
| Memorized-only responses | Next major training update (6-12 months) |
Where people go wrong (and when to call a pro)
Two failure modes: (1) getting angry at OpenAI or Google and expecting them to fix it (they won’t, and they can’t at scale), or (2) accepting the hallucinations as immutable and losing money on every quote. Call a pro when you want a real source audit — every listing your business appears on, cross-referenced for consistency, plus a plan to make your website the authoritative source. It’s about 4-6 hours of work initially, then monthly monitoring. For any small business getting real inbound leads, it pays back in days.
Frequently Asked Questions
Can I sue an AI provider for defamation if they say something wrong about my business?
Theoretically yes, in practice almost never worth it. The legal landscape is still forming. You’ll spend more on lawyers than on fixing the sources.
What if the AI is stating a competitor’s info about my business?
Focus on making your own information more visible and consistent. AI models resolve conflicts by picking the source with the strongest, most consistent signal.
Should I use “AI schema” or an “AI file” on my site?
There’s no such thing as an “AI schema” tag. Standard structured data (LocalBusiness, Service, FAQ, Product) is what AI models read. Skip anyone pitching you “AI-specific” markup.
Are hallucinations getting better or worse?
Better slowly. AI providers have improved their citation and source-linking. But hallucinations about small local businesses will persist for years, because thin online footprints will always leave gaps.
Losing customers to AI hallucinations about your business? Let’s fix the sources and monitor the answers.