You ask ChatGPT, or Google Search with its AI answer: “How much does service X cost at [your company] and until what time are they open?” The answer comes back smooth, with numbers and not a single “maybe”. Except the numbers may not be yours. And a customer who asked the same thing will follow that answer: call, show up or, the other way round, stay away.
Below we explain what an AI “hallucination” is, how it differs from repeating someone else’s fabrication, what has actually happened to companies, and what a small business owner should check personally in 5 to 15 minutes. We checked the facts against the BBC Future article, a Canadian tribunal decision and news coverage. Whatever we could not open at the original source, we say so plainly.
What a “hallucination” is, in plain words
A hallucination is when AI confidently states something that is not true. The reason lies in how it works. A chatbot does not pull a fact out of a database. It picks words that usually follow one another. Google’s guidance on AI-written text says the same: the model “predicts a likely sequence of words”. When it lacks the exact information, it does not always say “I don’t know”. It may put together a plausible answer.
That is why a fabrication sounds convincing: the same words and the same confident tone as a correct answer. You cannot tell them apart by tone. You can only tell them apart by checking against a document.
Two different cases, and it matters not to mix them up
First: AI invents it by itself. It has no facts, or has put them together wrongly, and fills the gap with plausible text. This is a hallucination in the narrow sense.
Second: AI repeats someone else’s fabrication. Modern AI assistants can search the internet and retell what they find. If something false sits on the web, they may repeat it as fact. The mistake here does not come from the AI’s “head” but from a page it read. For a business owner the consequence is the same, but the cause is fixed differently.
The BBC journalist’s story below belongs to the second case. To be fair: it is not a hallucination in the narrow sense.
Examples
The BBC journalist and the hot dogs (repeating someone else’s fabrication). Journalist Thomas Germain, BBC Future, 18 February 2026. On his personal website he posted an invented article with a “ranking” of tech journalists by hot dog eating. Preparing it took about 20 minutes. By his account, in less than a day ChatGPT, the Gemini app and AI Overviews in Google Search began repeating the invention.
Here the sources differ, so we give both sides. According to BBC Mundo, which retells the article, Claude did not repeat the fabrication. One Ukrainian retelling lacks this detail, and from it you could think that all of them fell for it. We rely on the Spanish-language version of the BBC article and on write-ups that retell it. Google and OpenAI responded, per the same publication, that they are working on protection and that their tools “can make mistakes”.
Air Canada (AI invents it by itself). In February 2024 a Canadian tribunal in Moffatt v. Air Canada (decision 2024 BCCRT 149) heard a passenger’s complaint. The chatbot on the airline’s website told him a bereavement discount could be claimed after the flight. The airline’s rules did not allow that. The tribunal ordered Air Canada to pay the passenger. The amounts differ across sources: some give 650.88 Canadian dollars (damages), others about 812 (including interest and fees). We have not read the text of the decision itself, so we do not claim which figure is exact. The substance does not change: the company was responsible for what its bot wrote on its own website.
Wolf River Electric v. Google (a lawsuit, not a verdict). The US solar company Wolf River Electric sued Google in March 2025. According to the complaint, an AI Overview in search wrote that the company was being sued by a state attorney general, although there was no such lawsuit against it. This is only a lawsuit with no ruling on the merits, so we do not claim that Google is at fault. Sources also disagree on which Google product gave that answer: AI Overviews or Gemini.
The Cursor support bot (AI invents it by itself). In April 2025 the technical support bot of the Cursor code editor told a customer that, under the rules, one subscription could be used on only one device. A co-founder of the company replied that there is no such rule. In other words, the bot invented a term of its own service.
What this means for a small business
Only you know your real prices and terms. What can go wrong when a customer asks an AI:
- prices. The assistant names an old, someone else’s or “market average” figure, and the customer arrives with it.
- hours. “Open until 9 pm” when you close at 6 pm, or “open on Sunday”.
- terms. Warranty, free delivery, returns, prepayment: something that sounds plausible but is different at your company.
- reputation. Something foreign or non-existent attributed to you: from a service you do not offer to a complaint that concerns another company.
We do not know how often this happens specifically to small businesses in Ukraine: there is no reliable statistic, and the loud figures about the percentage of wrong answers come from vendors selling related services, so we do not quote them. But the cases above show it is possible. Moreover, Air Canada showed that “the bot wrote it” does not remove responsibility.
What to do yourself in 5 to 15 minutes
This is our own recommendation from practice. The BBC article contains no advice for businesses, and we do not attribute any to the sources.
- Ask AI about yourself. Open ChatGPT, Gemini and Google Search and enter 3 to 5 queries the way a customer would: “what company is [name]”, “how much does [your service] cost in [city]”, “until when is [name] open”, “what are the reviews of [name]”.
- Check against your documents. Open your price list and schedule next to the answer. Compare every figure, address, hour and term. Write down the mismatches: what exactly and in which assistant.
- Ask the same about a competitor. See what they say about them and from which pages (where the assistant shows sources). It shows where your site loses: perhaps the competitor has a page with prices and you do not.
- Find the source of the error. If the assistant shows links, follow them. It may be an old price from an aggregator, a review site or your own old page.
- Put the fact pages in order. Practitioners advise closing such gaps with short, precise pages: “Prices”, “How we work”, “FAQ”, rather than long texts. We recommend this too: your site should carry current prices (at least “from”), hours, address, and the terms of delivery, warranty and returns. Check the same in your business listing in Google Maps (6 things to check in your listing).
- Repeat once a month. AI answers and sources change, so a one-off check goes stale quickly.
⚠ An honest limit: we do not promise that a fix on your site will change the AI’s answer right away. Assistants update at different speeds, and some do not read fresh pages at all. But a current site gives them the right source, and that is the only thing you control. How long it takes depends on the assistant; we do not know of a guaranteed timeframe.
What not to trust AI assistants on without checking
- prices and discounts, especially other companies’ and “market averages”;
- legal terms: warranties, returns, contracts;
- any exact figures and dates you have not seen in a document;
- links and “sources”: open them and see whether it really says that;
- a bot in the chat on your site: before leaving it to answer customers, test it on 10 to 15 real questions. We wrote about an AI agent for phone calls in the article on calls after 6 pm.
The same applies to website texts written by AI. What to check before publishing is set out in the article AI writes your website text: what Google says to check by hand.
Summary
An AI assistant may invent your prices, hours or terms on its own, or pick up someone else’s falsehood from the internet, as in the BBC journalist’s case. It sounds equally confident either way. The Air Canada case showed that a company answers for what its bot says on its own website. And the Cursor case and the Wolf River Electric lawsuit against Google are a reminder that the customer sees the error before you do. So ask AI about your business, compare its answers with your documents, keep exact prices, hours and terms on your site, and repeat the check once a month.
How we run advertising for small businesses is in the section on advertising for local business.
Want a free audit of your Google Ads account? Send us a request: we will look at your account and tell you what is worth fixing.