Key Takeaways
- Learn what the word means before you worry about it. An AI hallucination is an answer that states something untrue or unsupported in the same confident voice as everything else. The assistant is not lying. It is completing a sentence.
- Stop reading confidence as evidence. An assistant sounds equally sure of a fact stated on three pages and a detail stated on none. Tone tells you nothing about what stands behind a sentence.
- Expect the gaps in your record to be the risk. Invented details tend to land where the public record is silent: price, founding year, program length, credentials. A blank is an invitation to guess.
- Separate invented from stale and confused. Most wrong answers about a business trace back to a real page that is old or about someone else. Only a detail with no page behind it is a hallucination.
- Publish the facts you least want guessed. You cannot switch hallucination off. You can state your price range, your start year, your credential and your program plainly, so there is something true to repeat.
A prospect tells you they read that your program runs twelve weeks and costs $4,800. You have never offered a twelve-week program. They did not get it from your website. They got it from an assistant, which said it as plainly as it said your name.
What is an AI hallucination? It is an answer in which an assistant states something untrue or unsupported in the same confident voice it uses for facts. And yes, it can happen to your business, most often in the places where your public record says nothing.
Below is the short answer, why an assistant makes things up, what a hallucination about a business looks like, which businesses it happens to, and how to lower the odds.
What is an AI hallucination?
An AI hallucination is an answer in which an assistant states something that is untrue, or that nothing supports, in the same fluent and confident voice it uses for everything else. Nothing in the wording marks it as a guess.
The companies that make these assistants use the word themselves. OpenAI's help page Does ChatGPT tell the truth? describes a hallucination as the model producing responses that are not factually accurate, and lists three kinds: incorrect definitions, dates or facts; fabricated quotes, studies, citations or references to sources that do not exist; and overconfident answers to ambiguous or complex questions.
Anthropic's guidance for people building on Claude says much the same. Its page on reducing hallucinations opens by saying that even the most advanced language models can sometimes generate text that is factually incorrect. The word is jargon, but it is the makers' own jargon, and it names a limit they state openly.
Why an assistant makes things up
An assistant makes things up because its job is to produce a likely next sentence, not to look up a record. When it has a source, the likely sentence and the true sentence are usually the same one. When it has none, it still produces a sentence, and that sentence is built from what is typical.
Think about what "typical" means for a coaching practice. Programs often run twelve weeks. Coaches often hold a credential. Practices are often described as founded in some particular year. An assistant that has read thousands of coach pages knows the shape of the paragraph, and when your details are missing it can fill the shape in.
Two things make this hard to spot:
- The voice does not change. OpenAI's page puts it as "confidence isn't reliability": the model may express high confidence even in incorrect answers.
- The invented detail is plausible. It is not a wild claim. It is the ordinary thing a business like yours might well have said.
Assistants that search the web while they answer have something real to write from, which helps. The difference between an answer written from pages and one written from memory is covered in what grounding is and why some AI answers cite sources. It lowers the rate. It does not bring it to zero, because an assistant can still add a sentence that none of the pages it opened contains.
What a hallucination about a business looks like
A hallucination about a business usually looks like one or two invented details sitting inside an otherwise accurate answer. The name, the city and the specialty are right, which is exactly why the invented price or founding year gets believed.
Asked: "Tell me about Northgate Coaching." The answer says Northgate Coaching...
- is an executive coach in Denver How sure it sounds Certain Pages that state it 3 of 3. Website, LinkedIn, a directory listing
- works with first-time managers How sure it sounds Certain Pages that state it 2 of 3. Website, LinkedIn
- holds an ICF credential How sure it sounds Certain Pages that state it 1 of 3. Website only
- has been in practice since 2009 How sure it sounds Certain Pages that state it 0 of 3. Nowhere. Filled in.
- runs a 12-week program for $4,800 How sure it sounds Certain Pages that state it 0 of 3. Nowhere. Filled in.
All five statements sound certain. Three have a page behind them and two have none. From the wording alone you cannot tell which two.
The example is illustrative. Northgate Coaching is a made-up name and not a client. Of the five statements in the answer, three have public pages behind them: an executive coach in Denver (three pages: website, LinkedIn and a directory listing), works with first-time managers (two pages: website and LinkedIn), and holds an ICF credential (one page, the website). Two have none: in practice since 2009, and a 12-week program for $4,800. Every one of the five is worded with the same certainty.
Invented details about a business come in a few recognizable kinds. The table sets them beside the three kinds OpenAI lists.
| Kind | Example | Why it happens | What lowers the odds |
|---|---|---|---|
| Invented detail about a real business | A price, founding year or program length you never published | Your record is silent, so the typical value is filled in | State the fact plainly on your own site and profiles |
| Invented source or quote | A review, article or link that does not exist | The answer has the shape of a cited claim with nothing to cite | Publish real reviews and results where they can be read as text |
| Invented business | A plausible practice name in a list of recommendations that turns out not to exist | Few real names are known for a narrow question | Be a real, well-described name for that narrow question |
| Overconfident answer to a vague question | "The best coach in Denver is..." stated as settled fact | The question has no single answer and one is given anyway | Nothing on your side. Treat single-name answers to broad questions with care |
Can it happen to your business?
Yes, it can happen to your business, and the odds rise with how little is published about you. An assistant with a full record to draw on has less reason to guess. One with only a name and a specialty has most of the paragraph still to write.
That gives a small practice a different risk than a large one. A national brand is described on hundreds of pages, so an invented detail is contradicted by the record. A solo coach may be described on three. Four signs that you are more exposed:
- Your price, or even a price range, is not stated anywhere.
- Your site says what you believe but not what a client actually gets, for how long.
- Your credential appears as letters after your name and is never spelled out.
- Your profiles disagree with each other, so there is no single version to repeat.
Before you call something a hallucination, check that it is one. Most wrong answers about a business have a real page behind them. The page is out of date, or it describes a different business with a similar name. Those are stale and confused answers, and the fix for them is to correct or separate the source. A hallucination is what is left when you have looked and no page says it at all.
The sorting matters because assistants lean on agreement between pages rather than checking facts one by one, as explained in how AI assistants verify what's true about your business. Where several pages agree, the answer follows them. Where there are none, there is nothing to follow.
How to lower the odds, and what to do when it happens
Lower the odds by publishing the facts you least want guessed, in plain sentences, in more than one place. You cannot switch hallucination off, and nobody honest will sell you that. You can leave fewer blanks.
Work through it in this order:
- Ask two or three assistants about your business by name. Go through each answer a sentence at a time and look for the page that states it.
- List the sentences with no page behind them. Those are your blanks: usually price, dates, program length and credentials.
- Write one plain sentence for each blank and put it on your site as text, not inside an image or a PDF.
- Repeat the same facts on the profiles you control, worded the same way.
- Ask again in a few weeks. Answers vary from run to run, so ask each question twice before you conclude anything.
If the invented detail is harmless, such as a founding year that is off, publishing the true one is the whole job. If it is costing you clients or misstates a credential, treat it as a correction to manage, and the steps for that are in what to do when AI says something wrong about you.
One honest limit. Publishing a fact does not guarantee an assistant repeats it, and nobody can give you a date by which an answer will change. Assistants that search as they answer can pick up a corrected page once it has been found again. Answers written from what a model learned in training can lag for months.
Checking what is said about you by name is a test you run yourself. The free check covers the other half: whether you are named at all when buyers ask without using your name. It asks ChatGPT 25 real buyer questions from your niche at $0 and emails a plain-English readout: your score, who got named in your place, and the one fix to start with. It runs as soon as you ask for it, and the report lands within the hour between 7am and 5pm Central, or by 10am the next morning outside those hours.
Questions we hear the most
What is an AI hallucination?
An AI hallucination is an answer in which an assistant states something that is untrue or has nothing behind it, in the same confident voice it uses for facts. Typical examples are a wrong date, an invented quote or a link to a source that does not exist.
Can AI make up things about my business?
Yes. An assistant can state a price, a founding year, a credential or a service that appears on no page anywhere. It is most likely where your public record says nothing, because the assistant fills the gap with what is typical for businesses like yours.
Why does AI sound so confident when it is wrong?
Because the confident voice is how it writes everything. An assistant produces fluent sentences whether or not it has a source for them, so tone does not drop when support does. OpenAI says the model may express high confidence even in incorrect answers.
Is every wrong AI answer a hallucination?
No. Many wrong answers come from a real page that is out of date, or from a page about a different business with a similar name. A hallucination is the case where no page says it at all. The three need different fixes.
Can I stop AI from hallucinating about my business?
No, not completely, and nobody can promise it. What you control is how much there is to guess. Stating your basic facts plainly, on your own site and on your profiles, leaves fewer blanks for an assistant to fill.
How do I check whether AI is making things up about me?
Ask two or three assistants about your business by name, then go through each answer one sentence at a time and look for the page that states it. A sentence you cannot find on any page, yours or anyone else, is the one to note.
Does the AnswerHalo free check look for hallucinations?
Not as such. The free check asks ChatGPT 25 real buyer questions from your niche at $0 and emails your score, who got named in your place and the one fix to start with. It measures whether you are named, not whether each sentence about you is accurate.
How fast does the free check report arrive?
It runs as soon as you ask for it. The report lands within the hour between 7am and 5pm Central, or by 10am the next morning outside those hours. It arrives by email, and there is nothing to log in to.
Your buyers are already asking. Find out what AI tells them.
The free check asks ChatGPT 25 real buyer questions about your niche and sends you the report within the hour, at $0. If it shows you are already getting named everywhere, we will say so in plain words.
Get my free check →