Customer Impact

SEO & GEO

What is an AI hallucination and why should your brand take it seriously?

Copy for AI

An AI hallucination is a confident AI answer that’s factually wrong or made up.

The model sounds sure of itself, phrases things fluently and presents the error as an established fact.

The verdict: for your brand, an AI hallucination is a sales problem you can’t see. You don’t fix it by correcting the AI, but by leaving it nothing to invent.

When ChatGPT, Gemini or Perplexity invents a wrong price, a non-existent service or wrong contact details, the prospect hears it as the truth.

Below: what a hallucination is, why AI produces them, what brand damage they cause and how to reduce the risk.

What is an AI hallucination exactly?

An AI hallucination is text that sounds plausible but doesn’t match reality.

A misleading metaphor

The word “hallucination” is a somewhat unfortunate metaphor. The model doesn’t see things that aren’t there.

A large language model simply predicts the most likely next word, based on patterns in its training data.

If those patterns don’t produce an exact answer, the model fills the gap. It writes something that looks like a good answer.

The tone gives nothing away

The treacherous part is the tone.

A human who doesn’t know something hesitates audibly. A language model hallucinates with the same fluent confidence it uses for correct facts.

There’s no warning light attached.

Why does AI hallucinate in the first place?

AI hallucinates because a language model is built to predict language, not to look up the truth.

Prediction, not lookup

During training, the model learned billions of text patterns. It answers by choosing the statistically most likely word each time.

It has no built-in sense of “I know this” versus “I don’t know this”.

Training rewards guessing

OpenAI researchers go a step further. In a September 2025 paper, they argue models hallucinate because training and evaluation reward guessing over admitting uncertainty.

So faced with a question about a niche brand, it still picks the most likely answer, even when it’s wrong. That happens most when the training data held few or contradictory sources.

When the risk goes up

Hallucinations become more likely in three situations:

AI HALLUCINATION RISK When does the model start guessing? Little public information a smaller or niche B2B company Gaps get filled with likely text Specific, fast-changing facts prices, dates, figures Old or invented numbers appear Contradictory info online old pages, directories, mixed names The model picks one version at random The less clear your public story, the more the model improvises.

The common thread is an information vacuum.

Our experience: if you can’t tell your positioning clearly, the model fills the gap and starts to hallucinate.

Grounding: live search or memory

That’s why it matters when a model searches the web live and when it answers purely from memory.

That difference is called grounding. It largely decides whether your facts get checked.

Without grounding, the model leans entirely on its possibly flawed memory. That’s when an invented brand fact is most likely.

Here’s an answer with no grounding visible. ChatGPT’s free plan gave three fluent bullets of about 90 words, but no sources:

ChatGPT free answer to a marketing prompt with three fluent bullets and no sources, an example of an ungrounded answer where an AI hallucination can go unnoticed ChatGPT free (temporary chat), 17 October 2026: three bullets, about 90 words, no sources.

There’s nothing to check against. Compare Perplexity, which listed the 10 web pages behind its answer:

Perplexity sources panel listing the 10 web pages behind one answer, a grounded AI answer that lowers the chance of an LLM hallucination Perplexity’s sources panel for our own test prompt, 17 October 2026: 10 web pages behind one answer.

A grounded answer can still be wrong. But at least you can see where it came from, and which page to fix.

Why is an AI hallucination a brand risk for your company?

AI has become one of the first places where prospects research your company. What it says there shapes your reputation, and you aren’t in the room.

The new research step

In the past, a prospect read your own website or a review.

Now that same prospect asks ChatGPT “is this company reliable?” or “what does service X cost with them?”. They get back a synthesis they trust as objective advice.

You don’t see the errors

You correct a wrong price on your own website in five minutes.

But an AI that tells thousands of users a wrong price about you does it invisibly. It happens in private conversations you never see.

You only notice the damage indirectly:

  • a lead who drops off over a price you never quoted
  • a prospect who expects a service you don’t offer
  • a call to a phone number that’s no longer yours

Leads, not vanity scores

At Customer Impact, we look at AI visibility from the angle of leads and revenue, not vanity scores.

That’s also why AI brand reputation management has become part of SEO work, not a side project.

What damage can a hallucination about your brand cause?

The damage ranges from annoying to downright costly.

These are the most common forms.

Wrong prices or terms

The AI names a price you never use, or a delivery time that’s wrong.

That sets an expectation you can’t meet.

Invented or missing services

The model attributes a service to you that you don’t provide.

Or it leaves out your core offering, because it found no strong source about it.

Incorrect contact or company details

An outdated address, a wrong phone number or a non-existent location.

It literally sends prospects the wrong way.

Confusion with a competitor

The model mixes your brand with a company of the same or a similar name.

It then attributes someone else’s characteristics to you, or the other way round.

Invented reputation elements

Made-up reviews, non-existent awards or a wrong founding year.

They undermine your credibility the moment someone checks them. Our guide to AI brand mention sentiment shows how to track the tone of those mentions.

Why these errors happen

What these cases share: they rarely come from ill will, but from an information vacuum.

Where you’ve left no clear, consistent and authoritative signal, the model fills the void itself.

Who is liable if AI says something wrong about your brand?

Companies can indeed be held responsible for wrong information that comes out via AI.

The Air Canada precedent

The best-known precedent is the Moffatt versus Air Canada case (2024 BCCRT 149).

In February 2024, British Columbia’s Civil Resolution Tribunal ruled that Air Canada was liable. Its own chatbot had given a customer wrong information about bereavement fares.

That chat took place in November 2022, so the bill arrived well over a year later.

Here’s how it played out:

ElementWhat happened
The chatbot’s answerWrong information on bereavement fares
Air Canada’s defenceThe chatbot was “a separate entity”
The tribunal’s findingNegligent misrepresentation
The awardPartial refund of CAD 650.88, plus interest and fees

The “separate entity” argument didn’t hold up. The tribunal found Air Canada “did not take reasonable care to ensure its chatbot was accurate”.

A refund of CAD 650.88 (about USD 482) is small money, but a big signal: the company stayed responsible for all information on its channels.

What it means for third-party models

That case was about a chatbot on the company’s own site, not an external model like ChatGPT. So the legal line to third-party hallucinations is less direct.

But the direction is clear. As AI becomes a fixed part of how customers get information, the expectation grows that a brand keeps its own facts in order.

Waiting until this is fully settled isn’t a strategy. Take responsibility now for the information circulating about you.

How do you reduce the chance that AI hallucinates about your brand?

Not by correcting the model afterwards, but by putting so much clear, consistent information online that it has nothing to invent.

A model guesses least about brands for which it finds the same, unambiguous story everywhere.

Three levers work strongest.

Lever 1: Entity consistency

Make sure your company name, services, location and core facts are identical everywhere. That means your site, directories, LinkedIn and mentions elsewhere.

Contradictory data is a direct invitation to guess. Our guide to entity consistency shows where to check.

Tip: add Organization structured data to your site. Google says it helps it understand your administrative details and disambiguate your organization from others, with properties such as address, telephone and sameAs.

That feeds the same entity picture behind Google’s knowledge graph.

These are the facts to keep identical everywhere:

Brand factWhere it must match
Company name and legal nameSite, LinkedIn, directories, invoices
Address and phone numberSite footer, Google Business Profile, directories
Services you offerService pages, LinkedIn, partner listings
Pricing structureYour own pricing or quote page
Founding year and team sizeAbout page, LinkedIn

Watch out: an old directory listing with your previous address is enough to give a model two versions to choose from.

Lever 2: Brand mentions in authoritative places

The more often reliable sources describe your brand correctly, the more firmly the right story sits in the models.

In AI visibility, those mentions often weigh more heavily than classic backlinks.

Lever 3: Be the best source yourself

Write your core facts, pricing structure and services explicitly, clearly and extractably on your own site.

Then a model that grounds finds your version instead of improvising.

This is exactly what our work around AI findability aims at: making your brand the reliable source that AI falls back on.

We promise no guarantees on positions, because nobody can honestly give those. But we systematically reduce the risk of invention.

How do you monitor hallucinations about your brand?

By regularly asking the questions your prospects ask, and checking what the AI answers.

Run a recurring prompt check

Ask ChatGPT, Gemini and Perplexity questions such as:

  • What does [your company] do?
  • What does [your service] cost with them?
  • Is [your company] reliable?

Note every deviation from the facts. Do it recurrently, because models and their sources change.

The engines don’t all answer the same way, as our comparison of ChatGPT vs Claude vs Gemini shows.

MONITORING Catch hallucinations in time repeat & refine 01 Ask prospect questions ChatGPT, Gemini, Perplexity → 02 Check the answer Note every discrepancy → 03 Correct the facts Before a prospect acts on it Models and their sources change, so do this recurrently.
Monitoring your brand facts in AI is a recurring cycle, not a one-time check.

Use a tool when the prompt list grows

For a structured approach, there are tools that track brand mentions and factual accuracy in AI answers.

Otterly.AI, for example, lists the engines it monitors right on its homepage:

Otterly.AI homepage listing the AI engines it monitors for brand mentions, including ChatGPT, Google AI Overviews, Perplexity and Copilot, a way to catch AI hallucinations about your brand Otterly.AI’s homepage, October 2026, listing the AI engines it monitors.

Peec AI shows the same idea as a dashboard, with visibility and sentiment per competitor:

Peec AI homepage with a demo dashboard comparing brand visibility, sentiment and position against competitors in AI answers Peec AI’s homepage, October 2026. Visibility, sentiment and position sit next to each competitor.

The goal isn’t to chase every nuance. It’s to catch the errors that can tip a buying decision, and fix the underlying information before a prospect acts on it.

Prefer a quick one-off check? Our AI visibility checker or a full AI visibility audit is a good start.

The broader framework, from fundamentals to measurement, is in our complete guide to GEO.

Frequently asked questions about AI hallucinations

Can you stop an AI hallucination about your brand completely?

No. No one controls what a model generates. You can lower the risk sharply by keeping your facts consistent, clear and well sourced across the web, and by monitoring the answers.

Does correcting ChatGPT in a chat fix the error?

Not for other users. A correction in your own conversation doesn’t change the model’s training or the sources it finds for someone else. Fix the information at the source instead.

Is an LLM hallucination the same as a lie?

No. A lie needs intent. A hallucination is a side effect of predicting likely text without checking it against facts.

The short summary

An AI hallucination is a confident but wrong answer. It arises because a language model fills gaps with likely text.

For your brand, that’s a real risk: AI tells prospects wrong prices, invented services or wrong details, invisibly and with authority.

You fight it not by correcting afterwards, but by putting consistent, authoritative information online and monitoring your brand facts in AI.

That lowers the chance a model invents something, and raises the chance it tells your correct story.

Want to know what AI says about your company today? Contact us and we’ll run the check with you.

Free website scan

Enter your website and get an automatic scan within minutes, with concrete technical and SEO improvements. No sales pitch.

Where should we send your report?

We only use your details for your scan. No spam, unsubscribe anytime.