AI search
Hallucination
A hallucination is a confident, fluent statement from a language model that is not supported by any source.
What is hallucination?
A hallucination is a confident, fluent statement from a language model that is not supported by any source. It matters commercially when a model states something false about a brand’s pricing, services, coverage or credentials.
Why it matters
A wrong fact stated confidently in an AI answer reaches buyers without any chance to correct it in the moment.
How it is actually done
Reduced on the brand’s side by making the correct facts consistent, machine-readable and present on the sources the assistant retrieves from, because a model with no clear source guesses.
Where this gets confused
There is no way to edit an AI answer directly. Correction works indirectly, by fixing the sources the system draws on and waiting for re-retrieval.
The definition is simple; the implementation is where the arguments happen.
Where this comes up in the work
It is part of AI Automation and Workflows, and more specifically of CRM integration. Across the rest of the work it shows up in:
- Paid Social and Meta Ads, which cannot be done properly without it
- Conversion Rate Optimization, where getting it wrong shows up in fixing the slowest page on the revenue path first
- Website Design and Development: it is the reason keeping third-party scripts from degrading a site that was fast on launch day is scoped the way it is
- AI Automation and Workflows, where it is the difference between ending the monthly rebuild of the same report working and appearing to
What to check
Two questions that establish whether this is actually in order on a given site:
- Whether it holds on a mid-range phone and not only on the machine it was built on
- Whether it survives a logged-out visit from a cold cache
Common questions
Is hallucination the same as anything else?
There is no way to edit an AI answer directly. Correction works indirectly, by fixing the sources the system draws on and waiting for re-retrieval. The distinction matters because the two are fixed differently. If the brief is still open, Paid Social and Meta Ads and Technical SEO are the two that change it most often.
Will fixing hallucination move rankings?
A wrong fact stated confidently in an AI answer reaches buyers without any chance to correct it in the moment. The honest framing is that it removes a constraint rather than adding a boost. Server-side rendering or Client-side rendering is the reference page for this. Subdirectory or Subdomain is the one with the examples.
How do you check hallucination on a site?
Reduced on the brand’s side by making the correct facts consistent, machine-readable and present on the sources the assistant retrieves from, because a model with no clear source guesses. The check is cheap; the fix is where the work sits. The specifics live in Are backlinks still important?. Do AI assistants send any traffic? is the context that makes them make sense.
Whose job is hallucination?
Whoever is accountable for turning an internal report into something worth covering. Problems with this tend to persist because nobody is. There is a fuller account in Zero-click search. A/B test is the part that tends to get skipped.
Is hallucination expensive to get right?
It depends entirely on where the problem originates. Establishing that is the first piece of work.
How urgent is hallucination?
High if it is actually broken, low if it is merely imperfect. The check tells you which.
Get a scoped answer rather than a proposal
Share the site and the target. We will say what we would check first and what we would leave alone.