AI hallucination, in a GTM research context, is a research agent stating an invented fact, like a funding amount or investor name, with the same flat confidence as a verified one, because the model is pattern-completing an answer rather than retrieving it from a real source.
An LLM asked a question with no grounding data doesn't reach for a 'don't know' button by default. It doesn't have one built in. What it has instead is a next-token prediction problem: given everything it's seen about, say, funding announcements, what's the most statistically plausible completion for a company that looks like this one. So it pattern-completes. A dollar amount sized to the company's stage. An investor pulled from firms that fund companies like it. None of it retrieved. All of it generated to sound retrieved.
Contrast that with a properly grounded call: the agent fetches a specific page, finds a specific sentence, and reports what that sentence says. That's retrieval, a different operation entirely, and most agents don't distinguish between the two modes in their own output. A pattern-completed guess and a retrieved fact come back in the same confident sentence, same punctuation, same tone. Nothing in the raw text tells a reader which one they're looking at.
An agent never told the difference between 'I found this' and 'this seems likely' will keep collapsing the two. That makes hallucination a prompting problem as much as a model problem: instruct the agent to say UNVERIFIABLE instead of guessing, require a quote before a claim counts, and the rate of confident invention drops without needing a different model at all.
Test any research prompt by asking what it does when no source supports an answer. If the prompt has no instruction to say so, the agent will invent something plausible instead, because that's the default behavior of a model with no permission to come back empty.
Hedged language gets treated as a safer middle ground between a confident fact and an admitted gap. It's often worse than a flat guess, because a phrase like 'likely raised' or 'probably around' reads as careful while doing the exact same job as an outright invention: filling a gap with words instead of admitting the gap is there.
Tell us how your motion runs today. We'll show you what we'd engineer.
Contact us