Field notes
An AI-Generated Supplier Shortlist Isn't Sourcing Data — It's a Starting Point
Ask a general-purpose AI model to list manufacturers of a given component and it will answer instantly, in a confident, well-formatted list — company names, locations, sometimes even a product description and a domain. For a sourcing consultant working against a deadline, that's a tempting shortcut past the usual slog of directory searches and trade-show catalogues.
It's also, on its own, not sourcing data. These models are good at pattern-completing what a plausible supplier list looks like for a given product category, which is a different thing from reporting which real companies actually make it. The known failure modes are specific enough to check for directly, the same way we check for any other oversell pattern:
- Invented or merged companies. A name that reads as plausible for the category but doesn't resolve to any real, independently findable business.
- Domain mismatch. A domain the model states with confidence that either doesn't exist, is parked, or belongs to an unrelated business — security researchers have already documented large-scale "hallucinated domain" squatting aimed at exactly this gap between confident AI output and real registration records.
- Scope creep. A real company, but not actually in the product category asked for — the model filled the gap with something adjacent rather than leaving it blank.
- Unsourced specifics. A stated certification, address, or "manufacturer not trader" claim that isn't attributable to anything the company itself published — it reads like a fact because the sentence is fluent, not because it's grounded in a page you could go check.
A starting point, not a finding
None of this makes an AI-generated shortlist useless as a starting point. It makes it exactly as unverified as a cold search-engine hit or a trade-show badge-scan list — a set of names worth checking, not a set of findings. The same evidence discipline applies either way: does the domain resolve to a real, non-parked site for that specific company; does that site's own stated scope actually match what you asked for; does a claimed registration or certification show up on the actual issuing registry rather than just in the company's own copy.
For a sense of how wide the gap between a confident supplier claim and a verified one can be in practice, see our write-up on validating a batch of "manufacturer" names against real evidence.
Evidence, not a verdict
Treat an AI-written list as a set of hypotheses to run through verification — evidence-gathered, human-reviewed — before any name on it reaches a shortlist you'd act on.
Doing supplier verification for clients? Supplier Intel is built for exactly this workflow — get in touch if you'd like to compare notes.