Anyone can wire an LLM to a webhook. The difference between a demo and a pipeline you can trust is what happens when the model is wrong. Gatework templates ship with verification gates — adversarial checks, fail-closed parsing, false-positive detection — extracted from production systems that processed thousands of real records. Free templates do the job. Pro versions do it on a schedule, at scale, with alerts.
Bulk-checks a lead list's websites and classifies each one — live, parked, dead, or unreachable — with page titles extracted along the way. No credentials required; runs on a pasted list.
Form-triggered web research: generates search queries, gathers sources, extracts the claims that matter — then a second model pass tries to refute every claim against its own source before anything reaches the report.
Every template starts from a failure that actually happened in a production pipeline — and ships with the check that would have caught it.
A naive checker marks every non-200 response as a dead lead. But a security layer blocking your bot returns 403 for a business that's very much open. Whole segments of a list get discarded for no reason.
gate: waf-signature detectionAn LLM summarizing web sources will, some fraction of the time, state things the sources never said — fluently, with conviction. If nothing checks the claim against the source, it lands in your report as fact.
gate: adversarial second-pass verificationA model returns almost-JSON, the workflow shrugs and moves on, and a malformed verdict gets treated as approval. Anything a parser can't read should count against the claim, not for it.
gate: fail-closed parsingPro versions run the same verified pipelines unattended — scheduled re-checks, failure alerting, and batch scale. They're in build. The waitlist is the only list we keep.