Seven red flags in a prospect enrichment dump that mean rewrite before send
The CSV looks complete. Title, company, funding, tech stack, LinkedIn URL. The model writes a confident first line. The prospect replies: "I left that company nine months ago."
Enrichment dumps fail loudly when you send them. The rewrite gate belongs before send, not after the screenshot lands in a founder Slack.
Explorium's 2026 write-up on prospect data hallucination names the mechanism: empty, stale, or mismatched fields get filled with plausible text because generation hates saying "I don't know." CrustData and CUFinder's enrichment QA pieces keep pointing at the same ops fix: standardize, dedupe, and refuse bad rows before they become copy.
Here are seven red flags that should force a rewrite or a suppress, not a send.
1. Title and seniority disagree with the LinkedIn URL
The dump says "VP Sales." The LinkedIn URL resolves to an individual contributor, or a different person with a similar name. That is unresolved identity wearing a complete-looking row.
Action: Hold send. Re-resolve the person. If match confidence is missing entirely, treat the row as untrusted.
2. Company name and email domain do not belong together
acme.com email with company field set to a different brand, a subsidiary, or a name-collision firm. Explorium calls this unresolved identity: every other field can look populated and still be attached to the wrong business.
Action: Rewrite nothing from firmographics until domain ↔ company is fixed. Prefer omit over invent.
3. Enrichment older than a quarter with no refresh timestamp
Industry decay figures commonly cited around 2–3% per month mean a static import is a different dataset by mid-quarter. Stale titles and phantom funding rounds are the classic outbound failures.
Action: Require enriched_at (or equivalent). Past your freshness SLA, re-enrich or strip claims that age poorly (title, funding, headcount, stack).
4. Funding, executive, or integration claims with a single thin source
One vendor field says "Series B, 2024." No second corroboration. Models love turning that into a personalized opener. Prospects love correcting it in public.
Action: Material claims need provenance. No source trail, no claim in the email. GTM's own product stance is provenance on every field for a reason.
5. Catch-all or role-based email treated like a verified personal inbox
info@, sales@, or a catch-all that "verified" as accept-all. Deliverability and personalization both lie here. You are not writing to the buyer you think you are.
Action: Route catch-alls out of personal sequences. Use a different motion or suppress.
6. Duplicate people under slightly different keys
Same human as j.smith@ and jane.smith@, or two LinkedIn URLs that are the same profile with tracking junk. Apollo-style duplicate guidance exists because dual sequences to one human is how you earn spam complaints and angry replies.
Action: Dedupe on normalized email + LinkedIn + name/company before write. Merge preferences and suppression history.
7. Empty fields that the draft somehow filled anyway
The enrichment export has blank tech_stack or recent_news, yet the draft mentions a specific tool or announcement. That is gap-filling. Explorium's checklist is the right instinct: if the field was not verified in context, the agent should omit, not improvise.
Action: Diff the draft against verified fields only. Anything not in the verified set gets cut before QA.
A pre-send rewrite checklist that fits agency volume
For each client workspace batch:
- Fail closed on missing match confidence for identity-sensitive rows.
- Drop or re-enrich rows past freshness SLA.
- Strip unverified funding / exec / stack claims from copy.
- Block catch-all and role accounts from personal cadences.
- Dedupe before the write step, not after the first bounce.
- Log rewrite reasons so you can see whether the enrichment vendor or the prompt is the leak.
None of this requires inventing a case study. It requires refusing to send fiction dressed as research.
GTM's write step is supposed to run from prospect evidence inside client-isolated workspaces, with a compliance gate still required before send. Enrichment that cannot survive these seven checks is not evidence yet.
If you want write-from-evidence with field provenance in the motion, join the waitlist.