Mistake: AI-personalizing first lines without a fallback or sanity check

Mistake: AI-personalizing first lines without a fallback or sanity check
Clay or an LLM writes 'Loved your work at {company}' off thin scraped data and produces nonsense — referencing a job they left, a blog that 404s, or a competitor. Mass-personalized garbage is worse than an honest template.

The fix: require a data-quality gate. In Clay, only fire the AI line when the source field is populated and validated; otherwise drop to a clean generic fallback. Spot-check 20 rows per batch before sending — the failure mode is silent and at scale.

Pros / Cons of gated AI personalization:
— Pro: real personalization where data is good, safe fallback where it isn't
— Pro: avoids embarrassing factual misfires
— Con: fallback rows lose the personalization lift
— Con: gating logic in Clay takes setup time

Verdict: 4/5. Ungated, it's 4/5 the time great and 1/5 the time a reputation hit.

Who it's for: anyone running LLM first-lines across thousands of rows unchecked.