Voice guard catches the AI tells before the prospect does.
The single biggest signal that a cold email was written by AI is the punctuation and vocabulary. Em-dashes. Buzzwords. Cohort claims with no source. We built a three-layer voice guard that rejects every one of those patterns before the sequence ever loads.
How it works
-
Layer 1, generation rules.
The generator prompt includes 16+ hard rules. No em-dashes, no en-dashes, no spaced hyphens. No banned words. No bullet points in email bodies. 3-line signature with a URL. Under 120 words. Natural opt-out in emails 2 and 3. These are CRITICAL RULES the model is told it will be rejected for violating.
-
Layer 2, validation at generation time.
After the model returns a draft, an automated review pass runs a battery of checks. Banned-word detection. Punctuation pattern detection. Anchoring checks that flag stats without a cited source or first-person experience. Framing checks that catch laundered authority phrasing.
-
Layer 3, pre-load guard.
Before any sequence ships to the sending tool, a final review pass scans every email subject, body, and LinkedIn message. Any disallowed pattern refuses the entire load. Strict mode is the default for every client. There is no "the model missed one" path to production.
-
Cited stats verified against an approved library.
Any sequence that cites a benchmark gets its verbatim quote checked against an internal source library where every entry has a verbatim quote, a live URL, and a verified date. A paraphrase fails. A source-name swap fails. A one-word drift fails. If it does not match, the sequence is regenerated.
Want proof the voice holds up?
On the discovery call we run 5 real sequences against your ICP, live. Voice guard output included.