Most cold-email drafts get sent without ever being looked at for spam triggers. The presend analyzer is there so you don't need to be a deliverability expert to catch the ones that will hurt you.
- Tier 1 — deterministic: stop-words, shouty caps, link-to-text ratio, tracking-pixel presence, domain intel on every link.
- Tier 2 — AI human perception: LLM evaluates the draft the way a reader would. Does it smell like cold outbound or like spam? What would a human classify it as?
- Tier 3 — link intelligence: resolve every URL, follow redirects, flag domains with bad reputation or recent spam history.
What it catches before you send
- Spam-trigger phrases: “100% free”, “limited time”, “click here”, and 200+ others.
- Over-capitalisation and excessive punctuation.
- Link-heavy templates (two+ links for cold outbound is a flag).
- Tracking pixel presence (often a negative on Gmail for cold).
- URL shorteners, recently-registered domains, sites on blocklists.
- Mismatched claim-and-reality: content says “from Alice” but From is
no-reply@.... - Over-formatted HTML that triggers content filters.
The AI human-perception layer
The interesting tier. We pass the draft to a reasoning model with instructions to classify it like a busy recipient would. The output is qualitative (“sounds like a copy-paste outreach” vs “sounds like a specific, researched message”) and ranks you from 0–100 with concrete rewrite suggestions.
This catches what rule-based systems miss: drafts that pass Tier 1 but still read like spam to a human. And since Gmail's own classifiers are increasingly LLM-shaped, this is a closer proxy to provider evaluation than a stop-words list.
Integration points
- Web UI: paste draft, get score + suggestions in 5 seconds.
- API:
POST /presendwith the draft payload. - MCP tool:
presend_analyze— agents call it before the sequencer sends. - Zapier/Make: trigger in your draft review flow.
Paste a draft. Free, no signup. The 5 seconds you save here is the one that prevents a one-hit-you-to-Spam campaign.