Symptoms#
- Messages landing in spam despite previously good placement
- Bounces mentioning policy, authentication, or reputation
- Provider dashboards showing a drop in reputation or authentication pass rate
Diagnosis steps#
- Confirm the configuration is live: query DNS from an external resolver and check the sending platform's settings page.
- Send to a seed mailbox and read the full headers, especially Authentication-Results.
- Compare the domains in From, Return-Path, and DKIM d= for alignment.
- Check provider dashboards (Google Postmaster Tools, Microsoft SNDS) for reputation and error rates.
- Review recent changes: new vendors, DNS edits, list imports, volume spikes.
Likely causes#
- Hallucinated details about the recipient, which destroy trust instantly.
- Over-personalization that reads as surveillance.
- Varying so much per message that you cannot learn what works.
Fix and re-verify#
- Enrich contacts with 3 to 5 verifiable data points: role, company, tech stack, recent news, mutual context.
- Write a prompt template that constrains tone, length, and forbidden claims.
- Generate a first line and a value proposition per contact; keep the rest of the message stable.
- Human-review a sample of every batch before sending.
- Track reply rate per template and iterate weekly.
Frequently asked questions#
Does AI-generated email get flagged as spam?
Filters judge sender behavior and recipient reaction, not authorship. Poorly targeted AI mail gets flagged because it is poorly targeted.
How do I keep AI drafts in my voice?
Provide examples of your real sent mail and constrain style in the system prompt. MailMaid's Draft with AI learns from thread context for exactly this reason.