Cold email deliverability when an AI is doing the sending
Deliverability is the constraint that quietly decides whether an AI SDR works. You can have perfect targeting and excellent copy, and none of it matters if the mail lands in spam. AI makes this harder in one specific way: it removes the natural rate limit that a human typing imposed, so mistakes that used to take months to compound now take days.
Start here: the three records
If SPF, DKIM and DMARC are not all passing, nothing else on this page will save you. Since 2024 both Google and Yahoo require authentication for bulk senders, and enforcement has only tightened.
SPF
A DNS record listing which servers may send for your domain. One rule people get wrong constantly: you are limited to ten DNS lookups. Chaining several providers silently breaks it. Check the lookup count, not just whether the record exists.
DKIM
A cryptographic signature proving the message was not altered in transit. Use a 2048-bit key where your provider supports it. Rotate it if it has ever been exposed.
DMARC
Tells receivers what to do when SPF or DKIM fails, and — the part most people skip — sends you reports. Start at p=none with a reporting address, read the reports for two weeks, then move to p=quarantine and eventually p=reject. Going straight to reject without reading reports is how organisations discover they were breaking their own invoicing email.
If yourcompany.com is where your invoices, contracts and support mail come from, do not put cold volume on it. Buy a secondary domain — a close variant, redirected to your main site — and send from there. If the reputation burns, you have lost a £10 domain rather than your ability to bill customers.
Warmup, and why the curve matters
A brand-new domain sending 500 emails on day one looks exactly like a spammer, because that is what spammers do. Warmup means starting at a handful of sends a day and climbing over four to eight weeks while engagement signals accumulate.
The mechanism that matters is not volume, it is reply and engagement rate. Mailbox providers are watching whether recipients open, reply, mark as important, or move to spam. A slow ramp with real replies builds reputation far faster than a fast ramp with silence.
Practical shape: 5–10 sends a day in week one, roughly doubling weekly, never more than about 50 a day per mailbox at steady state. Want 500 a day? That is ten mailboxes, not one mailbox working harder.
What breaks deliverability fastest
- Bad list hygiene. Bounces above 2–3% are read as a signal you did not verify the list. Verify before sending, every time.
- Spam traps. Recycled addresses that exist only to catch senders using scraped or purchased lists. One hit can affect a domain for months.
- Volume spikes. A sudden jump from 50 to 500 a day reads as a compromised account.
- No replies. A campaign nobody answers is a campaign providers learn to filter. Reply rate is a deliverability metric, not just a performance one.
- Links and images. Tracking pixels, link shorteners and image-heavy HTML all cost you. Plain-text-looking mail with at most one link performs best cold.
- Unsubscribe friction. If people cannot leave easily they hit the spam button instead, which is far more damaging than an unsubscribe.
The AI-specific risks
Volume without judgement
A human writing 40 emails a day self-limits. An AI does not tire, so caps have to be explicit and enforced by the system rather than by good intentions.
Homogeneous copy
If an AI writes 2,000 emails from one template with variables swapped, the near-identical structure is itself a filterable pattern. Genuine per-account variation is a deliverability feature, not just a conversion one.
Fabricated personalisation
An email that congratulates someone on a funding round that never happened gets marked as spam by an irritated recipient. Hallucination is a deliverability problem as well as a credibility one.
No back-pressure
The critical question for any AI sending on your behalf: does it throttle itself when reputation dips? A system that keeps sending at full rate while complaint rates climb will destroy a domain in about a fortnight.
A monitoring baseline
| Metric | Healthy | Act now |
|---|---|---|
| Bounce rate | Under 2% | Over 4% |
| Spam complaint rate | Under 0.1% | Over 0.3% |
| Reply rate | Above 3% | Under 1% |
| Open rate | Above 30% | Under 15% |
| Sends per mailbox per day | Up to ~50 | Over 100 |
Google Postmaster Tools is free and shows you domain reputation directly from the largest receiver. Set it up before you need it.
If you are already in spam
Stop sending from the affected domain entirely — do not taper. Fix authentication. Purge and re-verify the list. Warm a fresh domain from zero. Re-engage your most responsive contacts first to rebuild positive signal. Recovery takes weeks, not days, and some domains never fully recover, which is the real argument for not risking your primary one.
GTMHack guards sending like infrastructure
Guided SPF, DKIM and DMARC with live verification, built-in warmup curves, per-domain daily caps and automatic throttling before reputation damage — not after.
See the deliverability engine← All guides · How buying-signal scoring works · UK GDPR & PECR checklist