Deliverability is the quiet reason most outbound programmes stop working. Nothing announces it. Reply rates drift down, a rep says "the market is quiet", and six weeks later you discover half your sends have been landing in spam since the day you doubled volume. AI made this worse in both directions: it made sending at scale trivially easy, and it gave mailbox providers much better tools for spotting mail that nobody wanted. This guide covers what actually determines whether you reach the inbox in 2026.
What the filters are judging
Spam filtering is no longer mostly about words. It is about whether people who receive your mail behave like people who wanted it.
| Factor | Weight | What you control |
|---|---|---|
| Recipient engagement | Highest | Who you send to and whether the message is relevant |
| Domain and IP reputation | High | History, consistency, complaint rate |
| Authentication (SPF, DKIM, DMARC) | Gate | Setup. Fails here and nothing else matters |
| Volume pattern | High | Ramp, consistency, spikes |
| List quality | High | Bounce rate, spam traps |
| Content | Lower than people think | Links, images, formatting |
The practical consequence: you cannot write your way out of a reputation problem. Teams spend weeks rewriting copy when the issue is that they sent 2,000 emails from a domain that had sent 50 the month before.
The setup you need before sending anything
- Use a separate domain for outbound. Not your main company domain. If outbound damages a domain's reputation, you do not want that domain to be the one carrying your invoices and customer support.
- Authenticate properly. SPF, DKIM and DMARC on every sending domain. DMARC should start at p=none while you watch the reports, then move to quarantine.
- One mailbox, modest volume. Treat each mailbox as having a ceiling. More mailboxes, not more per mailbox, is how you scale without looking like a machine.
- Warm up over weeks, not days. Start at a handful of sends per day and ramp gradually. A new domain sending hundreds on day three is the single clearest spam pattern there is.
- Verify the list before sending. Bounces are weighted heavily, and a high bounce rate on a new domain can end it immediately.
Why AI-era outbound breaks deliverability
- Volume became free. The constraint used to be how many emails a human could write. That constraint is what kept most programmes accidentally within safe limits.
- Personalisation became shallow. Mail-merged "I saw your company does X" reads as personalised to a filter and as spam to a human, and the human's behaviour is what the filter learns from.
- Lists got bigger and worse. Enrichment tools make it easy to add thousands of contacts who were never a real fit, which pushes down engagement, which pushes down reputation.
- Everyone is doing it. Your prospect now receives many AI-written emails a day. Their threshold for marking something as spam has dropped, and those complaints are the strongest negative signal available.
This is why sending more usually produces fewer meetings after the first few weeks. See the autonomous SDR for where that volume pressure comes from.
What to watch, weekly
- Bounce rate. Keep it low. A rising bounce rate means your list source has degraded.
- Spam complaint rate. The metric with the least tolerance. A small percentage is a large problem.
- Reply rate by domain. A sudden drop on one sending domain while others hold steady is a reputation problem, not a copy problem.
- Open rate by provider. If Google holds and Microsoft collapses, that is a filtering difference, not a market shift.
- Seed inbox placement. Send to your own test accounts across providers and check where it lands. This is the only direct measurement available.
Open rates are unreliable individually because of privacy protection, but the trend per provider still tells you something useful.
If you are already in trouble
- Stop sending from the affected domain. Continuing while you diagnose makes recovery longer.
- Find the trigger. Usually a volume jump, a new list, or a sequence that ran without a reply check.
- Clean the list. Remove unverified addresses and anyone who has not engaged across several campaigns.
- Re-warm slowly, starting with your most engaged recipients, since their opens and replies are what rebuilds reputation.
- Reduce the steady-state volume you return to. If the old volume caused it, the old volume will cause it again.
Recovery takes weeks. There is no reset button, which is why the warm-up discipline at the start is worth the patience it costs.
The uncomfortable trade
Most deliverability advice is operational, but the real lever is strategic: send fewer emails to people more likely to want them. Every tactic above is a way of managing the symptoms of sending to people who did not. A smaller, better-targeted programme has better deliverability almost automatically, because engagement is the dominant signal.
That is also why signal-based selling helps here. Timing outreach to a real event raises reply rates, and reply rates are what protect the channel. AI email outreach covers the message itself.
