Why email deliverability cratered

The numbers

Cold email reply rates by year, B2B average:

2020: 4–6% all-reply, 1.2–2.5% positive.

2022: 3–4.5% all-reply, 0.8–1.6% positive.

2024: 2–3.5% all-reply, 0.5–1.2% positive.

2026: 1–3% all-reply, 0.3–1% positive.

Source: aggregated from public benchmarks (Lemlist, Apollo, HubSpot) and our own customer data. The decline is unambiguous and continuing.

Force 1: Apple Mail Privacy Protection

Apple shipped MPP in 2021. It preloads remote images for Apple Mail users (which is roughly 50% of B2B email users globally), which historically meant tracking pixels.

Effect: open rates broke completely. Most cold-email tools now show 80–95% open rates because MPP is auto-opening on behalf of users who never read the email.

Operators optimizing for open rate were chasing a broken metric. Reply rate was always the better signal; MPP forced everyone to acknowledge it.

Force 2: Gmail's filter aggression

Gmail (~40% of B2B email globally) has rolled out increasingly aggressive spam filters since 2022. The 2024 release (BERT-based filter) catches subtle template patterns in cold-email copy.

Effect: cold email from new domains lands in spam at 30–60% rates. Even from warmed-up domains, suspicious content patterns kill deliverability.

Specifically targeted by Gmail's filter: any-link-density above 1 link per 80 words, generic templates, 'I hope this finds you well' openers, repeated subject patterns across many recipients.

Force 3: AI-detectable spam

Both Gmail and Apple Mail (and Outlook to a lesser extent) deployed ML models in 2024-2025 that detect AI-generated content patterns in incoming email.

Effect: poorly-written AI cold email gets flagged as spam more aggressively than 2020-era manual cold email did. The AI-arms-race created a ceiling.

Critically: high-quality AI personalization passes through. The detection targets repetitive AI-feel, not all AI-generated content. Same dynamic as LinkedIn's spam detection.

What's still working in 2026

Warmed-up sub-domains. Don't send cold email from your primary domain; use a sub-domain (send.yourcompany.com) warmed up specifically. 4–6 weeks of low-volume warmup before serious sending.

Plain-text emails. HTML cold email is increasingly suspicious. Plain-text or minimal-HTML formats land in inbox more reliably.

Personalization quality. Real personalization (referencing the recipient's specifics) passes through filters more reliably than generic content.

Conservative volume. 30–50 cold emails per day per domain is the safer sustaining volume. 100+ per day per domain triggers reputation problems within weeks.

Reply-coordination across channels. When a prospect replies to your LinkedIn message, follow up via email (warm context). This pattern lands much better than cold-first email.

What to stop doing

Don't optimize for open rate. The metric is broken. Optimize for reply rate or for booked-meetings only.

Don't send cold email from your primary domain. Reputation problems cascade to your transactional email. Use a sub-domain.

Don't include tracking pixels for EU contacts. ePrivacy compliance issue, plus they don't work anyway with MPP.

Don't include images in cold emails. Anything beyond plain text raises spam scores. Save images for transactional or replied-to threads.

Don't use 'I hope this finds you well' or 'I noticed your background.' Pattern-flagged.

The multi-channel approach that recovers what was lost

Single-channel cold email is dying. Multi-channel coordination recovers most of the lost reply volume.

Standard sequence: LinkedIn invite Day 1 (low-friction, high-acceptance) → Email follow-up Day 4 (warmer because you've already touched on LinkedIn) → LinkedIn message Day 7 if connected → Final email Day 11.

Reply rates on this sequence land at 12–22%, vs 1–3% on email-only. The same volume of effort, dramatically more replies.

Why it works: the LinkedIn touch creates context. By the time the email arrives, the prospect has seen your name and profile. The email reads as 'someone I'm vaguely familiar with' rather than 'someone I've never heard of.' Filters and humans both treat it more favorably.

How Infonet handles email deliverability

Per-customer warmed-up sending domains. SPF, DKIM, DMARC properly configured. Real-time deliverability monitoring — if a domain's deliverability degrades, automatic pause until recovery.

Multi-channel orchestration: email is one of three channels in a coordinated sequence, not a primary channel. Email touches go after the LinkedIn warm-up.

Plain-text default; HTML opt-in for replies and warm contexts only. Tracking pixels disabled for EU recipients.

Volume pacing: maximum 40 cold emails per day per sending domain. Custom volume override available for established senders.

FAQ

Should I give up on cold email entirely?

No, but it can't be the primary channel anymore. Use it as the second touch in a multi-channel sequence after LinkedIn. The math works there; it doesn't work alone.

How long does email warm-up take?

4-6 weeks of low-volume sending (5-15 emails/day, primarily to known contacts) before the domain reaches serious sending capacity. Skip warm-up and you'll spend the next 8 weeks fighting deliverability.

Are there email-only tools that still work in 2026?

Lemlist, Smartlead, and Instantly still work for warmed-up senders running conservative volume. They don't fix the underlying deliverability problem — they just optimize within the constraints. Multi-channel beats email-only on every metric.

What about Apple Mail Privacy on the sender side?

MPP affects the recipient side (their Apple Mail preloads images). On the sender side, you can't detect MPP-protected opens, so any open-rate metric is inflated. Just don't optimize for open rate.