Your Inbox Used to Mean Something
Email used to operate on an implicit trust contract. If something landed in your inbox, there was a reasonable assumption it mattered — relevant, contextual, or at least sent by someone with enough understanding of who you were to make a real attempt at relevance.
That assumption is breaking down.
What I see now is a steady volume increase paired with a steady relevance decrease. Messages that are not just generic, but structurally disconnected from reality. Offers for services that do not apply to my business. Outreach in the wrong language. Proposals built on assumptions that are clearly, demonstrably wrong. Not slightly off. Completely off.
Knowledge workers receive 50 to 100 emails a day, and only 24% of them warrant any attention — the rest is newsletters, automated notifications, internal updates, and threads where someone was unnecessarily CC’d [5]. The cost of sorting that pile is real: 28% of the workweek goes to managing email, and 70% of workers identify email as their primary source of workplace stress [6].
To understand why the volume keeps growing while relevance falls, look at the system generating these messages.
The Cost of Sending Hit Zero — and Everything Broke
Before LLMs, sending a personalized email at scale required human time. A campaign of 10,000 prospects, written individually enough to feel real, cost a measurable number of person-hours. That cost was the economic brake. It limited how much volume any sender could push through the channel before someone in finance asked whether it was worth it.
That brake is gone. Spammers now save 95% on campaign costs by using LLMs, according to the 2026 StrongestLayer threat report [4]. Ninety-five percent. This is not an incremental efficiency gain — it is a phase change in the channel’s economics.
The volume response was immediate. 51% of all spam is now AI-generated, having tripled in two years [1]. And the dynamic is not confined to bad actors: 63% of marketers admit to using AI in email campaigns, with another 25% likely doing so without disclosure [11] — nearly nine in ten outbound marketing teams running on at least partial LLM output.
“Personalization at scale” turned out to produce the least personal messages most of us have ever received. When the same model template hits the same prospect from three different vendors in the same week, the prospect develops pattern recognition. The technology built to make outreach feel handcrafted ended up training everyone to recognize the seams.
The clearest case study is what analyst Zak El Fassi calls The Cold Swarm — an AI-generated communication crisis systematically destroying professional outreach at industrial scale. As AI cold-email platforms (Instantly.ai, Smartlead, Salesforge) proliferated through 2023–2025, B2B reply rates fell from 8.5% in 2019 to 5% in 2025 — a 41% reduction, with the steepest decline coinciding with mass AI adoption [2]. In the same period, average open rates dropped 23% and reply rates settled at 5.1% across industries. 95% of cold emails now generate zero response. As El Fassi puts it: “When every email claims to be personally crafted but feels artificially generated, recipients develop pattern recognition that rejects both genuine and artificial outreach indiscriminately.”
The problem reaches the person who has a legitimate reason to write. Volume taught recipients to discount the entire category, so careful outreach now meets the same suspicion as the messages that diluted attention.
So Filters Escalate. So Does Everything Else.
On one side: systems generating high-volume outreach, hoping something gets through. On the other: increasingly aggressive filters protecting attention. Neither side improves the quality of interaction. They escalate.
The collateral damage is measurable. Between Q1 2024 and Q1 2025, Microsoft tightened anti-spam heuristics across Office 365 — the platform most enterprise B2B mail passes through. Inbox placement collapsed from 77.4% to 50.7% in a single year. Organizations sending 1M+ emails monthly fell from 50.0% to 27.6%. Mailgun dropped 27.8 points to 26.1%. MailChimp dropped 19.6 points to 32.3%. The filtering operated silently — no bounces, no warnings, no notification to recipients that messages had been blocked. Critical business communication simply vanished. Fully authenticated domains, by contrast, achieved 2.7x higher placement, opening a stark divide between prepared and unprepared senders.
Then the platforms escalated again. In November 2025, Gmail shifted from soft warnings to hard rejection at the SMTP protocol level: non-compliant bulk emails from senders of 5,000+ per day no longer get routed to spam — they receive permanent 550 failure codes and bounce back outright [7]. As Google explained when it announced the underlying policy in early 2024, the rules — SPF, DKIM, DMARC, one-click unsubscribe, a 0.3% spam complaint threshold — are “intended to help protect users from spam, spoofing, and phishing.” Yahoo followed with parallel requirements for any sender above a small volume. Globally, inbox placement fell to 84.6% in 2024–2025, the steepest year-over-year decline in a decade — 1 in 6 emails now never reaches its destination [3]. Security vendors are catching one malicious email every 19 seconds.
What you have just watched is one cycle playing out across two platforms. Call it The Channel Erosion Cycle:
- Cost Collapse — the marginal cost of sending approaches zero.
- Signal Contamination — noise starts mimicking signal; structurally irrelevant messages flood the channel alongside real ones.
- Filter Escalation — receivers, human and automated, build increasingly aggressive defenses.
- Trust Decay — recipients stop assuming inbox contents matter.
- Channel Triage — senders and receivers either fight for the channel, migrate critical communication elsewhere, or run a hybrid.
Office 365’s collapse is step 3 in motion. Gmail’s hard rejection is step 3 going binary. The 67% of US adults who avoid responding to email (next section) is step 4 already in the rearview. The cycle loops: lower engagement triggers compensating volume from senders who did not understand what just happened, which feeds the next turn of Cost Collapse. It looks less like communication and more like a war of attrition.
Maybe the Channel Itself Is the Casualty
Two-thirds of US adults — 67% — now avoid responding to emails because any given message might be spam or a scam [8]. A legitimate sender has to reach someone who is already weighing that risk.
Two-thirds of the adults you might be trying to reach have already decided that not responding is safer than trying to distinguish real from fake. That is not a spam problem. It is a channel trust collapse.
The sender-side data tells the same story from a different angle. Open rates have dropped 23% in the same period AI entered mainstream sales workflows. Reply rates settled at 5.1% across industries [2]. 95% of cold emails generate zero response. The medium has shifted from high-trust — if it landed in my inbox it probably matters — to low-trust — anything could be here, and important things might not be. A 2025 ScienceDaily summary of a study of 1,100 professionals found that managers’ regular AI-email use can erode trust with their teams, even when the polished output reads better than the unaided draft. The mechanism the authors propose is reduced authenticity signaling; the effect is contested, but it points in the same direction as everything else.
When more than half your audience is filtering everything from your category by default, what does “open rate” even measure?
Maybe I Am Wrong: Email Has Survived Every Death Sentence
There is a reasonable counter to all of this, and it deserves a real answer.
Some claim email has been declared dead every decade and never quite died — which is true. Bayesian filters tamed the first spam wave in the early 2000s. Email survived social media, mobile, Slack. The current crisis, on this view, is another adaptation cycle. Google and Yahoo’s 2024 sender requirements show the system self-correcting in real time.
There is an even stronger technical version. AI-powered filtering has structural advantages that AI-powered spam does not: platform-side detection runs on billions of training signals with full metadata access — sender reputation, engagement patterns, authentication, network behavior. Google processes 15+ billion emails a day. No spammer can match that dataset. Defenders win these arms races, even when they take a few years to catch up.
Both points are real. Email will not die. It will survive as a medium. The honest framing is degradation, not extinction. And degradation is the more dangerous outcome — because a dead channel forces a migration; a degraded one quietly absorbs critical messages until they fail to land.
This cycle differs from previous ones in a specific, mechanical way. Previous spam waves had economic friction: buying lists, running servers, writing copy that did not read like a robot. AI removes that friction entirely. 95% cost reduction is not “spammers got a little more efficient” [4]. It is the brake disappearing. The reason filters caught up in past cycles is that volume could not scale indefinitely on a human budget. Volume can now scale indefinitely on a $200/month tool.
And even when defenders prevail at the platform level, the cost is borne by bystanders. The 84.6% inbox placement rate already shows aggressive filtering catching legitimate senders. Per IBM’s 2025 analysis, only a third of organizations have implemented the SPF, DKIM, and DMARC stack that Google, Yahoo, and Microsoft now require [9]. Two-thirds absorb the collateral damage. The arms race may have a winner. The channel does not.
What Works Now Without Making It Worse
The response starts with how senders choose whom to contact and how receivers route messages that must arrive.
The instinct, when engagement drops, is to send more. The data is unambiguous on what that does.
Cold email campaigns under 100 recipients achieve 5.5% reply rates. The same campaigns sending 10+ contacts at the same company drop to 3.8% [10]. Campaigns targeting 1,000+ recipients fall to 2.1%; campaigns under 50 maintain 5.8% — nearly 3x higher than the bigger blasts. Volume and effectiveness are now inversely correlated. There is no scale at which this inverts.
It would be too easy to blame the tool. AI does not send emails — people and organizations do, and the incentive to spray-and-pray existed before LLMs. The agency is human. What AI removed is the natural economic brake that kept the human incentive from running unchecked. Policy and cultural responses calibrated to human-speed spam are insufficient for AI-speed spam.
For senders, restraint means smaller lists, genuine research per prospect, and fewer messages with a clear reason to reach that person. The Cold Swarm post-mortem reads identically across the agencies that survived: the ones that grew did the opposite of scaling.
What works for receivers is channel triage — step 5 of the cycle turned into a practical move. Classify your messages by criticality. The must-reach ones — contract changes, time-sensitive offers, anything with legal or financial consequence — should not be living in the same probabilistic channel as a 5,000-sender newsletter blast. Move them to verified paths: authenticated portals, direct messaging with read receipts, scheduled calls. Keep email for lower-criticality updates where a 16% miss rate is survivable.
There is a counter to even this: maybe email degradation is a feature, not a bug — it accelerates a long-overdue migration to Slack, Teams, WhatsApp, and closed channels with verified identities. There is something to that. For real-time collaboration inside an organization, closed channels are already winning. But closed channels solve noise by building walled gardens, which sacrifices what made email uniquely useful: first-contact communication across organizational boundaries. The answer is triage, not wholesale migration. Use email where it still works. Use other paths where it does not. Stop pretending one channel can carry everything.
Email is one node in a larger pattern, the one I explored in Brand Crisis: AI intermediation degrades signal in any channel where the cost of producing plausible content falls below the cost of verifying it. Email is the first channel where the failure is undeniable.
Four Questions to Sit With
The closing actions are simple enough to do this week:
- Measure your signal-to-noise ratio. What percentage of your outbound actually gets a human response, and has that number changed in the past year? If reply rate is down and volume is up, the channel is telling you something.
- Stop compensating with volume. The data shows it accelerates degradation, not results. Cut your sends. See what happens.
- Audit which communications still belong in email at all. For the must-reach ones, what is the verified alternative — and what is the cost of moving them?
- Build authentication as table stakes. SPF, DKIM, DMARC. But do not confuse compliance with trust. Authentication gets your mail accepted by filters. It does not get it read.
The inbox now carries less assurance that a message matters. The people using it can rebuild some of that assurance by choosing recipients carefully and protecting a reliable route for the messages that must arrive.
References
- [1] “Email Deliverability in 2025: New Rules, Higher Standards” — 1827 Marketing, 2025; DeBounce Email Spam Statistics, 2026.
- [2] “The Cold Swarm: Why Your AI-Personalized Pitch Is Dead on Arrival” — Zak El Fassi, 2025; MailReach Deliverability Statistics, 2025.
- [3] MailReach, “Email Deliverability Statistics 2025: Benchmarks & Trends,” 2025; IBM Think, “The Hidden Email Crisis That Costs Companies Billions,” 2025.
- [4] StrongestLayer, “AI-Generated Phishing: The Top Enterprise Threat of 2026,” 2026.
- [5] Mailbird, “2025 Survey: Email Overload’s Impact,” 2025; PMC peer-reviewed, “Drowning in Emails,” 2024.
- [6] Mailbird, “2025 Survey: Email Overload’s Impact,” 2025; PMC, “Drowning in Emails,” 2024.
- [7] PowerDMARC, “Gmail Enforcement 2025: Google Begins Rejecting Emails,” 2025; IBM Think, “Hidden Email Crisis,” 2025.
- [8] 1827 Marketing, “Email Deliverability in 2025,” 2025; DeBounce, 2026.
- [9] IBM Think, “The Hidden Email Crisis That Costs Companies Billions,” 2025.
- [10] Zak El Fassi, “The Cold Swarm,” 2025.
- [11] Zak El Fassi, “The Cold Swarm,” 2025.
- Inline-cited enrichment: Google sender requirements blog, 2024; Yahoo sender guidelines, 2024; ScienceDaily — AI emails and trust at work, 2025.