AI Email Statistics 2026: How AI Changed Your Inbox

There is no defensible global percentage of email “written by AI.” The strongest bounded figures we could reopen answer narrower questions: a 2025 survey of 1,234 email senders found 79% used or planned to use AI in their email programs and 27% used it regularly; Microsoft's preliminary randomized field study of 3,000 people at 60 Copilot customers found users read 11% fewer individual emails and spent 4% less time interacting with email. These are different populations, questions, and outcomes—not a single adoption series.

Verification note: This is a documentary source review checked against current sources on August 3, 2026. We did not hands-on test the named product or workflow during this review, so claims are limited to the cited documentation. Interfaces can vary by account, region, rollout, and app version.

This reference separates four constructs that roundups often mix: automated sending, generative text classification, sender adoption, and recipient-side productivity. Every number below keeps its population, date, denominator, and key limitation attached.

How many emails are written by AI?

The public evidence does not support a population-wide answer. “Automated” includes deterministic marketing, receipts, alerts, and bots; it is not equivalent to generative-AI authorship. AI-text detectors also require a disclosed sample, validation set, operating threshold, false-positive rate, and error analysis before their percentages can be interpreted.

Two widely repeated claims fail that bar in the public sources we could reopen:

Repeated claim	Public source available	Missing evidence	Publication decision
13% of email traffic was human-written	TechRadar coverage of a Hostinger analysis	Primary dataset, sampling frame, denominator, collection window, and classification method	Not used as a global or headline statistic
51% of spam and 14% of BEC were AI-generated by April 2025	Barracuda vendor summary	Exposed message count, detector validation, threshold, false-positive rate, and reproducible method	Recorded as a vendor-reported result, not promoted as a benchmark or directional finding

The answer to “what percentage is AI-written?” is therefore unknown with current public methods. That gap is useful: it prevents automated traffic, detected prose, and all global email from becoming one invented denominator.

Is AI growing total email volume?

We could not verify a public primary dataset that isolates generative AI's contribution to global daily email volume. Commercial forecasts and secondary provider coverage did not expose a sufficient sampling frame, denominator, method, and uncertainty for that causal question, so this page publishes no AI-attributed volume estimate. General traffic forecasts answer a different question; inbox volume is covered separately in email overload statistics.

How many email senders use AI?

The Sinch Mailgun Email Impact Report surveyed 1,234 email senders from late September through early November 2025. Its method note says 769 respondents were customers or other email-community participants and 465 came from an Alchemer high-volume-sender panel in the U.S., U.K., France, Germany, and Spain. It separately analyzed more than 400 billion messages sent through Sinch infrastructure in 2025; that platform traffic is not the denominator for the survey percentages.

In that sender survey, 79% said they used or planned to use AI in their email programs, while regular use was 27%. “Use or plan” is a broad self-report and should not be compared as a trend with an older survey that asked only about planned adoption. We therefore removed the prior 2023-versus-2026 series.

Among respondents in the same report, the published use-case shares were:

AI use case among senders	Share	Source
Copy generation	41%	Sinch Mailgun, 2026
Content personalization	36%	Sinch Mailgun, 2026
Dynamic content	29%	Sinch Mailgun, 2026
Send-time optimization	27%	Sinch Mailgun, 2026
Data analysis	27%	Sinch Mailgun, 2026

The report also says 23% of respondents using AI reported no improvement. Its AI and non-AI groups were not randomly assigned, so self-reported program changes cannot isolate AI from company maturity, list quality, budget, or measurement practice. This supports a dated adoption snapshot, not a causal performance verdict.

How many people use AI to handle the email they receive?

Recipient-side evidence measures workplace use or product interaction, not the authorship share of incoming mail.

Microsoft's Work Trend Index reported that 75% of surveyed knowledge workers used generative AI at work, with 46% of those users saying they started within the prior six months. That is a broad work-use measure, not an email-assistant rate. Gallup's U.S. employee panel found 40% used AI at work in any form in 2025, 19% used it a few times a week or more, and 8% used it daily. Again, none isolates email.

For email specifically, Microsoft's 2024 report gives preliminary results from a six-month randomized field study of 3,000 people at 60 Copilot customers. Users read 11% fewer individual emails and spent 4% less time interacting with them. That is a vendor-published treatment result among participating customers, not a 2026 population benchmark or proof that less reading improved work quality. The same report's commercial-user telemetry says 85% of observed emails were read in under 15 seconds and about four were read for each one sent; those telemetry observations use a different denominator from the experiment.

Google reported in May 2017 that Smart Reply accounted for 12% of replies in Inbox on mobile. That is a first-party product share for a discontinued app and specific surface—not a current Gmail or population rate. The sources above cannot establish a sender-versus-recipient trend because they measure different populations and questions. For separate time measures, see email productivity statistics.

Which figures can be compared safely?

Compare results only inside the same population, question, and method:

Evidence unit	Population / denominator	What it can answer	What it cannot answer
Sinch Mailgun sender survey	1,234 senders; customer/community and external high-volume panel, fielded Sep.–Nov. 2025	Self-reported sender adoption and use cases	Global sender prevalence, recipient use, or causal performance
Microsoft Copilot field study	3,000 people at 60 Copilot customers over six months	Treatment difference in measured email interaction among participants	Population-wide time saved, work quality, or all assistants
Gallup employee panel	U.S. employees, all workplace AI use	Frequency of workplace AI use	Email-specific adoption
Google Smart Reply product observation	Replies in Inbox on mobile in 2017	Share of replies using that feature on that surface	Current Gmail or general AI-writing prevalence

Do not average these percentages, subtract one from another, or describe them as one time series. They are useful precisely because their boundaries are visible.

Our intended contribution to the receiving side is deliberately un-generative. We build Flick (flicked.email), a swipe-to-triage email client with a finite deck, and maintain a 150-sender Exit Gap work-list for future unsubscribe-outcome research. It is not yet a dataset of sender behavior: as of August 3, 2026, four entries have capability-only observations and none has a published honor grade. Capability means a machine exit route was present, not that the sender respected a "no." Each future result must rest on at least three probes across at least 48 hours with preserved evidence; the founder-run burner-mailbox observation cycle has not occurred.

FAQ

What percentage of emails are written by AI in 2026?

No population-wide percentage is supported by a public method we could verify. The repeated 13% human-written claim lacks a public primary sampling frame and classifier method. Barracuda's 51% spam claim lacks an exposed message count and detector validation in its public summary. We record those method gaps rather than promote either number.

How many emails are sent per day in 2026?

This page does not publish a 2026 global total or AI-attributable share because the public sources we found did not expose the sampling frame, denominator, method, and uncertainty needed for that question. Sender-adoption and recipient-productivity studies answer narrower questions and cannot be converted into message volume.

Do AI email assistants actually save time?

Microsoft's preliminary field study found Copilot users read about 11% fewer individual emails and spent about 4% less time interacting with email. It covered 3,000 participants at 60 Copilot customers and was vendor-published. It did not establish population-wide time savings, work quality, or performance for other assistants.

Is AI making spam worse?

The public sources reviewed here do not establish a causal or population-wide answer. Barracuda reported AI-classified spam and BEC shares, but its public summary did not expose the message count or detector validation needed to interpret the percentages. Treat AI-enabled abuse as a threat model, not as a measured global trend from this evidence.

How many people use AI to write their own emails?

No current population estimate isolates personal email writing. Gallup's 40% covers any workplace AI use, while Microsoft's 75% covers surveyed knowledge workers using generative AI for work; neither is email-specific. Google's 12% Smart Reply share applied to Inbox mobile replies in 2017, not today's global email.