We analyzed 213 million AI email actions. Only 2.5% were sends

When group ideate an AI delegate alongside admission to their email, they normally image it sending messages: following up alongside clients, contacting suppliers, negotiating deals. That’s not what we found.

We looked at data from Agentic Mail, our email assistance built for AI agents, which launched at the end of May 2026. It lets small businesses, developers, and individuals nexus the AI delegate of their choice to a mailbox, so the delegate can search, read, organize, and dispatch email on their behalf.

Over 14 weeks, these agents performed 213 million actions. Only 2.5% of them were email sends. Most of the remainder went to searching and study messages (the delegate beginning an email and handling its text). Even by the most traditional estimates, AI agents made concerning 18 hunt and peruse actions for all email they sent.

Agents mostly peruse email; sending is the exception.

Usage grew 15x in 3 months

Weekly AI email actions rosy from 2.6 myriad to 38.3 myriad in fourteen weeks, concerning 15 times higher. Weekly energetic users rosy from concerning 3,000 to 25,000 (8x), and energetic mailboxes from 3,000 to 39,000 (13x).

Users are additionally connecting additional mailboxes. The average rosy from concerning 1 per person to 1.6, which suggests group discover it helpful adequate to hand complete additional of their email.

The mix of action has additionally shifted complete time. In the archetypal four weeks, searching was the chief action alongside concerning 38% of actions. But in the final three weeks it cut to concerning 17%, during inbox listing rosy from 25% to concerning 41% and study messages risen from 15% to concerning 25%.

Agents peruse far additional than they send

If we set apart specialized operations, concerning 193 million actions engaged operating immediately alongside email: searching, reading, sending, or organizing messages.

Some inbox-listing actions arrive from agents often checking for new mail, a procedure called polling. We can’t yet distinct these from deliberate browsing actions, so we study a range.

• The traditional figure: If we depart out inbox listing entirely, searching and study motionless document for 45% of all actions. That’s concerning 18 actions to collect data for all email sent.
• The high figure: If we obtain all inbox listings as research, the portion rises to 83%, or concerning 33 actions for all email sent.

The genuine figure sits location in between. Either way, the form is apparent – agents expend most of their attempt understanding what’s in the inbox before taking action.

This proportion has been remarkably stable. The traditional fig was concerning 17 actions per dispatch in the archetypal seven weeks and concerning 18 in the following seven, equal as weekly action grew approximately 15x.

Meanwhile, weekly sends rosy from concerning 60,000 to concerning 750,000, but sends made up concerning 2% of actions in the latest three weeks, downward from concerning 3% earlier.

Agents arrange less, and mostly poll

A smaller portion of actions went to regular inbox work: flagging, moving, deleting, and filing messages. At 0.22% of all actions, that motionless adds up to additional than 470,000 inbox actions group didn’t have to do themselves.

Deleting is particularly rare: concerning 22,000 delete actions against 5.4 million sends, or approximately one delete for all 250 emails sent.

Some agents are additionally using webhooks, which vigilant an delegate whenever a new communication arrives, so it doesn’t have to keep checking. They create up a small portion of action so far (0.15% of actions), so most agents motionless inspect for new communication by polling.

Roughly 20 years of email work

It’s tempting to multiply 213 myriad actions by a few seconds all and call it period saved. But that would overstate the result. A sole project a individual gives an agent, akin “reply to the latest communication from Peter,” can affect a dozen or additional actions. A individual doing the identical job would fair open the inbox and scan it.

The fair difference is tasks, not actions. There is one project we can figure directly: agents sent concerning 5.4 myriad emails. If we evaluation that all open-read-reply takes concerning two minutes for a person, that’s approximately 179,000 hours, or concerning 20 years of non-stop work, for thousands of our customers.

This is an explanatory number, not a measurement. It’s additionally conservative, since it leaves out read-only tasks akin summarizing an inbox or finding a document, which create up most delegate activity.

The bigger change may be that “checking your email” stops being item group do at all.

A note on privacy

This inspection is based lone on the type of all action, specified as “search” or “send,” which is recorded automatically whenever an delegate uses the service. Hostinger does not peruse or analyze the satisfied of customers’ emails. We provision the tools that let AI agents activity alongside a mailbox. Customers choose which AI delegate or tongue example to use, and they decide what it does.

Methodology

Hostinger analyzed concerning 212.8 myriad AI email actions in Agentic Mail complete fourteen successive weeks, W29–W42 (July 13 – October 18, 2026). The inspection starts in W29 whenever complete logging was in place.

An AI email action is one API call, the sole procedure an AI delegate performs, specified as listing inbox contents, searching, beginning an email, or sending one. One project a person gives an delegate can affect many actions. Counts shield calls to 15 tracked API endpoints, grouped into the action categories above.

An active user is a person whose AI delegate performed at smallest one act in Agentic Mail during the week. An active mailbox is a mailbox alongside at smallest one act during the week.

Percentages are rounded. User and mailbox counts are rounded to the nearest thousand, so early-week growth figures are approximate. Derived figures, specified as ratios, totals, and period estimates, are approximate.