Nobody Opened Your Email. Their AI Did.

I still subscribe to things. Retailers still send me offers. Companies still announce products. There are still far too many messages sitting in the Promotions tab.

I just increasingly don't read them.

AI does.

Every so often, instead of going through the emails myself, I ask AI to look at what has come in and tell me if there is anything actually worth my attention.

Most of the time there isn't.

Sometimes there is a genuinely good offer, something I had been waiting for, or a piece of information that matters. In that case, tell me about it and I'll go look.

From my perspective, this is fantastic.

I still get the upside of being subscribed without having to perform the tiny administrative ritual of opening, scanning and deleting everything myself.

But from the sender's perspective, something fairly fundamental just happened.

The first reader of the email wasn't me.

It was an AI working for me.

And I think that has some interesting consequences for marketing.

The first reader of the email wasn't me. It was an AI working for me.

The basic mechanics of email marketing haven't really changed for a long time.

Get into the inbox. Write a subject line good enough to earn an open. Make the email interesting enough to hold attention. Get the click. Get the action.

We've built an enormous amount of machinery around those steps.

Subject-line testing. Preview text. Send times. Hero images. Personalization. CTA placement. Templates. Mobile optimization.

All of it rests on an assumption that has always been so obvious that nobody really needed to say it: the recipient is the person consuming the email.

That assumption is starting to get shaky.

Google now lets Gmail users summarize conversations, ask questions about information buried in their inbox, and increasingly interact with email conversationally rather than manually searching through it. Outlook's Copilot can summarize email threads and attachments as well.

So the journey can now look more like this: email arrives, AI reads it, AI decides whether it matters, AI summarizes it, surfaces it, or effectively ignores it. Then maybe I open the original.

That last step is no longer guaranteed.

The funny thing is that marketers were already losing the open as a clean measure of attention.

Apple's Mail Privacy Protection prevents senders from reliably seeing whether somebody using the feature actually opened an email.

So even before AI enters the picture, an 'open' isn't necessarily what we like to pretend it is.

AI makes the question stranger.

Imagine a company emails me: 30% off the product you looked at last week. Ends Friday.

I never open the email.

My assistant reads it and tells me: That thing you were considering is 30% off until Friday.

The communication worked. I received the information. I may even go and buy the product. But I never opened the email.

So what exactly happened?

Was that email unsuccessful because it generated no human open? Or was it extremely successful because the information made it all the way to me without requiring one?

Our measurement systems aren't really designed for that distinction.

This is where it gets more interesting.

If AI increasingly sits between the sender and the recipient, marketers aren't really writing for one audience anymore.

They're writing for the human. And the human's agent.

Those two audiences do not necessarily value the same things.

A beautiful hero image may matter once I decide to look at the email. It doesn't matter very much if my AI has already concluded there is nothing useful inside it.

A clever subject line can make me curious. An AI doesn't need to guess from the subject line. It can read the message.

And that means some of the least glamorous questions in marketing suddenly become much more important.

What are you actually saying? What is new? Why does it matter? Who is it relevant to? Is there a deadline? Is there genuinely something here that this person needs to know?

These sound like painfully basic questions.

Maybe that's the point.

For years we've become extraordinarily good at optimizing the packaging around a message.

AI may force us to become better at the message again.

This is the part I keep coming back to.

An AI working on behalf of the recipient has completely different incentives from the sender.

Its job isn't to make sure your campaign performs well. Its job is to protect my time.

If ten companies send me broadly similar promotional emails, I don't actually want my assistant to faithfully reproduce all ten.

I want it to tell me: Nothing useful here. You've seen this before. This offer isn't particularly good. This one actually matters.

That could make AI an extremely unforgiving gatekeeper.

Humans get distracted. We get curious. Sometimes we click something simply because somebody wrote a ridiculous enough subject line.

An AI can just read the thing and make a judgement based on its contents.

Which creates an uncomfortable possibility.

If somebody's AI consistently decides that your emails aren't worth showing them, the answer may not be a better subject line.

The AI may just be right.

There's another behavioural change happening here that I find just as interesting.

We normally think of email as a queue.

Newest message at the top. Work your way down. Deal with whatever came in.

AI makes the inbox behave more like a database.

Instead of reading everything, I can ask: What did my kids' school send me this week that I actually need to act on? Did any airline send a fare worth knowing about? What did my bank say about this? Was there anything useful in the promotional emails I received this month? What has this company told me about that product recently?

That is a fundamentally different interaction with email.

The information doesn't necessarily need to win my attention at 9:04 on Tuesday morning anymore.

It needs to be useful when I eventually ask for it.

And in some ways, that might make email more valuable.

Less interruptive. More retrievable.

The email becomes part of a body of information that my assistant can reason across later.

This gets particularly interesting to me because I work in pharma, where communication with healthcare professionals already comes with far more constraints than normal consumer marketing.

Think about what happens when an HCP receives several emails from a company over a period of time and later asks an assistant: What has this company sent me recently about this treatment?

The assistant may be perfectly capable of finding and summarizing those communications.

But now a new set of questions appears.

What survives the summary? What gets compressed? Does important context remain attached to the information? When should the person be directed back to the original communication? What happens when the email itself can be summarized but the destination sits behind HCP authentication that the assistant cannot simply walk through?

The sender no longer fully controls how the information is consumed.

That doesn't mean marketers should start trying to 'optimize for AI summaries.' I think that would be a very strange direction to take regulated communication.

It means clarity matters even more.

If information is going to pass through a compression layer before reaching the human, we need to understand what that compression does to meaning.

That's a much bigger question than email design.

If somebody's AI can understand a message without the person opening it, the open rate becomes less useful.

And eventually the click may become more complicated too.

Maybe the better question is the one we probably should have been asking all along: Did the communication accomplish anything useful?

Did somebody learn what they needed to know? Did they take an appropriate next action? Did they seek more information? Did the message meaningfully contribute to an outcome?

Those things are harder to measure than opens and clicks.

Unfortunately, being harder to measure doesn't make them less important.

We've spent years measuring the things digital channels made easy to count.

AI may force us to get much better at measuring whether the communication actually did its job.

Did the communication accomplish anything useful?

There is also something slightly ridiculous about where all of this could end up.

The sender uses AI to analyze the audience. AI helps write the email. AI generates variations. AI decides when to send it.

Then my AI receives the email. My AI reads it. My AI decides whether their AI-generated email is worth bothering me with.

Two machines could end up conducting an entire marketing interaction on behalf of two humans who barely participate in it.

At that point, producing more content isn't much of an advantage.

Everyone can produce more content.

The scarce resource is still exactly what it has always been: attention.

The difference is that the recipient now has a much better gatekeeper.

And I'm not sure that's a bad thing.

I don't think AI kills email.

If anything, I think email remains incredibly useful precisely because it creates a durable record that an assistant can later retrieve.

What changes is the assumption that the message travels directly from brand to human.

Increasingly, there is something in the middle.

An assistant. A summarizer. A filter. An agent working for the person whose attention you're trying to earn.

So I don't think the question for marketers is: How do we get past the AI?

That misunderstands whose side the AI is on.

The better question is: What could we send that a person's AI would genuinely think was worth telling them about?

That's a much higher bar than getting an open.

It might also be a much healthier one.

What could we send that a person's AI would genuinely think was worth telling them about?