A startup sends an onboarding sequence. The first email goes out the moment someone signs up. It explains what the product does, lists the main features, and ends with “let us know if you have any questions.”
Open rate: decent. Click rate: near zero. And most users who opened it never completed setup.
The problem wasn’t the design or the subject line. It was that the email described the product instead of prompting the next action. Someone who just signed up already knows what the product does. They need to know what to do in the next ten minutes.
What AI optimizes for in email
Ask an AI to write an onboarding email and it will write a good-looking email. Clear paragraphs, friendly tone, a call to action at the bottom. It will mention the product’s key features because that’s what onboarding emails are supposed to do, according to every example it trained on.
What it won’t do is know which single action predicts whether this user will stick around. That’s specific to your product. The AI doesn’t have it.
So the email becomes a tour instead of a prompt. “Here’s the dashboard. Here’s where you can invite your team. Here’s where to find your settings.” The user reads it, thinks “okay,” and closes the tab. There’s no moment that says: do this now, before you close this tab, because this is the thing that makes everything else work.
The activation moment is almost never “explore the dashboard.” It’s something smaller and more specific: connect the first integration, add one real piece of data, see one output that couldn’t come from anywhere else. If the email doesn’t name that exact step, it’s describing the product, not onboarding the user.
Try it on a standard onboarding email
Typical AI-written welcome email. Friendly, complete, and unlikely to get a click.
The edits that matter
Cut everything after the first call to action. AI-written emails tend to keep going after the main ask: another paragraph of context, a list of secondary features, a reassurance that support is available. Each additional element gives the reader a reason to keep reading instead of clicking. One ask, then stop.
Replace the feature list with a single next step. “Here’s what you can do” is a product tour. “Here’s what to do right now” is onboarding. The step needs to be specific enough that the user knows exactly where to click and what they’ll see. “Explore the dashboard” isn’t specific enough. “Connect your first data source” is closer. “Import your last 30 days of feedback from Intercom” is exactly right.
Check the time claim. “Takes two minutes” appears in AI-generated emails because it appeared in every example the model trained on. It reads as filler when it’s there by default. If something actually takes two minutes, say so. If it takes five, say five. Specific times feel honest. The generic version reads like a placeholder that nobody replaced.
Check what triggered the email. AI doesn’t know when in the journey the email is sending. An email that goes out immediately after signup can say “you just created your account, here’s one thing to do before you close this tab.” An email sent three days later should acknowledge that the user may have already logged back in, or may have forgotten the product exists. The same body copy can’t do both, and AI will default to the generic version that fits neither.
Read the call to action out loud. “Get started” is not a call to action. It’s a placeholder for one. Same with “learn more,” “explore,” “take a look.” Each of these delays the moment the user understands what they’re supposed to do. The button text and the sentence before it should name the exact thing: “Add your first team member,” “Connect Intercom,” “See your first report.” If saying it out loud doesn’t create a clear mental image of what happens next, it needs to be rewritten.
The second email
The first email failing silently is common. But there’s a signal that’s easy to miss: users who open the first email and don’t click anything are telling you the email didn’t give them a reason to. They were interested enough to open. Something in the copy, the call to action, or the ask didn’t land.
AI will generate a second email that picks up where the first left off. “In case you missed it, here’s a quick reminder about our features.” This is the wrong email. The user didn’t miss the features. They decided the first ask wasn’t worth their time.
The second email should do something the first didn’t: give a concrete reason why this week is a good time to set this up, name a problem the user is probably having right now that the product solves, or show what a completed setup actually looks like so the user can see where they’d end up. Not a reminder. A different angle.
That angle requires knowing why users come to the product in the first place. The AI doesn’t know. Someone does.