The send button is a design problem

Everyone who has run a newsletter for more than a year has a send-button story. Mine is a merge tag that went out as {{first_name}} to eleven hundred people, which is fine, and a second one where I sent an issue twice in four minutes, which was worse.

I used to think these were attention failures. I now think they are interface failures, and that the interface is shaping more than just the mistakes.

What the button actually is

Consider what we are asking of this control. It performs an action that cannot be undone, at a scale where every error is multiplied by the list size, on a piece of work the author is emotionally invested in, usually late, usually tired, after a session of small anxious edits.

Almost every other irreversible action in software has a safety net. You can unsend a Slack message, recall a Gmail for thirty seconds, roll back a deploy. The one action where the stakes are highest and the author is most tired ships with a modal that says “Are you sure?” — which is not a safety net, it is a formality everyone clicks through.

The habits that work

I stopped trying to be more careful, because being careful is not a plan. These are mechanical.

Send yourself the real thing. Not a preview render — an actual send, to a real consumer address, with a real merge record attached. Preview modes fill merge tags with sample data and hide exactly the class of bug that embarrasses you.

Read it on a phone, in the morning. Not on the desktop where you wrote it, and not at eleven at night. Roughly two thirds of your readers will see it on a phone and the errors that survive proofreading are almost always errors of scale.

Schedule, never send. Schedule for ninety minutes out. The button becomes reversible for ninety minutes, which is the entire safety net, and it costs nothing. Every provider supports this and almost nobody uses it for anything other than time zones.

Run the link check last. After every edit, click every link in the sent test. Links break during the final pass more than during writing, because the final pass is when you move paragraphs.

If you are about to send while feeling any emotion at all about the issue, schedule it instead and look again in the morning. Feelings at 23:40 are not editorial judgement.

The part that is not about mistakes

Here is the thing I did not expect. Once I started scheduling instead of sending, I started writing differently.

When the button is irreversible, the last hour of work goes into anxiety — re-reading for errors, checking the subject line for the fifth time, hovering. When it is reversible for ninety minutes, that hour goes into the ending instead. I cut more. I strengthened more closings. The issues got better in a way I can point at, and the only thing that changed was that I had made a decision reversible.

Interfaces shape work. This one shapes yours twice a month.

That is issue thirteen. Next time we start on subject lines, which deserve a lot less mysticism than they get.