We ran a two-option poll in a newsletter to 40,000 subscribers. Click-through went up 19% against the previous send. Then we ran the same format again, and again, and the lift evaporated. Not reversed—just gone. That plateau is more interesting than the initial win, because it says something about how people decide whether to click an email at all.

The first send worked because it lowered the cost of deciding

The email asked one binary question: which of two subject lines would you open? Two buttons, no text field, no login. That's it.

What made it work wasn't the poll. It was that the ask had a clear end. Behavioral economists call this the paradox of choice—Sheena Iyengar's famous jam study found that a display of 24 jams drew more browsers but a display of 6 drove ten times the purchases. Two options sits comfortably inside the range where people actually commit.

There's a second mechanism underneath. A binary choice removes the risk of being wrong in an interesting way. Nobody can pick a "bad" answer between two subject lines, so the social cost of clicking drops to near zero. Compare that to a five-question survey where a half-finished response feels like a small failure. The two-option poll lets people participate without committing to a position they'd have to defend.

The second send stalled for the most ordinary reason

We expected a smaller lift. We got essentially none.

The explanation is habituation, and it's well documented. Repeated exposure to the same stimulus produces a declining response—this is the same curve that flattens novelty in everything from product launches to joke punchlines. The first poll was a small event. The second was a format. By the third, subscribers had learned that clicking led to a page that asked them to vote on something they didn't care much about, and the learned response was to skip it.

Variable-ratio reinforcement, the schedule that makes unpredictable rewards so sticky, doesn't apply here. Our poll was perfectly predictable. Predictability is efficient for the sender and boring for the reader.

What the plateau actually tells you

The 19% wasn't a lie. It measured genuine curiosity about a new thing. The mistake would be reading it as evidence that polls lift engagement, then building a content calendar around polls.

A cleaner interpretation: the lift came from novelty plus low friction, and we only controlled for one of those. Friction stayed low across all four sends. Novelty didn't.

This is the same trap people fall into with subject-line testing generally. A pattern interrupt works once. Run it weekly and it becomes wallpaper. The metric you should watch after any format succeeds isn't the lift—it's the decay rate on the second and third repetition. A format that holds 60% of its lift after three sends is worth keeping. One that drops to zero is a one-off, and you should treat it as a lesson about your audience's attention, not as a channel.

Where to take this next

Stop optimizing the poll and start optimizing the surprise budget.

Concretely: keep a running list of low-friction interactive formats—binary polls, one-question quizzes, "pick one of these two" comparisons—and rotate them so no single format appears more than once a quarter. Track the decay curve per format, not per send. If a format's second outing produces less than half the first lift, retire it for six months before testing again.

The forward-looking question isn't "do polls work in email?" It's "how fast does my audience learn my tricks, and can I stay ahead of that curve?" For most lists, the honest answer is: faster than you'd like, and only if you keep changing the shape of the ask rather than the wording.