I Changed One Word on a Button and Conversions Went Up 20%

TLDR: I once changed a single word on a button, from "could you try it" to "would you try it," and conversions went up 20%. Twenty percent, from one word. The lesson is not about that word. It is that you cannot find results like this in a brainstorm or a prompt. You find them by testing.

One word, twenty percent

"Could you try it" is a question about ability. "Would you try it" is a question about willingness. That tiny shift reframed the ask, and the number moved hard. If you had put both options in a meeting and asked people to vote, nobody would have predicted the gap. That is the whole point.

Why brainstorms and LLMs cannot give you this

A brainstorm produces opinions. A large language model produces the most probable phrasing based on everything it has read. Both are useful for generating candidates. Neither can tell you what your specific audience will actually do, because that lives in behavior, not in consensus or in training data. The only thing that settles it is putting both versions in front of real users and counting.

This is exactly why retention work has to be run as experiments, not guesses. The tactics that move your churn curve are often small, unglamorous, and impossible to predict in advance. You have to ship the test.

How to actually run these

1. Pick one thing at a time. One word, one headline, one flow step. If you change five things and the number moves, you learn nothing about why.
2. Decide the metric before you start. Conversion, activation, retained-at-30-days. Write it down first so you cannot rationalize afterward.
3. Run it long enough to trust it. A 20% lift on 40 visitors is noise. Give it real volume before you call it.
4. Keep the winners in a log. Small wins compound. Ten of these a year is a different business.

If you want a structured way to do this for retention specifically, the ship retention experiments in 15 days playbook lays out the cadence, and the full experiments library gives you tested tactics to put through your own A/B pipeline. To sanity-check whether a lift is even worth chasing, the MRR Impact Simulator shows what a few points of improvement is worth at your scale.