Your first prompt is usually your worst because you send it before your brain fills in the details — the format, the audience, the constraint that rules out the obvious answer. The model then guesses at all of it, and you spend the next three messages correcting a gap you could have closed in one line.
Call it the re-run tax. It’s small each time, which is exactly why it’s easy to ignore. Multiply one wasted round-trip across every prompt across a year and it’s a genuinely large amount of time and focus spent re-explaining yourself to a machine.
Why the gap keeps happening
It isn’t a knowledge problem. Most people already know to state the format and lead with the request. They skip it under time pressure because nothing is there to catch it. The prompt box is a blank rectangle with a send button, and the send button is always the fastest thing to click.
So the fix isn’t "learn prompting." You already know enough. The fix is moving the check to before you send, when the prompt is still cheap to change.
The 2-minute fix: a pre-send pass
Before you hit send, read your draft once and ask three questions. It takes about ten seconds and saves the round-trips.
- Format: did I say what shape the answer should take? (bullets, table, word count, JSON)
- Context: does it include what the model can’t guess? (audience, constraints, what I tried)
- Request-first: is the actual ask in the first line, or buried under setup?
If all three are yes, send it. If one is no, that no is almost certainly the reason your next answer would have been mediocre.
| Approach | Time before a good answer | Round-trips |
|---|---|---|
| Send the first draft, then fix | 3–5 minutes | 3–4 |
| Ten-second pre-send check | Under a minute | 1 |
Can’t I just tell the AI to improve my prompt?
You can, and sometimes it helps. But asking mid-conversation adds a round-trip of its own, and the model rewrites based only on what you already typed — it can’t add the context it never had. It often makes the prompt longer and more confident without making it more specific.
The cheaper version is to catch the gap before the first send. That’s the habit BeforePrompt automates: it reads your draft in the prompt box as you type and points out the missing format or context before you commit, so the first answer is the good one. The manual ten-second check gets you most of the way there today.