A long list of things not to do
Prohibitions accumulate one incident at a time and are rarely removed.
Every production prompt grows a "never do this" section, one postmortem at a time. Nothing ever gets deleted from it, because deleting a prohibition feels like inviting the incident back.
The token cost is the visible half. The other half is that a long list of negatives describes the space of wrong answers rather than the shape of the right one, and models follow a positive specification more reliably than a negative one.
Convert where you can. Ten rules about what not to include in a summary usually collapse into one sentence describing what the summary should contain. That is shorter and works better.
This is advisory: rewriting prohibitions correctly needs judgement about your task, so the optimizer will not attempt it.
Do not use bullet points. Never exceed 200 words. Do not mention pricing. Never use jargon. Do not speculate...Write one paragraph under 200 words in plain language, covering only what the ticket states.Check your own prompt
The analyser checks this pattern along with the other 25, prices each finding against your request volume, and hands back a rewritten prompt. It runs in your browser — nothing is uploaded.
Run the analyserMore on structure
- The same rule, stated twicePrompts grow by accretion. Instructions get re-added rather than edited.
- More examples than the model needsFew-shot examples are the most expensive content in most prompts, and they have a ceiling.
- Encoded data pasted into the promptBase64 and data URIs tokenize terribly — roughly one token per two or three characters.