The same rule, stated twice
Prompts grow by accretion. Instructions get re-added rather than edited.
This is the single most common expensive problem in a mature prompt, and it is invisible from inside. A rule gets added in January. In March somebody hits the same failure, cannot find the existing rule in four hundred lines, and adds it again in different words. By August the prompt says "always cite sources" in five places.
The cost is duplicated tokens on every request. The subtler cost is that near-duplicate instructions drift apart over time until they contradict each other, and then the model has to pick a winner.
Deduplicating is the highest-value edit available, but it is the one you should review by hand. Two sentences that look alike to a similarity metric occasionally differ in a way that matters, so this rule is off by default and only runs when you ask for an aggressive rewrite.
Always cite your sources.
...
Remember to always cite your sources when answering.Always cite your sources.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
- 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.
- A long list of things not to doProhibitions accumulate one incident at a time and are rarely removed.