Fix the cover letter prompt
Fix the cover letter prompt
A friend is applying for a product analyst role and asks for help with the AI prompt they have been using. Their last two letters read well in the moment, but in one interview they were asked to explain a claim they could not back up.
Rewrite the prompt so the model tailors the letter to the posting using only real evidence — and asks for missing facts instead of inventing them.
The weak prompt
Take my CV and write a cover letter for this product analyst job. Make it really impressive and confident — add anything that would make me stand out, and make my experience sound more senior than my CV shows. Don't worry about length.
A strong version (self-assessment — not machine-graded)
Here is the job posting's requirement list and my evidence bank. Map each requirement to the closest evidence, or mark it as a gap — do not fill gaps with anything I did not supply. Then write a 210-word cover letter that uses only the evidence bank: same facts, same scale, nothing upgraded or stretched. If a requirement needs a fact I have not given you, ask me for it instead of inventing one. Finish with the three strongest requirement-evidence pairings as a short bullet list, and flag any sentence in the letter you could not trace back to the evidence bank.
Hint
Replace every invitation to embellish (“impressive”, “more senior”) with an instruction about fidelity.
The model needs two artefacts to be honest: the requirements list and the evidence bank — name both.
Why this is the answer
The weak prompt asks for confidence and authorises invention — and the failure shows up weeks later, in an interview. The fixed prompt changes the job from “make me sound good” to “map my real evidence onto their requirements, and say what is missing”. That is why the criteria matter together: the boundary keeps claims honest, the posting list keeps the letter relevant, the evidence bank gives the model legitimate material, the format keeps it readable, and the self-check catches the sentence you would have been asked about.