Stop trusting AI to "be honest" about your resume because it

I put this to the test with a fintech backend role. I had three candidates: Priya (the hidden gem who meets every unwritten need but has "wrong" keywords), Daniel (the keyword-perfect candidate who has never actually handled live money), and Marcus (a front-end dev as a control). When using a standard prompt, the AI consistently overrated Daniel because he checked the explicit boxes, while missing the red flags that a seasoned hiring manager would spot instantly.

The fix is a two-step AI workflow. Instead of feeding the CV immediately, you force the LLM to build a "hidden rubric" first. By analyzing the job ad in isolation, the AI can infer the implicit expectations—the things a manager wants but didn't write down—before it ever sees a candidate's name. This prevents the model from simply mirroring the CV's strengths back to you.

The Prompt Engineering Breakdown

If you want a real-world gap analysis, you have to separate the rubric creation from the evaluation. Here is the exact process I used to break the pleasing effect.

Step 1: Extract the Implicit Rubric
Run this prompt with only the job description. No resumes allowed yet.

Here is a job description.

[job description]

Do NOT evaluate any candidate. Reconstruct the IMPLICIT expectations
behind this posting — the things that are NOT written down but are
actually in the head of the hiring manager and whoever else signs off
on this hire. What would make someone succeed or fail here that the
ad never says out loud? Read between the lines: the wording, the
omissions, the ordering, the company stage, the "nice to haves."

Then build an evaluation scale:
- The unwritten must-haves, and why each matters to THIS org
- Which stated requirements are actually soft, and which stated
 "nice to haves" are actually hard requirements
- Disqualifiers that would never appear in the posting
- For each dimension: what a VERY GOOD match looks like, what a GOOD
 match looks like, and what a BAD match looks like — in terms of
 evidence you could actually find on a resume
- Relative weighting across dimensions

Step 2: The Evidence-Based Audit
Now, feed the generated scale and the resume back into the model.

[job description]
[the scale from step one]
[resume]

Place this candidate on the scale. Go dimension by dimension and cite
the specific resume evidence for each placement. Where the scale
asserts a criterion the job description doesn't support, say so and
discount it.

This method transforms the AI from a polite assistant into a critical auditor. By establishing the "success criteria" before seeing the candidate, the LLM is far more likely to flag the difference between someone who just knows the tech stack and someone who can actually do the job.