Should I vibe code
Compare a resume with a user-supplied job description using transparent term and structure checks
Comparing your CV to a job ad is text analysis on two documents you already own.
?
Their verdict, the Premium price and the build-time estimate come from their entry, MIT-licensed. Checked 2026-08-03.
?
Our verdict, the regret score and everything below it. Editorial and unsponsored — nobody can pay to be moved.
The honest answer
why the verdict is what it is
No accounts, no storage, nothing at stake. You paste two things in and get a diff, which is a very good use of an afternoon and $50 a month saved.
What actually breaks
not "if". the specific failures.
- The premise, because a match score is a guess about an opaque system nobody outside the employer can see
- Keyword stuffing, which the score rewards and human readers immediately notice and dislike
- Optimising toward a claim about applicant tracking systems that varies enormously between employers
- Parsing your own CV, since PDF layout makes columns and tables read in an order you did not intend
- Encouraging you to describe experience you do not have, which is the one genuinely bad outcome here
The stakes are your own job search and the failure is self-inflicted. You optimise toward a high match score by working every phrase from the advert into the CV, and it works — the document contains everything the score wanted. A human reads it and finds a page that reads like the job advert rather than like a person, with a skills section listing eleven things in an order that fools nobody. The score went up. The thing the score was a proxy for went down.
Is that you?
the verdict is a default, not a law
- It reports overlap and gaps descriptively rather than producing a score to maximise
- It runs locally on two documents you already have
- You use it to find genuine gaps rather than to reach a number
- It produces a single score presented as a probability of success
- It suggests adding terms for experience you do not have
- You are optimising a document a human will read for a machine you cannot see
- It uploads your CV somewhere to do arithmetic on two text files
If you build it anyway
the checklist, then the prompt that enforces it
- Report overlap and gaps, not a score. A number invites maximising, and the thing being maximised is a proxy nobody can verify.
- Never suggest adding a term the person cannot substantiate. The tool's only genuinely harmful output is encouraging an untrue CV.
- Run locally. This is text comparison on two documents you already have, and uploading a CV to do it is an unnecessary transfer.
- Extract text from the CV the way a parser would, and show that extraction — a two-column layout reading out of order is a real and fixable problem.
- Frame findings as prompts for the human: 'the advert emphasises X, your CV does not mention it — do you have that experience?'
- Do not claim to model any specific applicant tracking system, because the behaviour differs per employer and per configuration.
Before you build a CV and job description comparison tool, apply these and push back if I ask you to break them. 1. Report overlap and gaps descriptively rather than producing a single match score. Tell me a number invites optimisation, that the number is a guess about an opaque and highly variable system, and that optimising it produces a document that reads badly to the human who decides. 2. Never suggest adding a skill, tool or phrase that I cannot substantiate. If a gap is genuine, say the gap exists — do not offer wording to conceal it. This is the only output of this tool that can actually harm me. 3. Run everything locally. This is text analysis on two documents I already have, and there is no reason for my CV to leave the machine. 4. Show me how the CV text was extracted, in order, so I can see whether a two-column layout or a table reads the way I intended. That extraction issue is real and fixable, unlike the score. 5. Frame every finding as a question to me: the advert emphasises this, your CV does not mention it, do you have that experience. 6. Do not claim to model a specific applicant tracking system. Behaviour varies by vendor, version and employer configuration, and any such claim is marketing. 7. Distinguish required from preferred requirements in the advert where the text allows. 8. Store nothing between runs unless I ask. 9. Out of scope unless I ask: rewriting my CV, generating cover letters, salary estimation, application tracking.
That one keeps you out of trouble. For the prompt that actually builds it, canivibecodeit.com has one.
their build prompt ↗Or don’t build it
the boring option, and the way back out
$49.95 a month is a lot for text comparison on two documents you already own, which is what the hot take is pointing at. Build it locally, use it to find real gaps, and be sceptical of any tool in this category that reports a confident score.
Nothing accumulates — it compares two documents and reports. Keep your CV in a plain source format you control, which matters far more for your job search than any tool that reads it.
Active open-source resume builder with structured data and PDF export.
Questions
Why is a match score worse than a list of gaps?
Because a score is a target. Once there is a number, the natural response is to raise it, and the cheapest way to raise it is to work the advert's vocabulary into the document — which is exactly what makes a CV read badly to a person. A list of gaps prompts a decision; a score prompts optimisation toward a proxy nobody can validate.
Is the extraction-order problem actually common?
Yes, and it is the one genuinely useful thing this category does. Two-column CV templates, text in tables, and headers in text boxes frequently extract in an order that scrambles the reading. Showing the extracted text is a real check with a real fix — unlike the score, it tells you something verifiable about your document.
Every week, someone ships something they shouldn’t have.
New verdicts, the worst thing that landed in the trap, and the occasional incident report. No other email, ever.
last reviewed 2026-08-03 · verdict is editorial and unsponsored · shared entry data from canivibecodeit under MIT · not legal advice