What actually filters you out
In the next ten minutes you will have a list of the exact words one real employer used that your CV does not — the highest-value edit available to you today.
You need two things open: a job posting you would genuinely take, and your current CV. Rough is fine. Nothing else.
Run this first
Paste the whole posting, including the boring parts at the bottom, then your CV.
Compare my CV against this job posting.
Give me exactly three things:
1. The 10 exact words or phrases the posting uses that my CV never uses.
Their spelling, their casing. Quote them.
2. For each one, whether I have actually done that thing based on my CV.
Answer "yes", "partly", or "no evidence" — do not be generous.
3. The one line of my CV that is furthest from what this posting asks for.
Use only what is written in the two documents. If something is not stated,
say "not stated" rather than guessing.
THE POSTING:
[paste]
MY CV:
[paste]You should see
Ten quoted phrases, an honest yes / partly / no-evidence against each, and one line of your CV singled out. If it starts praising your CV or suggesting rewrites, say "just the three lists, no advice yet".
Look at the "partly" column. That is usually the biggest one, and it is the whole opportunity: things you have genuinely done, described in words this employer does not search for.
Check it
- Ten phrases, quoted exactly from the posting rather than paraphrased
- Every phrase has a yes / partly / no evidence verdict
- The verdicts are honest — at least one "no evidence" if that is the truth
- Nothing is claimed that your CV does not actually support
- One specific line of your CV is named as the furthest off
Why that list is worth more than a rewrite
Because of how you actually get filtered. Not by a robot scoring your CV out of a hundred — that is the version everyone believes, and optimising for it makes your CV worse for the human who decides.
Four things really stand between you and a reply:
- Parsing. Your CV is turned into structured text before a person sees it. Multi-column layouts, tables, text inside images, and details stranded in headers or footers are what get mangled. Send a PDF unless the form asks for something else, and keep it one column.
- Knockout questions. The short form after the upload — work authorisation, notice period, years in a named tool. These are automatic, and they are usually the only automatic rejection in the process.
- Recruiter search. A recruiter types terms into the applicant tracking system and reads the shortlist that comes back. You are not competing against a score. You are competing to appear in a search — which is exactly why the ten phrases you just extracted matter.
- Volume. A popular role draws hundreds of applicants. Your CV gets a scan, not a reading, and the first half of the first page is what reliably gets scanned.
Scope note: this describes the systems most Western employers use — Greenhouse, Lever, Workday. On Indian resume databases like Naukri, and on Turkish boards like Kariyer, recruiters search an even larger pool, which makes matching their vocabulary matter more, not less.
What Claude can and cannot do here
It is very good at the part you are worst at judging: reading a posting closely, spotting which of your real experience matches, and rewriting your own bullets in the employer's vocabulary.
It cannot give you experience you do not have. Every prompt in this course is built to surface and sharpen what is already true about you. If a draft starts claiming things you have not done, that is a draft to reject — not because it is dishonest in the abstract, but because you will be asked about it in the interview.
What you just did
- Extracted the ten exact phrases one employer uses that your CV does not.
- Marked honestly which of them you can actually evidence.
- Learned the four real filters: parsing, knockout questions, recruiter search, and a fast human scan.
- Established the rule the whole course runs on: sharpen what is true, never invent.
Next: reading the rest of that posting properly, so the rewrite has something to aim at.
practice
Open Claude Cowork on your own computer and try this on real files. Each module ends with a checkpoint where you submit what you made.
Open in ClaudeNext
Read the posting like a brief