Skills/Resumes and job search
5 skills · 0 measured · no account, no upload, nothing to buy

Claude resume and job search skills, without the folklore

Applications, interviews and offers, built on what applicant tracking vendors and employment regulators actually publish, which contradicts most of the advice in circulation. How we test →

All 5 shown

5 skills we have not measured yet

Written, reviewed and free to take. No run behind them, so no claim about what they do to an output. What that means.

Resumes and job searchNot yet measured ATS resume tailorer Separates the knockout questions that actually reject applications from the match score that only sorts them, then rewrites the CV against what parsers and recruiter search are documented to do. skill · 3,334 words · MIT Read the write-up
Resumes and job searchNot yet measured Recruiter and hiring manager outreach writer Builds the exact search surface that makes you findable, then rations a free account's few 200-character notes across the targets whose trigger actually justifies an approach. skill · 3,207 words · MIT Read the write-up
Resumes and job searchNot yet measured Role-matched cover letter writer A six-branch rule for whether a letter is worth writing at all, then one claim attached to evidence a reference check could confirm, inside a stated word budget. skill · 3,195 words · MIT Read the write-up
Resumes and job searchNot yet measured Salary negotiation script writer Builds the script around who is legally obliged to name a number first and who is barred from asking your pay history, then scripts the exchanges that follow from that. skill · 3,197 words · MIT Read the write-up
Resumes and job searchNot yet measured STAR interview story bank builder Turns a career history into 8 to 12 indexed stories, each mapped to every competency it can serve, plus a coverage table naming the competencies with nothing behind them. skill · 3,199 words · MIT Read the write-up
no account, no upload, nothing to buy

The list

Two claims sit underneath almost every piece of resume advice written in the last decade. The first is that around seventy five per cent of applications are rejected by software before any person sees them. The second is that you should not submit a PDF because the software cannot read it.

Both trace to the same interested party: Preptel, a company selling resume-optimisation software to job seekers. The rejection figure reaches us through a CIO.com article of 1 March 2012, and the original sentence is not the one in circulation. What Preptel actually claimed was that these systems "kill 75 percent of candidates' chances of landing an interview", which is a claim about interview probability made by a vendor with something to sell, not a claim that three quarters of applications are automatically rejected. The number mutated in transmission and has been quoted in its mutated form ever since. The PDF claim comes from a Preptel blog post of 26 May 2011 asserting that PDFs "can't be read correctly by about 90% of the software being used in the employment market", with a journalist rather than a test as its stated basis, and it predates essentially every parser now in use. No methodology, no sample and no list of software tested was ever published for either figure. Preptel went out of business in August 2013. Today PDF sits on the accepted list of every major vendor, and Ashby's documentation recommends converting to it.

Replacing folklore with documentation changes the advice substantially, and in one case it inverts it. The system that scores your application and the system that rejects it are different systems. Every vendor examined documents automatic rejection running on structured application-form answers, the yes-or-no and multiple-choice questions, and both Greenhouse and Lever state in writing that their resume-matching AI does not read those answers at all. Greenhouse goes further and says its matching "does not automatically advance or reject candidates". So the knockout questions are where an application dies, and the wording of the CV is a ranking and findability problem instead. That is a different job, done in a different place, with different rules.

The formatting advice, which sophisticated readers have learned to dismiss as superstition, turns out to be the part with real vendor documentation behind it. Greenhouse publishes its own list of what breaks parsing: columns, tables, content in headers and footers, graphics, letters with spaces between them, and abbreviated job titles such as "Sr. Account Exec" rather than "Senior Account Executive". There are also silent size traps nobody mentions, where a file uploads successfully and is never parsed because the parsing limit is far below the upload limit.

The last thread is that candidates have documented rights almost no guide mentions. Several jurisdictions now require the salary range, ban questions about your pay history, or require an employer to disclose that AI screened you and to let you refuse. Those rules are dated, they differ sharply by country and state, and knowing which one applies to you is worth more in a negotiation than any script.

Every figure in these files carries the date it was checked, and anything we could not verify is named as unverified rather than repeated.

What the two labels mean

Some of these carry a number. Most do not, and they say so.

Measured means the skill was given a realistic task on real material, then the identical task was run again with the skill removed, five runs each way. Each output was graded alone, against a rubric written by someone who had never seen the skill, by a session that was not told the other arm existed. Whether it passed was decided by a rule written down before any run executed. Those pages carry the worst case, the median, the p value, and what the skill costs you as well as what it buys.

Not yet measured means exactly that. It is written, it has been read, it is free to take, and we have run no experiment on it, so we make no claim about what it does to an output. It is not a skill that failed. Skills that failed are not published at all, in either state, and their numbers are in the results table.

Measuring one skill properly costs roughly twenty model sessions. We are working down the queue and moving skills from the second group into the first. Read the full method, or go back to all skills.