Most applications you submit online are read by a machine before a person sees
them: an applicant tracking system that parses and ranks, and increasingly an
LLM that summarises and scores the parsed text. A resume a human would like can
still be mis-read, down-ranked, or flagged as generic on that first pass - and
you get no feedback about why.
You can check the machine-readable layer yourself in about 20 minutes. Here is
the method I use.
1. Do a parse test, not a "does it look nice" test
Open a fresh chat in ChatGPT or Claude. Paste the plain text of your resume
(copy it out of the PDF - if the order comes out scrambled, that is finding #1).
Then ask, one at a time:
- From this resume only, list: full name, current job title, current employer, city/country, total years of experience. Say "unclear" for anything ambiguous.
- List every job as Title | Company | start date | end date. Flag any date that is ambiguous or any gap over 4 months.
- What are the top 8 skills, based only on evidence in the bullet points (not the skills list)? Quote the bullet you inferred each from.
- What seniority level is this person? Justify using scope evidence - team size, budget, systems owned.
Score each answer 0-3 (3 = correct, 0 = wrong on the basics). If the model can't
reliably pull your title, level, or dates, neither can the screener in front of
the recruiter.
2. The format things that break parsers
These are cheap to fix and they matter more than wording:
- Text-based PDF or .docx, never a scanned image or a design-tool export that rasterises text.
- Single column. Two-column layouts frequently parse out of order.
- No tables for layout, no text in headers/footers/text boxes.
- Standard section headings ("Experience", "Education", "Skills") - not "Where I've Made an Impact".
- Dates in one consistent format on the same line as the role, every time.
- One job = Title, Company, Location, Dates, then bullets, in that order, always.
3. Make each bullet carry evidence
Screeners (human and machine) reward specificity. The formula:
[strong past-tense verb] [what you did] [a number, comparison, or result]
"Rebuilt the nightly ETL in dbt, cutting run time from 3h to 40min" beats
"Responsible for ETL." Aim for a quantified outcome in at least 60% of bullets,
and keep the most impressive two or three in the top third of the page.
4. Check it doesn't read as boilerplate
Ask the model: Does any part of this read as generic template language rather
than lived experience? Quote the weakest 3 lines. Then cut or rewrite those.
"Results-driven professional with a proven track record" is invisible; a specific
combination of domain, skill, and result is not.
Re-test after each change
Run the same parse prompts again and compare the scores. That loop - test,
change one thing, re-test - is the whole method.
I put the full version of this into a small kit ($15): a 35-point AI-screening
audit, an ATS-safe resume structure with a filled example, and the 6-prompt
parse test with a scoring sheet and a trend log. It's a method and a set of
templates - no interview or hiring outcome is promised, since that depends on
the role and the employer. If it's useful:
Resume AI-Screening Kit.
Made by the Revenue Operator project. These materials were drafted with AI
assistance under human direction.
