Most “ATS optimization” advice tells you to stuff keywords, which wrecks the resume for the human who reads it five seconds later. Modern parsers don’t need that. Workday, Greenhouse, and iCIMS all run NLP layers that already know “Python programming,” “Python development,” and “Python scripting” are the same skill. The real job is finding the honest overlap between the job description (JD) and your actual record, then making it parse cleanly. These 12 prompts do exactly that.
TL;DR
- ATS is near-universal: 90%+ of employers use one, and Jobscan’s State of the Job Search 2026 report says 99.7% of recruiters apply keyword filters. Roughly 48% now layer AI screening on top.
- Aim for a Jobscan-style match rate of about 75% (80% is ideal; many land interviews at 65%). Past ~75%, you’re keyword-stuffing, not improving.
- Format beats tricks: reverse-chronological, single-column, system font (Arial/Calibri/Helvetica) at 10-12pt, saved as a text-selectable PDF parses at roughly 97% across major systems. Use
.docxonly when the posting asks for it. - Use the right model: Claude Opus 4.7 inserts
[bracket]placeholders and refuses to invent numbers in most tailoring tests; GPT-5.5 will quietly fabricate metrics and skills if you let it. Tailor with Claude, fact-check everything.
Who this is for
Job seekers customizing a resume per role, career switchers translating prior experience into a new field, and candidates whose applications get filtered out before a recruiter ever sees them.
How a 2026 ATS actually reads your resume
It helps to know what you’re optimizing for. A modern parser extracts structured fields — name, contact, job titles, employers, dates, skills, education — by guessing structure from layout: big text is probably a heading, words close together are probably one paragraph. Two things follow from that:
- Layout is the parse. Tables, text boxes, multi-column layouts, icons, and headers baked into images all break the guess. Workday in particular parses single-column PDFs well but mangles multi-column ones, and it weights job-title-to-seniority match heavily — if your last title doesn’t map onto the target level, your score drops no matter how strong the skills are.
- The AI layer reads meaning, not just exact strings. Greenhouse AI (launched September 2025) adds match scoring and candidate summaries on top of the parse. So you don’t need the JD’s exact phrasing 40 times; you need the concept present, honestly, in standard sections the parser recognizes.
When not to use these prompts
Don’t use them to add skills you don’t have — the AI layer and the eventual human interviewer both catch it. Don’t over-engineer formatting; standard layouts parse fine. And don’t chase a 100% match rate. Jobscan’s own guidance caps the useful target around 75%; beyond that you’re writing for a robot a human will reject.
Prompt anatomy: six elements every ATS prompt needs
- Role: candidate, recruiter, or hiring manager — name the persona the AI plays.
- Context: target role, company, level, and your real background.
- Goal: one deliverable — an overlap analysis, a bullet rewrite, a format audit.
- Constraints: word cap, banned phrases, must-include facts, “do not invent numbers.”
- Tone: confident / measured / specific — pick 2-3 anchors.
- Examples: paste 1-2 of your actual bullets so the AI matches your voice, not a template.
12 copy-ready prompt templates
Placeholders use [brackets] — replace them with your real text before running. All 12 are model-agnostic; we note where one model behaves better.
1. JD ↔ resume overlap
JD: [paste JD]. My current resume: [paste resume]. Compute keyword and
phrase overlap. Output three lists: (1) terms in the JD missing from my
resume (top 10), (2) terms I can add HONESTLY because I have actually done
this work, (3) terms I should NOT pretend to have. Do not invent anything.Swap: the JD and resume text.
2. Bullet rewrite per JD
For each bullet in my resume, rewrite it to align with this JD where
relevant: (1) use the JD's verbs naturally, (2) add quantification only
where I supply a number, (3) cap each bullet at 25 words, (4) skip bullets
that aren't relevant. Output a table: original | rewritten | confidence.3. Skills section sanity check
My skills section: [paste skills]. JD: [paste JD]. Re-order so the most
relevant skills appear first, group them logically (languages, tools,
methods), and drop any skill I haven't used in 3+ years unless the JD
lists it. Keep the total under 20 distinct skills.Swap: skills, JD. (The 20-skill cap is deliberate — sprawling skill lists dilute match scoring.)
4. Action verb upgrade
Find weak verbs in this resume (helped, supported, worked on, was
responsible for) and replace them with stronger, specific verbs (led,
designed, shipped, scaled, decommissioned). Do not change any facts,
numbers, or scope.5. ATS-safe formatting audit
Audit this resume for ATS-hostile formatting: (1) tables or text boxes a
parser can't read, (2) multi-column layouts that scramble text order,
(3) icons or non-standard fonts, (4) headers baked into images,
(5) non-standard section names. Output a prioritized fix list.6. Title alignment
My real title was [real title] but this JD calls a similar role
[JD title]. Suggest 3 honest ways to surface [JD title] (e.g., a
parenthetical) without misrepresenting my level. Pick the cleanest, and
flag if the seniority doesn't actually match.Swap: real title, JD title. (Workday weights this heavily — worth getting right.)
7. Hidden-keyword check
Some advice says hide keywords in white or 1px text. Audit my resume and
flag any white, tiny, or off-page text. Recommend removal: modern parsers
catch it and some platforms flag it as a misrepresentation.8. Quantification gap
List every bullet in my resume that has no number. For each, tell me what
KIND of number would strengthen it (revenue, users, latency, headcount,
% change, time saved). Do not invent figures — ask me which numbers I have.9. Career-switch resume
I'm switching from [from field] to [to field]. Identify which of my past
experiences map to the new field, and rewrite the top 3-5 bullets so the
connection is obvious to a non-expert reader — without dropping the
original context or overstating the transfer.Swap: from field, to field.
10. Education and certs strategy
My education: [education]. Certs: [certs]. This JD requires: [requirement].
Decide for each item: (a) keep it prominent, (b) move it down, or (c) note
a cert worth adding to close the gap. Output a priority order with one
reason each.Swap: education, certs, requirement.
11. Resume length decision
I have [number] years of experience targeting [region/industry, e.g. US
tech]. Decide: 1-page or 2-page? Give me the trim list if 1-page and the
add list if 2-page. US tech generally accepts 2 pages past ~8 years; flag
if my target market expects otherwise.Swap: years, region/industry.
12. Final ATS + human readability check
Final pass on my resume: (1) generate a plain-text version — does it still
read coherently with no scrambled sections? (2) confirm every section uses
a standard heading (Experience, Education, Skills). (3) in 30 seconds, can
a human tell my current role and last 2 jobs? List anything that fails.Which model to run these with
For tailoring work, model choice matters more than people think, because the failure mode is invented facts on a document recruiters will verify.
| Model (June 2026) | Behavior on tailoring | Best for |
|---|---|---|
| Claude Opus 4.7 | Inserts [bracket] placeholders and asks for real numbers in ~80% of tests; refuses to fabricate metrics | Rewriting bullets, the honest-overlap prompt (#1), final pass |
| GPT-5.5 | Strong rewriter but will silently add metrics and skills the JD mentions but your resume doesn’t | Brainstorming verbs and phrasing — then verify hard |
| Gemini 3.1 Pro | Live Google Search baked in | Checking the company’s real name, products, and recent news |
Claude Pro and ChatGPT Plus are both $20/month as of June 2026; Google AI Pro (the tier formerly called Gemini Advanced) is $19.99. Any free tier is fine for a single resume pass — you don’t need a paid plan for this.
Common mistakes
- No real context. “Optimize my resume” with no JD, company, or level produces generic mush.
- Letting the model invent numbers. Always add “do not invent figures — ask me.” GPT-5.5 especially will fill gaps with plausible-looking metrics.
- Stuffing past ~75% match. Recruiters and the AI layer both notice unnatural keyword density.
- Non-standard section names. “What I Bring” instead of “Experience” can get the whole block ignored.
- Trusting the first draft. First passes read AI-flavored; one tightening round fixes most of it.
- Skipping the plain-text test. If the plain-text export scrambles, so does the parser’s read.
How to push results further
- Paste 2-3 of your real bullets first so the AI anchors to your voice.
- Run the tailored resume through one free ATS scanner (Jobscan, Resume Worded) to confirm the parse before you submit.
- Keep one master resume and generate per-role variants from it, rather than editing in place.
- Read the final version aloud — anything that sounds like a press release gets cut.
- Save phrasings that landed interviews into a personal bank for reuse.
FAQ
- Do I still need keywords if modern ATS understands synonyms?: Yes, but fewer. The NLP layer maps related terms, so you need the concept present in standard sections — not the exact phrase repeated. Aim for ~75% match, not 100%.
- PDF or Word for ATS in 2026?: Default to a text-selectable PDF; it parses at ~97% in Workday, Greenhouse, and Lever. Switch to
.docxonly when the posting explicitly asks, since a few legacy systems (e.g., older Taleo) parse PDFs less reliably. - Will the ATS or recruiter notice an AI-written resume?: Only when it reads generic or makes claims you can’t back up. Specific, true bullets in your own voice don’t trigger anything.
- Can AI fabricate things on my resume?: Yes. In tests, GPT-5.5 added metrics and skills that were never in the source. Claude Opus 4.7 inserts
[bracket]placeholders instead, which is why it’s the safer tailoring model. Either way, fact-check every line. - Does white-text keyword hiding still work?: No. Modern parsers strip and read hidden text, and some platforms flag it as misrepresentation. Use prompt #7 to find and remove any.
- How many skills should I list?: Keep it under about 20 distinct skills. Long lists dilute match scoring and read as padding to a human.
Related
- Resume prompts
- JD matching prompts
- Cover letter prompts
- LinkedIn bio prompts
- AI Work Presentation Outline: 10 Slides, One Message, Real CTA
- Career & Interview Prompts hub
- Case Interview Prep Prompts for Consulting and PM Loops
- Resume Keyword Matching Prompts: 15 Templates to Beat the ATS Filter
Tags: #Prompt #Job search #ATS #Resume