TL;DR
Paste the full JD plus a short background summary into Claude (Sonnet 4.6, free tier is fine) or ChatGPT (GPT-5.5) and ask it to do four jobs at once: separate must-haves from nice-to-haves, flag hollow buzzwords, map your experience to each requirement, and predict the interview questions. The model is good at pattern-matching across thousands of JDs and bad at internal company context, so treat its scoring as a hypothesis, not a verdict. Don’t talk yourself out of applying over one gap: surveyed candidates apply at roughly 52% (men) to 56% (women) of stated qualifications, not 100%.
The task
A JD just dropped that looks like a fit at first glance. Half the bullets are buzzwords that hiring managers paste into every role; the other half is the real bar. You want to know which is which, where your background actually maps, where your gaps are, and which questions to expect in the interview. Without this triage you either talk yourself out of applying, or apply with no plan and freeze in round two.
There’s a second reason to do this carefully. As of June 2026, more than 90% of employers run applications through an applicant tracking system (ATS), and the average un-tailored resume is missing about 52% of the keywords in its target JD (Jobscan, 2026). Pulling the real requirements out of a JD is the first step to a resume that clears the keyword filter and reads well to the human behind it.
When AI helps, and when it does not
AI is excellent at separating signal from buzzword (it has read thousands of JDs and knows the patterns), mapping requirements to your CV, and predicting interview questions. It is poor at internal context. Sometimes a “nice-to-have” is mandatory because the hiring manager wrote it casually, and sometimes a scary-looking must-have is negotiable for the right person. When the model labels something “nice-to-have,” confirm with a recruiter screen if you can.
It also cannot tell you the pay band, the manager’s reputation, or whether the team is on fire. Those come from people, not text.
Which model to use
Any current frontier model handles this well. As of June 2026:
| Tool | Free tier good enough? | Why pick it |
|---|---|---|
| Claude (Sonnet 4.6) | Yes | Strong at structured extraction and honest gap analysis; pasting a JD plus background fits comfortably in context |
| ChatGPT (GPT-5.5) | Yes on Free, fewer messages | Default since ~Apr 2026; the Thinking mode in the picker is worth it for the ranking step |
| Gemini 3.1 Pro | Yes (Google AI Pro $19.99/mo for higher limits) | Good if your resume already lives in Google Docs/Drive |
You do not need a paid plan for a single JD. Paid tiers only matter if you’re analyzing dozens of roles or want the model to read a long resume PDF plus several JDs in one session.
What to feed the AI
- The full JD text, including title, level, and location
- Your CV or a background summary (concise, not the full resume)
- The company name and team, so the model can guess at conventions
- Your real seniority, not the title you wish you had
- What you cannot move on: location, hours, comp floor, visa
- Your specific worry (for example, “I have no Kubernetes experience”) so the model addresses it directly
The prompt
Analyse this JD for me.
Company and team: [line]
My real seniority: [years, level]
What I cannot move on: [list]
My specific worry: [line]
JD:
"""
[paste]
"""
My background summary:
"""
[2-3 paragraphs]
"""
Return:
1. Must-have requirements (the ones a hiring manager will not waive) — with the JD line that signals each
2. Nice-to-have requirements (often waived for strong candidates)
3. Buzzword bullets (no real meaning — ignore)
4. Background-to-requirement map: which of my experience hits each must-have
5. Honest gap analysis: which gaps are dealbreakers, which can be talked around
6. 5 likely interview questions tailored to this role
7. The 3 questions I should ask back in the screen
8. A "do not apply" flag if there are dealbreakers I cannot move onFor senior roles, append: “Add a section on the strategic narrative: what story should I tell about my last 2 years that ties to this role?”
To prep your resume at the same time, add: “List the 15 keywords from this JD an ATS would screen on, mark which already appear in my background, and suggest where to add the missing ones without lying.”
What good output looks like
A clean 4-column table (requirement / my fit / gap severity / how to address), a buzzword list called out separately, predicted questions as a list, and a closing “should I apply” verdict. Check it against this bar:
- Must-haves trace to specific JD lines, not vibes
- Buzzwords are recognisably hollow (“rockstar,” “ninja,” “10x,” “self-starter”)
- The background map references your actual experience, not generic strengths
- Gap analysis is honest. If the model says “you’ll be fine” on every gap, push back
- Predicted questions look like questions this team would ask, not a generic top-10 list
Decoding the coded language
Models are decent at this, but know the patterns so you can sanity-check. Common JD phrases and what they usually signal (Ongig, 2026):
| JD phrase | Often means |
|---|---|
| ”Fast-paced environment” | Long hours, under-staffed, expect after-hours availability |
| ”Wear many hats” | Role isn’t defined; you’ll do work outside it |
| ”Self-starter” | Little onboarding or support |
| ”Work hard, play hard” | Long hours with the occasional pizza |
| ”Ownership culture” | Less help than you’d like; you’re on your own |
None of these is automatically disqualifying. They’re prompts to ask sharper questions in the screen.
Common mistakes
- Treating the AI’s match score as gospel. It’s a starting hypothesis
- Not feeding your real background. The model can’t map gaps without it
- Treating every “nice-to-have” as critical and burning prep time
- Skipping the “questions to ask back.” Recruiters notice the absence
- Applying without a plan for the dealbreaker gap. It surfaces in round two
- Over-trusting the keyword list. Stuff a JD verbatim and a human reviewer will notice; aim for natural coverage, not 100%
FAQ
Should I apply with a real gap? Yes, if it isn’t a dealbreaker. The “apply only at 100% qualified” idea is a myth: in a 2022 study of over 10,000 applicants, men applied at about 52% of qualifications and women at about 56% (Behavioural Insights Team). Address the gap in your cover letter or the screen.
What if the JD lists 12 must-haves? Real must-haves are usually 5 to 7. Ask the model to rank by how likely each is to be waived, and prep against the top tier.
How many JD keywords should my resume actually match? Target roughly 60-80% coverage of the meaningful terms, not all of them. Resumes that include the exact job title see far more interview invitations, so mirror the title and the top skills first.
Will an ATS auto-reject me for one missing keyword? Modern ATS tools rank and surface rather than hard-reject on a single term, but a low keyword overlap pushes you down the stack. Tailoring matters more than ever because most employers still filter on keywords.
Can the AI tell me the salary or whether the team is good? No. It can flag whether the JD hides the band or leans on red-flag language, but pay, manager quality, and team health come from recruiter screens, Glassdoor, and your network.
Related
- AI resume writing — tailor your resume to the JD
- JD matching analysis AI — deeper JD-to-resume match scoring
- Cover letter customisation prompts — a cover letter per JD
- STAR interview answers — prep behavioural answers for this role
- Mock interview AI — practise the predicted questions
- Self introduction — a 60-second opener
- Behavioural question prompts — anticipate behavioural questions
Tags: #AI writing #Job search #Workflow