How to Tailor Your Resume to a Job with AI

A universal resume is a resume optimised for nobody. It reads as competent and forgettable, which in a stack of three hundred is the same as being rejected — except slower, and without the feedback.

This is part two of a two-part method. In part one we found the roles worth applying to. Here we take one of them and rewrite the resume for it.

One warning before the method, because it decides whether any of this helps you. Tailoring means changing emphasis, order and language. It never means adding experience you do not have. A resume that wins an interview you then cannot survive has cost you more than a rejection would have.

The seven steps

1. Decode the posting into what is actually required
2. Run an honest gap analysis
3. Rewrite the bullets, outcome first
4. Use their words, not yours
5. Rebuild the top third of page one
6. Make sure a machine can read it
7. Review as an adversary
The prompts · Before you send · Common questions

Who actually reads your resume, and in what order

Tailoring makes sense only once you know what you are tailoring for. There are three readers and they want different things.

Each step below is aimed at a specific one of these. When a step feels like busywork, check which reader it serves — that is usually the part that was skipped.

Step 1. Decode the posting into what is actually required

What you do: give the model the posting and ask it to separate the real requirements from the wallpaper.

Why: job descriptions are written by committee. Some lines are what the team genuinely needs, some were copied from the last posting, and some are aspiration. If you weight them equally you will spend your best space answering the least important line.

What to ask for: the three to five capabilities the role genuinely turns on; the requirements that are real but secondary; the boilerplate; the problem this role exists to solve, inferred from the whole text; and the exact vocabulary the company uses for the work. That last one matters more than it sounds — see step four.

Step 2. Run an honest gap analysis

What you do: put your resume next to the decoded requirements and sort every one into three buckets — evidenced, real but invisible, and absent.

Why this is the step that decides the outcome: most people skip straight to rewriting, and end up polishing the things they had already said while the things they did but never wrote down stay missing. The second bucket — real but invisible — is where almost all the available improvement lives. It is work you genuinely did that your current resume does not mention or buries in a line about "responsibilities".

Then count. If more than about a third of the core requirements land in absent, this is a stretch application — fine to send, but do not spend three hours on it. If nearly everything is evidenced, the role may be below your level. The count is information; use it.

Step 3. Rewrite the bullets, outcome first

What you do: take the invisible bucket and rewrite each item as a bullet that leads with what changed, then how, then at what scale.

Why the order matters: a reader scanning for thirty seconds gets the first half of your line. If the outcome is at the end, it is not read. "Responsible for the migration of the billing service" and "Cut billing incidents by 70% by migrating the service off the legacy queue" describe the same work; only the second one survives a scan.

On numbers, honestly: not everything is measurable, and invented precision is worse than none. When you do not have a metric, scale still works — team size, request volume, number of services, how long it had been broken, how many people were unblocked. "Reduced onboarding for new engineers from two weeks to two days" is a number. So is "the first person to own this system across four teams".

Say what you personally did. Senior resumes fail most often on this. "We rebuilt the platform" tells a hiring manager nothing about you. Led, designed, decided, convinced, wrote — the verb is the content.

Step 4. Use their words, not yours

What you do: where you and the posting mean the same thing by different words, switch to theirs.

Why: the filter matches literally, and the recruiter is scanning for the terms in the requisition. If they say "observability" and you say "monitoring", you match a human's understanding and fail a machine's. This is not keyword-stuffing — the underlying claim does not change. Only the label does.

Where the line is: swap a term when it names the same work. Do not add a term for work you did not do, and do not paste a skills block of everything in the posting. Both are visible, and the second is visible to humans, who read it as desperation.

If your LinkedIn says something different from your resume, fix that too — recruiters check, and a mismatch reads as carelessness. That is what the LinkedIn packaging service exists for.

Step 5. Rebuild the top third of page one

What you do: rewrite the headline and summary for this specific role, and reorder what follows so the most relevant experience appears first.

Why: this is the only part you can rely on being read. A generic summary — "results-driven engineer passionate about technology" — spends your most valuable space saying nothing. Three lines that state your level, your domain, and the one thing you are demonstrably good at will do more than the next two pages.

Reordering is legitimate and underused. Within a role, put the bullets that match this posting first. Across roles, if the relevant work was three jobs ago, a short "selected relevant experience" block at the top solves what reverse chronology hides.

Step 6. Make sure a machine can read it

What you do: check the mechanics before you send.

Why: a resume the parser mangles is rejected by nobody — it simply never becomes a candidate. This is the cheapest failure to avoid and the most annoying to discover late.

Step 7. Review as an adversary

What you do: before sending, have the model argue against you — and then check its arguments yourself.

Why: you have been staring at this for an hour and you can no longer see it. A model asked to be critical will find the vague bullets, the unsupported claims and the questions an interviewer would open with. Asked to be helpful, it will tell you it looks great.

The last pass is yours and cannot be delegated: read every line and ask whether you can defend it under questioning. Anything you cannot, cut. This is the step that keeps AI-assisted resumes honest, and it takes five minutes.

The prompts

Paste the posting and your resume alongside these. Note how often they tell the model not to help — that instruction is doing most of the work.

Before you send

Common questions

Should I tailor my resume for every job application?

For every role you actually want, yes. A universal resume is optimised for nobody and reads as competent and forgettable. Tailoring means changing emphasis, order and language — never adding experience you do not have. If a role is a stretch, send it, but do not spend three hours on it.

Is using AI to write my resume cheating?

Using it to phrase and structure what you actually did is no different from asking a friend to help you word something. Letting it invent experience is a different thing, and it fails in the interview rather than at the resume stage. Ask the model explicitly to mark gaps rather than fill them.

How do I get my resume past an ATS?

Mostly by not breaking the parser: one column, standard headings such as Experience and Education, contact details as text rather than in a header, no tables and no text inside images. Then use the posting's own terms where they mean the same as yours. Copy your PDF into a plain text editor — what you see is roughly what the parser sees.

What if I do not have numbers for my achievements?

Invented precision is worse than none. When you have no metric, scale still works: team size, request volume, number of services, how long something had been broken, how many people were unblocked. "Reduced onboarding from two weeks to two days" is a number, and so is "first person to own this system across four teams".

How long should tailoring one resume take?

About an hour done properly, which is exactly why people stop after four applications. That is the argument for choosing the roles carefully first rather than applying widely.

Where this is going

An hour per application is the honest cost of doing this properly, which is exactly why people stop after four.

So we are building it into mentors.coach as a conversation: open a vacancy, upload your resume, and work through the rewrite with a bot that has already decoded the posting. It asks you the questions this guide tells you to ask yourself — what was the result, what did you personally do, what number can you stand behind — and gives you back a resume for that specific role.

A dialogue rather than a button, and deliberately so. The gap analysis only works if someone can say "yes, I did that, here is the number" or "no, I did not" — and only you can say either. That is also the guardrail: a bot that asks cannot invent, whereas a button that generates will. Join the waitlist to be told when it ships.

If you would rather not wait, this is what the resume rebuild and resume boost do with you rather than for you — and the prices are on one page, with no add-ons at checkout.

And if you have not chosen the roles yet, start with part one. A perfectly tailored resume sent to the wrong role is still the wrong application.