How honest AI resume tailoring works

October 16, 2026 · by the Offerkin team

Ask an AI to “tailor my resume to this job” and you get one of two failure modes: a timid nothing (“consider emphasizing collaboration”), or a confident fabrication (“Reduced PPA by 40%” — you did no such thing). The second one can cost you an offer and your credibility in the interview. Here is the architecture we built to make the ambitious version safe.

1. The AI proposes; code disposes

A frontier model (Claude Sonnet) reads your resume, the job posting, and your saved answers, and proposes structured edits: rewrite this line, reorder this list, add this one. But nothing it proposes reaches your resume directly. Every proposal passes a validator that runs before you ever see it:

Grounding: every content-bearing word in a new line must already exist in your cited resume lines or saved answers — including soft claims like “led”, “owned”, “end-to-end”.
Numbers: a rewritten line may only contain numbers that appear in the line it replaced or its cited evidence. New metrics are impossible by construction.
Protected lines: your name, contact, job titles and dates are never edited.
No silent losses: a rewrite can't quietly drop a skill the job asks for.

Proposals that fail are dropped with a reason — you can even see the top rejection reasons in our admin dashboard. The honest gap between you and the job stays visible instead of being papered over.

2. Every change shows its receipt

What survives comes to you as a diff with an evidence card: what the employer is looking for (quoted from the posting), why this change helps, and the exact resume line or saved answer each addition came from — with a “view source in resume” jump to the cited line. Rewrites highlight only the changed words; reordering is marked as movement, not painted as fake deletions.

3. You are the final gate

Accept or reject each change individually. Rejections stick: the tailored resume only ever contains what you approved. Change your resume, answer a new gap question, or re-run tailoring — the tool tracks which version of your background every decision was made against and refuses to apply stale ones.

4. The gaps you can't fake

The most useful output is often the list of what we didn't write: “the posting asks for wafer fab yield modeling and should-cost analysis — not in your resume; if any of it is real for you, say so and it goes in.” Tailoring can reposition real experience, but it can't manufacture it. A tool that pretends otherwise sets you up to fail the interview.

Can recruiters tell an AI-written resume?

Increasingly, yes — and not because of AI detectors. Experienced reviewers pattern-match the symptoms of generated text, and an invented resume has tells that are hard to fake your way past:

Metrics without a story. “Improved throughput by 37%” invites exactly one interview question — “how?” — and a fabricated number collapses there.
Uniform phrasing across bullets. Every line the same length, the same rhythm, the same “Spearheaded… leveraged… orchestrated” register. Real resumes are lumpier.
Skills the interview exposes. The resume claims Kubernetes; the screening call asks one question about it. This is the tell that costs offers.
A resume that doesn't match anything else — your LinkedIn, your portfolio, your answers.

The tells aren't “AI was involved.” A tool that repositions experience you actually have produces none of them. The tells are unsupported claims — which is exactly what validation removes.

What honest tailoring actually changes

Tailoring that stays honest is still a big lever — it just works on arrangement and framing instead of invention:

Order: the bullets this employer cares about move to the top of each role.
Framing: the same work described in the posting's vocabulary — your “built caching layer” becomes the “reduced API latency” they asked about, if the evidence supports both readings.
Selection: the skills line leads with the overlap, and irrelevant lines get out of the way.
Headline: your title line reflects the role's seniority and focus, supported by the body.

What it never does: add tools you haven't used, numbers you can't source, titles you never held, or scope you can't defend. The gap between you and the posting stays visible — as questions you can answer once (“do you have Terraform experience?”) and then reuse forever, not as lines the resume pretends already existed.

Try it

Paste any job posting link into Offerkin and open the tailored resume: you'll get the full “what changed and why” recap, the per-change evidence, and the honest gap list. Three AI tailors a month on the free plan.