How to Tailor a Resume Without Lying: Rules for You and Your AI

Tailoring a resume is not lying. Choosing which true things to say, in which order and in whose vocabulary, is what every good resume does. The line gets crossed in small steps: a title nudged up, a tool you watched someone else use, a percentage that "feels about right". AI tools make those steps easier, because they are built to produce plausible text and a job posting is full of plausible material. This post sets out where the line is and how to make an AI stay behind it.

#The line: what tailoring may change

The title row is the one people most often get wrong, because the line there is narrow. Changing a job title for ATS walks through what you may add next to a title and what you may not.

A simple test for every line: could you describe the work behind it, in detail, to the person who managed you at the time, without them correcting you? If yes, it is tailoring.

#Why AI tools cross the line

Language models are trained to produce fluent, likely text, and a gap in your resume next to a requirement in a posting is an invitation to fill it. A 2025 paper by Kalai, Nachum, Vempala and Zhang, Why Language Models Hallucinate, describes the mechanism plainly: models "sometimes guess when uncertain, producing plausible yet incorrect statements instead of admitting uncertainty."

In resume tailoring, those guesses take predictable forms:

• Term transfer. The posting says "Kubernetes"; your resume says "deployed services to AWS". The rewrite says "deployed containerized services with Kubernetes on AWS".
• Number invention. The posting rewards impact; the rewrite adds "improving efficiency by 30%".
• Level inflation. "Supported the migration" becomes "led the migration".
• Title drift. "Analyst" becomes "Senior Analyst" because the posting is for a senior role.

The pull is strongest for terms that appear in many postings. In the ATS Resume AI analysis of all 503 S&P 500 careers sites and 9,260 live job postings, June 2026 (role data), "AI / LLMs" appeared in 13.1% of software engineer postings and 23.8% of data scientist postings. A model that has seen that term next to those roles many times will reach for it, whether or not you have done the work.

#How a bracket guard works

The fix is to separate two jobs that a loose prompt mixes together: wording, which AI is good at, and asserting facts, which only you can do. A guard enforces the separation by checking every term in the output against the source resume.

Anthropic's guidance on reducing hallucinations recommends the same pattern for any document task: have the model find a supporting quote for each claim, and "If it can't find a quote, it must retract the claim." Applied to a resume, the logic is:

1. 1

   Extract the posting's terms

   Skills, tools, certifications, responsibilities and the job title.

2. 2

   Trace each term to the original resume

   If the term, or a plain synonym of it, appears in your resume, it may be used.

3. 3

   Bracket what cannot be traced

   A posting term with no trace comes back as a question, for example [confirm: Kubernetes], not as a claim.

4. 4

   Bracket missing numbers

   Where a bullet would be stronger with a figure you did not give, the output says [add: share of tickets resolved] instead of inventing one.

5. 5

   You resolve every bracket

   Replace it with a true fact, or delete it. Nothing in brackets is ever sent.

This is how our resume optimizer handles it: posting terms it cannot trace to your resume come back as [confirm: …], and missing numbers as [add: …]. You can get the same discipline from any chat assistant with the prompt below.

#The truthful-tailoring prompt

Tailor my resume to the job posting below, following these rules exactly.
 
You may:
- reword my bullets to use the posting's terms, when my resume shows I
  did that work;
- reorder bullets and skills;
- leave out content that is not relevant.
 
You may not:
- change any job title, employer, date, degree or certification;
- add any tool, skill or responsibility that does not appear in my resume;
- add or change any number.
 
When the posting asks for something my resume does not show, write
[confirm: the term] at the point where it would go, so I can check it.
When a bullet would be stronger with a number I did not give, write
[add: what number] instead of inventing one.
 
After the resume, list every [confirm] and [add] marker with the line it
appears in, and for each [confirm] quote the closest evidence from my
resume, or write NONE.
 
POSTING:
[paste]
 
MY RESUME:
[paste]

The list at the end is the review sheet. Work through it line by line before the resume goes anywhere.

#A worked example (fictional sample)

"Sample: D. Sato" is a made-up IT support specialist applying for a Systems Administrator posting that mentions Linux, Ansible, incident management and "reduced downtime".

Original bullet (sample):

Handled server tickets for 300 staff; wrote shell scripts to automate user setup on Linux servers.

Loose AI rewrite (what to avoid):

Administered Linux infrastructure with Ansible for 300 users, reducing downtime by 40% through proactive incident management.

Guarded rewrite (sample):

Resolved server incidents for 300 staff and automated Linux user provisioning with shell scripts [confirm: Ansible], cutting setup time from [add: before] to [add: after].

The loose version invented a tool, a percentage and a duty. The guarded version keeps the facts, uses the posting's vocabulary where true ("incidents", "provisioning"), and asks. If D. Sato has never used Ansible, the bracket is deleted; if he has, he adds it to his master resume first.

#Where the stakes are written down

For most private jobs, a false resume claim is an employment matter: it can end an offer or a job when it surfaces. For U.S. federal applications the stakes are written into law. The federal false-statements statute, 18 U.S.C. § 1001, covers anyone who knowingly and willfully "makes any materially false, fictitious, or fraudulent statement or representation" in a matter within federal jurisdiction, with penalties of up to 5 years' imprisonment.

Federal resumes are also built to be checked. OPM's applicant guidance asks for start and end dates in month/year and the hours worked per week for each job, which are facts an employer can verify against records.

For more prompts that keep AI output grounded in your resume, see ChatGPT prompts to tailor a resume.

#Sources

• Kalai, Nachum, Vempala and Zhang, Why Language Models Hallucinate, arXiv, 2025
• Anthropic, Reduce hallucinations
• Legal Information Institute, Cornell Law School, 18 U.S. Code § 1001
• U.S. Office of Personnel Management, Applicant Guidance on the Two-Page Resume Limit