AI Resume Kit Built on Verified Facts Only
An AI resume workflow that cannot write anything you have not already confirmed is true
Most AI resume tools make you sound better by adding things. A tool you used twice becomes “expert”, a one-time project becomes “daily”, and a team number becomes your number.
I built this kit to do the opposite. It tailors the wording for each job, and it cannot add a single claim that is not already a verified fact from your own work history.
Does the AI resume kit create anything fake?
No. Every line on every resume and cover letter traces back to a fact the person has already confirmed: their real titles and dates, the work they actually did, the skills they actually use, and the numbers they can source.
The AI chooses which verified facts fit a posting and how to word them. It does not decide what is true.
What the kit is allowed to change
- Order. The facts that matter most for a posting move up.
- Wording. A verified skill can use the term a posting uses, if it means the same thing.
- Selection. Some verified facts are left out to keep a resume to two pages.
What the kit is never allowed to change
- The facts themselves. No new skills, tools, titles, dates, or results.
- The qualifiers. Words like “about”, “average”, “in testing”, or “at one checkpoint” stay attached to a number everywhere it appears.
- The scope. A number that belongs to one job or project is never moved to another, and two true facts are never combined into a bigger claim.
Where do the verified facts come from?
They start with the person’s current resume and work history. The resume is the starting point, not the proof: during setup the kit turns each line into a question, and only what the person confirms becomes a fact it can use.
The confirmed facts live in a profile folder. A ranked order decides conflicts: the person’s latest statement wins, then the vetted experience file, then per-field skills files, then approved master resumes.
If a fact is not in those sources, it does not go on a document. The kit leaves it out or asks.
The metrics ledger
Every number lives in a ledger with its exact wording, its qualifiers, and its source. The kit never estimates, rounds up, or computes a new figure from other figures, and a new number means stopping to ask.
The do-not-claim list
Some things are listed so they never appear: tools the person has not used, wording they rejected, and results they cannot prove. They stay off the page even when a posting asks for them.
How does the kit catch drift before anything is delivered?
A final pass reads every line of the resume and cover letter next to the person’s own words. For each line it asks whether they said this, at this scope, and in this tense.
That pass looks for the small inflations AI tends to add while polishing: a quantifier nobody said, a past task written as ongoing, or a reason and a result the person never gave. In testing it caught slips that every automated check missed.
A script then checks each file for the layout problems that break applicant tracking systems: tables, text boxes, headers, unreadable text, and a page count over two.
How do the rules get better over time?
Every miss becomes a dated rule. A corrections log records the old wording, the truth, and every other file where the old claim needs fixing.
Several standing rules came from real misses:
- A missed email instruction. Any email or special instruction in a posting is now flagged in bold at the top of the reply.
- A fake posting. Every posting now gets a pre-check for legitimacy, pay, freshness, hard requirements, and scope before anything is built.
- Blank space on page one. The build measures the space at the bottom of page one and fills it with the next verified bullet instead of forcing a break.
- Letters that sounded generated. Cover letters now follow a plain voice rule with real stories and a list of banned stock lines.
What does a tailored application include?
The person pastes a posting. The kit checks it first, then builds a tailored resume and cover letter from the matching master resume, in markdown, Word, and PDF.
It also gives an honest match estimate and lists the gaps. Gaps are not hidden or bridged with invented experience; they are named so the person can prepare for them.
How did it become something other people can use?
The first version was full of my own facts. To share it, I split the method from the person.
The rules, templates, and tools became the shared kit. The facts moved into a profile folder that starts empty and is filled by a first-run questionnaire from the new person’s own resume and answers.
A packaging script builds a clean copy with no personal data. I wrote up the general approach in How to Write AI Skills Other People Can Reuse.
What does this case study not claim?
The kit is not public yet, so there is no link to it. I am not publishing application counts or response rates.
It does not make anyone more qualified than they are. It makes sure the qualifications they really have are shown accurately for each job.
Related
The method behind this kit is explained in Why Constraints in AI Development Matter. For a data pipeline built the same way, see the public data research agent.