Published April 23, 20266 min read

Seven Ways an AI-Edited CV Drifts Away From the Facts

Most CV errors are not elaborate fraud. They are quiet drift between documents, prompts, dates, and polished claims. A structured Career Record makes them reviewable.

The problem is usually drift, not a forged identity

A senior freelancer asks an agent to shorten a CV for a client. The agent merges two projects, rounds a result upward, and turns “supported the migration” into “led the migration.” The text sounds better. It is also a different claim.

Repeat that across a Google Doc, a website, a PDF, and several agent conversations and the professional no longer knows which version is authoritative.

Seven patterns worth checking

1. Date compression

Month-level dates become years, gaps disappear, and overlapping engagements look sequential. Keep the underlying start and end dates stable even when a short CV displays less detail.

2. Responsibility inflation

“Contributed to” becomes “owned”; “technical lead for one stream” becomes “led the program.” Preserve scope as a fact, not a tone choice.

3. Metric laundering

An estimate becomes an exact number after several rewrites. Store whether a metric is measured, approximate, client-provided, or intentionally omitted.

4. Technology time travel

A current stack is copied into an older project because the agent treats a skill list as universal. Attach technologies to the experience in which they were actually used.

5. Company and client collapse

For contractors, employer, consultancy, intermediary, and end client are not interchangeable. A structured project record can represent the relationship without pretending they were one organisation.

6. Translation drift

The English version gains stronger claims than the German source. Secondary-language fields should be translations of confirmed facts, not a second biography.

7. Output drift

The public profile gets updated, but the PDF in a shared folder does not. Generate both from the same record instead of maintaining copies.

A safer agent workflow

  1. Read before writing

    The agent reads the current confirmed record before proposing anything.
  2. Propose a narrow diff

    Dates, roles, results, and translations change as explicit fields rather than as a rewritten document.
  3. Show provenance

    The owner can see whether a fact came from an import, a manual edit, or an agent proposal.
  4. Require owner approval

    The agent that drafted the change cannot approve or publish it.

Verification helps, but it is not magic

A named client or peer can add first-hand context to a specific claim. A public artifact can support another. Neither proves the entire CV, and person.work should not pretend otherwise. The benefit is that evidence stays legible and attached to the work instead of becoming an opaque badge.

The practical standard

The goal is not to make every CV suspicious. It is to make professional memory stable enough that different agents and outputs do not quietly rewrite it. A confirmed Career Record gives the owner a place to inspect those changes and an open format to take elsewhere.

Frequently asked questions

Does using AI make a CV fraudulent?

No. AI can help structure and phrase real experience. The problem begins when generated wording, metrics, dates, or responsibilities are accepted without checking them against the underlying facts.

How does a Career Record reduce CV drift?

It keeps stable facts separate from presentation, records the source of changes, and requires owner approval before agent proposals become part of the confirmed record.

Can person.work guarantee that every claim is true?

No. The owner remains responsible for the record. Structured provenance and optional confirmations improve inspectability, but they do not replace judgment.

Maintain your existing CV with the agent you already use.

person.work stores your confirmed Career Record facts and renders web, PDF, and JSON from them. Non-commercial, with no feed and no sale of profile data.

Connect agent