AI verification policy

How ATSTune keeps AI-assisted resume changes reviewable

AI can improve structure and wording, but it can also produce unsupported claims. ATSTune therefore separates a useful draft from a verified final document and prevents unresolved additions from moving silently into an export.

Maintained by ATSTune · Updated

Identity-level facts stay fixed

  • Real name and contact information are not guessed from filenames or surrounding context.
  • Employer and school names, education, and employment dates are not replaced with invented alternatives.
  • New employers are not fabricated and dates are not changed to improve apparent experience.
  • Precise numerical achievements are not invented when the source resume does not support them.
  • Redacted, withheld, unreadable, or placeholder identity fields remain blank.

Three optimization modes, increasing in scope

ModeTypical changesVerification behavior
1 · Focused polishPolishes wording, adds major target skills, and strengthens the summary while keeping experience and projects close to the source.Unsupported additions are surfaced for verification or user evidence.
2 · Strong alignmentCovers most target skills and brings the summary closer to the target role while keeping experience and project changes primarily editorial.Unsupported additions remain review items and cannot bypass the export gate.
3 · Reference draftCreates a more extensive demonstration draft, including stronger target-role framing across experience and projects.Demo-level additions are explicitly treated as unverified until the user confirms, edits, or removes them.

Every unresolved addition has a review state

Review items include a stable identifier, destination, the actual text that needs attention, and a question asking the user to confirm the supporting experience. Users can confirm an accurate statement, edit it into an accurate statement, or remove it as unsupported or not applicable.

  • Pending: not yet reviewed and blocks export.
  • Confirmed: the user states the wording is supported by real experience.
  • Edited: the user revised the content to make it accurate.
  • Removed: the claim or skill was removed or marked not applicable.

Export requires two accuracy checks

All individual review items must be resolved before export. The user must then make a separate overall accuracy confirmation in the template dialog. Both the browser and server enforce these checks before PDF generation or any premium-template charge.

AI output can still be incomplete or wrong. The verification workflow reduces risk; it does not transfer responsibility away from the applicant.

See the method in a complete example

Review a fictional report with weighted dimensions, located keyword evidence, missing evidence, and an AI verification item.