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
| Mode | Typical changes | Verification behavior |
|---|---|---|
| 1 · Focused polish | Polishes 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 alignment | Covers 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 draft | Creates 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.