Set a factual boundary before prompting
AI can help organize a CV, identify unclear wording, or compare your evidence with a job description. It should not create experience, qualifications, or achievements that did not occur. Start with a verified fact sheet containing dates, responsibilities, tools actually used, and results you can support.
Tell the assistant to preserve those facts and flag missing information rather than fill gaps. If a suggested bullet implies commercial deployment, team leadership, or measured savings you cannot substantiate, reject it. A polished sentence is still false when it changes the underlying scope of your work.
Protect private and confidential information
Before uploading material, remove identity numbers, private addresses, account details, and information about clients or colleagues that you are not authorized to share. Replacing a name may not be sufficient if the remaining details identify a person or confidential project. Check the tool's current data and account controls rather than assuming every service handles inputs identically.
Use a sanitized summary whenever possible. For example, describe a synthetic reporting project instead of pasting a real customer export. NIST's AI risk framework offers a structured way to think about risks, but reading it does not certify a particular tool or make an unauthorized disclosure acceptable.
Use AI for critique rather than invention
Ask focused questions: which bullet lacks a clear action, where is the project context ambiguous, or which stated requirement has no evidence? These prompts can improve clarity without inviting fabricated achievements. Keep the original text and compare revisions sentence by sentence.
A useful exercise is to ask for two shorter versions of a truthful bullet, then check whether either removes an important limitation. A synthetic prototype should remain a synthetic prototype. If the assistant turns it into a deployed platform, correct the wording and investigate whether the same exaggeration appears elsewhere in the document.
Verify every technical and career claim
Check tool names, dates, certification titles, links, and metrics after editing. AI may produce confident but inaccurate descriptions of your work or a technology. Do not list a method simply because it sounds aligned with the advertisement. You should be able to explain the skill and show appropriate evidence.
Read the final CV aloud and ask whether you could defend every sentence in a detailed interview. Keep a small evidence register linking important claims to projects or records. This is especially useful when multiple AI-assisted revisions make the language more impressive while gradually moving away from the original facts.
Respect employer assessment rules
Employers may set specific rules for take-home tasks, writing samples, coding exercises, or live interviews. Follow those instructions. Assistance that is permitted for editing a CV may not be permitted during a timed assessment. If the rule is unclear, ask rather than assuming silence means unrestricted use.
Do not use AI to impersonate you in an interview, fabricate references, or conceal another person's work. When an employer requests disclosure of tool use, answer accurately. If disclosure is not requested, you still remain responsible for the submitted content and must not falsely imply that prohibited assistance was absent.
Keep your voice and judgement visible
Revise generic phrases into concrete descriptions you would naturally use. An interviewer needs to understand your reasoning, not merely read a fluent document. Practise explaining each project without the assistant and identify any sentence whose meaning you cannot comfortably unpack.
UNESCO's recommendation on AI ethics and NIST's risk framework emphasize broader responsible-use principles that support this careful approach. Apply them practically: minimize unnecessary disclosure, retain human review, and avoid misleading claims. The best AI-assisted CV is not the most elaborate one; it is an accurate, readable account of evidence that you can discuss with confidence.
Sources & further reading
Last editorial review: 27 October 2026. Examples are educational, not live employer data or guaranteed outcomes.