Use AI to improve a job application without inventing experience
Use a prompt with clear fact limits, a before-and-after example and a checklist to improve an application without creating false claims.
Panta field guide
5 min read
Published 9/8/2026
AI can compare a job advert with your evidence, flag vague writing and suggest clearer wording. It does not know what you did. Give it a small set of facts, tell it to mark gaps and check every proposed sentence before using it.
Choose a narrow task
“Improve my application” leaves too much room for guessing. Ask for one task you can check:
- Map the advert’s requirements to evidence you provide.
- Flag unsupported claims in a draft.
- Shorten a paragraph without changing its meaning.
- Ask questions that expose missing context.
- Offer two plain-language versions of one factual bullet.
Do not ask for a fit score. A percentage hides assumptions about how requirements are weighted and how missing evidence is treated.
Language models can produce plausible false statements. OpenAI calls these hallucinations and recommends uncertainty over guessing. Read OpenAI’s explanation.
Prepare a fact sheet
Do not begin with your full CV. Create a small evidence set for the target role.
TARGET ROLE
Role title:
Employer:
Requirements copied or summarised from the advert:
VERIFIED CANDIDATE FACTS
- Job, project or training:
- Action personally taken:
- Scope:
- Result:
- Tool or method actually used:
- Date or duration, if relevant:
KNOWN GAPS
- Requirement for which I have no evidence:
- Detail I do not remember or cannot verify:
WRITING TASK
What I want the tool to improve:
Tone and length:
Remove information the tool does not need. A writing task rarely needs your address, date of birth, references’ details, identification numbers or private information about colleagues. Check the service’s privacy and data settings. For ChatGPT, see OpenAI’s current Data Controls guidance.
Set clear fact limits in the prompt
Copy and adapt this prompt:
You are editing one part of a job application. Use only the candidate facts
between FACTS and END FACTS.
Rules:
1. Do not add tools, results, responsibilities, dates, qualifications,
language levels, preferences or motivations.
2. Do not turn a team result into the candidate's personal result.
3. If evidence for a requirement is missing, write "NO EVIDENCE PROVIDED".
4. Keep proposed wording separate from questions and warnings.
5. Prefer plain, specific language. Do not add praise, enthusiasm or
personality claims.
JOB REQUIREMENTS
[Paste the relevant requirements, not the entire page]
FACTS
[Paste your fact sheet]
END FACTS
TASK
First create a table with: requirement, supplied evidence, gap or question.
Then rewrite [one CV bullet / one letter paragraph] in no more than [limit].
After the rewrite, list every factual claim in it and show the supplied fact
that supports that claim.
Check the table before asking for prose. It is easier to spot a false connection there.
Fictional before-and-after test
This fictional example was tested on 8 September 2026 with a GPT-5-based Codex agent. The research note records the input, output and manual checks. Another model—or another run—may respond differently.
Relevant fictional advert requirements
- Coordinate monthly performance reporting.
- Build dashboards in Power BI.
- Present findings to senior managers.
Facts supplied by the fictional candidate
- Maintained a weekly Excel report for six retail branches.
- Checked submissions and followed up on missing figures.
- Proposed one standard returns definition, adopted by the branch managers.
- Presented the report in the monthly branch-manager call twice when the team lead was away.
- Has viewed Power BI dashboards but has never built one.
Before
Data-driven professional responsible for delivering powerful reporting solutions and influencing senior stakeholders across the business.
“Data-driven” and “powerful” say little. “Solutions,” “senior stakeholders” and the business-wide scope are not supported by the facts.
Proposed after
Maintained the weekly Excel performance report for six retail branches, followed up on missing figures and helped branch managers adopt one definition for product returns.
Claim check
| Proposed claim | Support |
|---|---|
| Weekly Excel performance report | Supplied fact 1 |
| Six retail branches | Supplied fact 1 |
| Followed up on missing figures | Supplied fact 2 |
| Helped managers adopt one returns definition | Supplied fact 3 |
| Power BI experience | Not claimed; no build experience was supplied |
| Senior-management presentation | Not claimed; only branch-manager presentations were supplied |
The rewrite is narrower because the evidence is narrower. Power BI belongs in the gap column, not in the candidate’s experience.
Treat output as proposed edits
Review one sentence at a time:
- Underline each factual claim.
- Find the source in your notes, CV or memory.
- Check whether the sentence enlarges your scope or ownership.
- Restore any useful qualification the model removed.
- Say the sentence aloud and rewrite it in your normal voice.
If you cannot support a claim, delete it or find the missing fact. Do not ask the model to hide the problem.
UZH’s current motivation-letter guidance advises applicants who use AI to check the text critically, adapt it to their own language and add their own examples. See the UZH guidance.
Common failure patterns
- Borrowed ownership: “supported” becomes “led.”
- Invented scale: “several clients” becomes “a portfolio of 20 clients.”
- Tool inflation: reading a dashboard becomes building one.
- False motivation: the model claims a lifelong interest you never stated.
- Employer flattery: copied brand language replaces a reason to apply.
- Smoothed gaps: a missing requirement disappears from the comparison.
- Generic voice: every applicant becomes “results-driven” and “passionate.”
- Privacy excess: the prompt contains personal data unrelated to the edit.
When not to use AI
Do not use AI when the application forbids it, the exercise tests your unaided writing, the service’s terms are unsuitable or the facts are too sensitive to share. Do not upload confidential employer documents or other people’s personal information.
The tool can ask why you want the role. Only you can answer.
Verify every AI-assisted edit
- The task was narrow and the supplied facts were clearly delimited
- Unnecessary personal and confidential information was removed
- Missing evidence remained visible
- No team result became my personal result
- Every tool, number, date, responsibility and language level is supported
- The wording sounds like me and I can explain it in an interview
- The final application was checked against the original advert
- I—not the tool—made the final decision
For a broader search task, first build a search brief. For application prose, use the evidence table with the motivation-letter guide.
Put the guide to work
Use what you learned to research employers and turn broad ideas into a focused shortlist.