Preparing a CV with AI: Tools and Prompts
Preparing a CV with AI: working tools, ready prompts and verifying the output — how to build a CV that proves your fit without looking artificial.

Preparing a CV with AI is possible, but AI must be used not as the author of your experience — rather as an editor that selects the real facts you provide to fit the vacancy and cleans up the language.
A fluent sentence is not a correct CV. The model can write a non-existent result convincingly, scramble your work history and present a keyword from the listing as a skill you own. Such text looks strong at first glance and puts its owner in a difficult spot at the interview.
This guide does not promise "the perfect CV in one click." Instead it gives the right input data, three tool types, seven copyable prompts, the personal-data boundary and the human check required after every output.
What should AI do in a CV — and what should it not?
| Work AI can be given | Decisions AI must not make |
|---|---|
| Grouping the vacancy's requirements and responsibilities | Deciding by itself which skills you have |
| Turning the event you provide into a short experience bullet | Inventing events, dates, tools and results |
| Simplifying long sentences and finding repetition | Inflating personal contribution with the team's collective result |
| Showing the proof gap between the CV and the listing | Filling the gap with new claims |
| Providing a checklist for format and sequence | Guaranteeing "it will pass ATS" without reading the CV |
A good CV does not compress your whole career onto one page; it selects the facts serving a specific vacancy's decision. Define the core structure first in the guide to writing a CV. AI can speed that structure up, but only you can confirm which facts are yours.
Which fact package must be prepared before the prompt?
Just dropping the old CV into AI and writing "make it more professional" is a weak start. The model does not know which vacancy you are applying for, which result was verified and which information must not change. Prepare this package first:
- the vacancy's full text and the application language;
- a confirmed work history with titles, companies and dates;
- per role, not the responsibilities but two or three real events and your concrete actions;
- the tools you use and your genuine level;
- the correct spellings of the education, certificate, project and language details;
- the source of any number — or, if unmeasured, the observed change;
- a list of confidential and personal data that must not enter the CV.
Remove the phone number, home address, identity documents, date of birth, references' contacts and confidential work files from the initial prompt. After the editing is done, add the necessary contact details to the final document yourself.
How do you choose between ChatGPT, Gemini and Copilot?
These tools do not verify a CV's truthfulness. The choice should be made not by the "smartest model" slogan but by the account you use, the accepted files, the privacy settings and the program you will edit the output in. Because features and retention terms change, reopen the official page on the day you send. For a broader feature and cost comparison, see the AI tools guide.
| Tool | Practical use | Point to verify |
|---|---|---|
| ChatGPT | Listing analysis, the fact map, bullet and language editing | Data Controls and the "Improve the model for everyone" setting |
| Gemini | Analysis over text and files, answers and document output | Gemini Apps Activity, retention and connected services |
| Microsoft Copilot | PDF/DOCX file analysis and work within the Word ecosystem | Different data rules for personal Copilot vs a work/school account |
| A local text editor | Formatting, spelling and final fixes without sending to AI | Automatic cloud sync and sharing permissions |
OpenAI's Data Controls explanation shows the option to switch off the use of conversations for model improvement on a personal account. Google's Gemini Apps Privacy Hub separates how conversations and files are processed when Keep Activity is on or off. And Microsoft's Copilot file upload document explains the supported formats, including PDF and DOCX, and the retention rules for personal Copilot.
This information is not permission to "upload the confidential CV freely." An employer's internal documents, client lists, sales data and third parties' contacts are not your personal CV facts. If a company account has its own admin policy, do not apply the general personal-account explanation in its place.
A safe workflow for preparing a CV with AI
- Freeze the listing: save its date and full text; do not mix a listing that changes mid-process with the old CV.
- Clean the personal data: keep only the necessary professional facts for analysis.
- Build the requirement–proof map: instruct the model to show gaps, not to fill them.
- Work one section at a time: first the profile, then one job; do not change the whole CV in one reply.
- Confirm line by line: match the dates, titles, tools, numbers, roles and results with the original notes.
- Format in an ordinary editor: fix the headings, order, spacing, links and file name separately.
- Run the PDF/DOCX test: select, copy, search the text, and check the mobile view and links.
If a fact is wrong at one stage, the next prompt can hide the error by making it more fluent. So every output stays in "draft" status. It moves into the master document only after human confirmation.
7 practical prompts for preparing a CV
Replace the bracketed parts in the prompts below with your own material. Do not include confidential data. For broader prompt structures you can use the prompt guide.
1. Extracting the listing's requirements
Analyse the vacancy text below.
1. Show the core responsibilities, required skills and practical terms in separate tables.
2. Do not add requirements not in the listing itself.
3. Merge repetitions with the same meaning.
Vacancy: [TEXT]
2. Building the requirement–proof map
Match the vacancy requirements against my confirmed facts.
For each requirement write: a matching fact, a partial match and the missing proof.
Do not suggest skills not in the input; mark them [NO PROOF].
Vacancy: [TEXT]
My facts: [CLEANED FACTS]
3. Writing the profile summary
Write a 45–65-word CV profile using only the facts below.
Show the job title, length of experience, field knowledge and 2 relevant proofs.
Do not use unproven adjectives like "professional", "dynamic", "results-driven".
Do not invent numbers or tools. Facts: [FACTS]
4. Turning experience into a CV bullet
Turn this real event into 1 CV bullet: action + context + verifiable change.
Separate my role from the team's result.
If there is no number, do not add a percentage. Do not exceed 28 words.
Event: [SITUATION, ROLE, ACTION, RESULT]
5. Finding weak and unproven sentences
Check every claim in the CV text.
Show in a table each adjective without visible proof, unexplained tool, vague result and possible contradiction.
Do not write new text; for each problem, give the fact question I should be asked.
CV: [TEXT]
6. Editing for vacancy fit
Adapt the CV to the vacancy, but use only the real facts in the CV.
You may propose deleting or shortening non-matching sections.
Do not copy a word from the vacancy into the CV as a skill.
Show the changes in a before/after table with reasons.
7. The final fact audit
Match the final CV below line by line against the fact list.
Flag contradictions in dates, titles, tools, certificates, language levels, numbers and results.
Keep lines whose accuracy is unproven as [HUMAN CHECK].
Do not fix anything automatically.
CV: [FINAL TEXT]
Fact list: [CONFIRMED SOURCE]
How do you turn a weak AI output into a correct CV bullet?
| Version | Text | Problem |
|---|---|---|
| Weak AI output | Increased customer satisfaction 40% with an innovative CRM solution. | The CRM, the 40% and the satisfaction metric are not in the input |
| Fact-true bullet | Gathered customer requests into a single sheet by arrival date, owner and status, making pending enquiries visible at shift handover. | The event and the observed change are preserved |
"40%" makes the sentence look strong, but it creates no fact. If there is no answer when the interview asks for the measurement's source, the AI has not improved the CV; it has created a risk. You can compare ready field structures on the CV examples page.
Which promises about AI, ATS and CV design should you not believe?
No general AI tool sees every employer's applicant tracking system. A score like "95% ATS compatibility" is not decision evidence without the test's name, the system used and the scoring rules. The file format the listing requires always wins.
The National Careers Service recommends a CV that is clear, brief, in a consistent style, readable with headings and bullets, with a second person reviewing the spelling. Europass emphasises adapting the document to the specific vacancy and focusing on the relevant facts instead of collecting the whole past. These are not "ATS scores"; they are the foundations of a document clear to humans.
- keep simple headings and a logical section order;
- do not hide important information only inside icons, images, tables and text boxes;
- use a term from the vacancy only if the matching skill and experience genuinely exist;
- test selecting, copying and searching in the DOCX and the text-layer PDF;
- paste the extracted text into a plain editor and check the order survives.
AI can give formatting advice, but a human eye and a real file test must confirm the final document. In interview preparation too, explaining every CV bullet with the same event is a condition.
What must be deleted before handing personal data to AI?
| Rarely needed for analysis | Only with necessity and permission | Can stay as a professional fact |
|---|---|---|
| Phone, personal email, exact home address | A reference's name and contact | Titles, tenure and responsibilities |
| ID card, FIN, bank details | Client and employer internal documents | Public projects and your own role |
| Date of birth and family details | Confidential sales, salary and contract detail | Tools, certificates and verifiable results |
Changing an account setting does not replace data minimisation. The safest file is the one where information the tool does not need was never present. Check the admin policy on work and school accounts, and the active data settings on a personal account, separately.
Questions about preparing a CV with AI
Can an employer tell a CV was prepared with AI?
There is no guarantee the employer will precisely identify which tool wrote it. The main risk is not a detector: it is generic language, repeated adjectives, sentences foreign to your speaking style and results you cannot explain. Match every line against your own facts and delete any claim you cannot back with a real event at the interview.
Is uploading the full CV file to AI safe?
There is no universal "yes." The account type, the tool's current privacy settings, the employer's policy and the file's contents vary. First delete the phone, address, document numbers, references' contacts and confidential company detail. Then check the official retention and model-improvement settings; do not upload what is not necessary.
Can AI add keywords to the CV automatically?
It can extract the vacancy's terms, but it must not automatically add them as skills you own. Every keyword must be grounded in real experience, a project, education or a genuine level. Copying the listing's word without proof is not tailoring; it is creating a new claim to be tested at the interview. That does not count as verified fit.
Which AI tool is best for a CV?
There is no single winner. Several tools can suffice for listing analysis and text editing; the choice depends on the accepted files, the working language, the account's privacy, working with Word or Google documents, and the final human edit. Compare the data boundary first, then the features, and convenience last.
How should an AI-written CV bullet be checked?
Match every noun and number in the bullet against the source fact: the company, title, dates, tools, your action and the result. Check the team's result has not been converted into a personal one. If the number has no measurement, delete it and write the observed change. Then explain the bullet aloud; if it collapses under a follow-up question, it is not ready.
Sources
I'm Anar Rustamli - a strategist, entrepreneur, and AI adoption leader working at the edge of growth, technology, and human thinking. Since 2016, my work has focused on helping businesses evolve in a rapidly changing digital landscape. I design growth systems, AI-powered workflows, and strategic frameworks that align performance with purpose. I believe real growth happens when strategy, data, and human insight work together - and my mission is to help businesses adopt AI in a way that strengthens both their results and their identity.

