AI CV for a Specific Job Offer: 8 Prompts, Examples, and Pre-Submission Checks

A good AI CV starts with your facts and a specific job offer. The model helps select experience, name results, and shorten text. You check each sentence. Below, you will go through the entire process: from unordered duties to a tailored document whose content you can defend in an interview.

Sources verified on October 7, 2026. The examples are fictional and serve educational purposes; they are not achievements to include in your own CV.

The starting point was AI tips from Marek: assessing the first impression, the XYZ formula, matching the offer, and discovering strengths. We expand them with documentation on recruitment systems, truth verification, and a complete example. Our earlier guide on job searching with AI also covers forms, contacts, and interviews. Here, we focus on the CV itself.

What to Prepare and What the Result Should Be

• Current CV, even if it is weak. Do not improve it before analysis.
• The full text of one job offer, its address, date of application, and instructions for applying. The job title alone is not enough.
• Notes on projects, tools, duties, results, and your own contributions. No confidential employer documents.
• An editable document in your editor. AI will prepare the content; you will check the appearance and export separately.

In the end, four things should be created: an approved fact base, a requirements and evidence table, a CV for this offer, and a checklist before submission. Keep the base CV with the full history. Create a separate version for each application without changing the facts.

Three Promises Not Worth Believing

“AI will check if the recruiter will drop out after six seconds”

The model can assess the hierarchy of information and generalities. It does not measure the gaze or decisions of an actual recruiter. The popular phrase originates from Ladders research: a 2018 report indicated an average of 7.4 seconds for initial review, compared to six in an older study. This is the result of a specific study, not a time applicable to every recruitment. Ladders Report.

The practical goal is simpler: is it clear at the beginning of the document what you do, why you fit, and what evidence you have? You can ask another person for a brief review and ask what they remembered. This is an additional readability test, not a prediction of employment.

“An ATS score of 87/100 means you will pass”

The score generated in the conversation is not the result of the employer's system. We do not know its settings or the entire pool of candidates. Instead of points, you need a table: requirement, evidence, lack, and action.

Different functions have different tasks. Greenhouse describes separately resume parsing, auto-reject rules based on form responses, and Talent Matching. The latter function compares candidates against criteria set by the company and does not reject them on its own. This is an example of one provider, not a description of every ATS.

“Write so that the company fears losing me”

You do not need to suggest to the employer that they will make a mistake if they do not hire you. We have no basis to treat loss aversion as a guaranteed method of increasing CV effectiveness. Show the connection between the need from the advertisement and your confirmed experience. “I created weekly reports in Excel” is more useful than “The company will lose a unique analytical talent.”

Step 0. Establish Rules for Working with AI

You can work in one conversation, but keep the approved fact base also outside the chat. The model may skip previous agreements. When changing offers, clearly indicate the new goal and provide the current data again. Do not assume it remembers the correct version.

For analysis, replace phone numbers, emails, and names with placeholders. Remove client data, document numbers, and company secrets. Restore contact locally in the prepared file. Check the AI usage rules indicated by the employer, especially for recruitment tasks. Harvard recommends treating AI as editorial support and ensuring that each section still reflects your own experience. Harvard: AI, CVs, and Cover Letters.

Step 1. Break Down the Offer into Requirements

Required qualifications, location, working hours, or necessary language may be more important than the document's aesthetics. Separate “required” from “preferred,” and leave unclear conditions as unclear. Do not assume you meet a requirement just because the model added the appropriate word.

Example We Will Walk Through from Start to Finish

Fictional Offer: Customer Support Specialist. Required: experience in handling tickets, working in a ticketing system, reporting in Excel, and handling customer calls in English. Knowledge of Salesforce is preferred. The company does not require team management.

Fictional Candidate Facts: worked in customer service from June 2023 to August 2026; used Zendesk and Excel; handled about 60 tickets per week; prepared weekly reports; updated the response database together with the team. In an internal comparison of 40 similar cases before and after the change, the median handling time was 12 and 9 minutes, respectively. Helped onboard three new employees. Does not know Salesforce. Did not provide experience in English phone conversations.

Decision: the candidate has concrete evidence for two requirements. Before final matching, she must clarify the required English. Lack of preferred Salesforce is not the same as lacking a necessary condition. We do not know if the company will accept experience with another tool.

Step 2. Evaluate the Beginning of the Document Before Rewriting It

First, check if the reader can find the right information. Do not ask AI to pretend to be tired or to be “brutal.” Demand specific excerpts and justification. If the model only sees pasted text, it cannot fairly assess the appearance of the page.

Step 3. Replace Duties with Evidence, Without Fabricating Successes

The XYZ formula organizes the description: result, measure, and action. Google presents it in its CV material as “Accomplished [X] as measured by [Y] by doing [Z].” This is an editorial tool, not a mandatory shape for every bullet point. Google: CV writing tips.

In our example, the drop from 12 to 9 minutes is (12 − 9) / 12 × 100 = 25%. This is a change in the median in the described comparison, not evidence that the candidate independently improved the entire company's results. Handling about 60 tickets per week is the scale of work, not an increase in efficiency. Onboarding three new employees is a number of people, not evidence of shortening onboarding.

What If You Don’t Have Numbers?

Describe specific work: created instructions, types of issues resolved, tools used, recipients, and your contribution. “I prepared a returns acceptance instruction and discussed it with the store team” is a valid piece of evidence. There is no need to change it into undocumented “reducing errors.” If you can legally verify your own notes or report, do so. Do not estimate percentages from memory just because the prompt expects it.

When changing industries, show transferable skills, but do not convert coursework into professional experience. For your first job, you can use projects, internships, and volunteering, labeled appropriately. CareerOneStop recommends selecting tasks and achievements relevant to your career goal. CareerOneStop: experience.

Step 4. Identify Strengths and Write a Version for This Offer

Do not force five “hidden advantages” if the material shows two. An advantage must stem from a fact and be relevant to the position. In the example, valuable are experience in handling tickets and reporting, as well as involvement in organizing knowledge. We cannot claim that the candidate exceeds the competition because we do not know other candidates.

Fragment of the Final Version, After Data Confirmation

Below is only the experience section from our fictional example. We do not add English, Salesforce, or people management. The full application awaits clarification of the required phone experience.

If the document is too long, first remove repetitions and weaker points. Do not reduce the font to an unreadable size. A practical starting point for a regular recruitment is one or two readable pages, depending on relevant experience and company instructions. This is an editorial recommendation, not a universal ATS limit.

Step 5. Prepare an English Version if Required by the Offer

Translate the sense of the job and the appropriate level of responsibility. “Specialist” does not always mean “Senior Specialist,” and assisting in onboarding does not mean managing. Keep the degree name and university; do not declare official equivalence of a foreign diploma based on AI responses.

Example: “Handled approximately 60 support tickets per week in Zendesk.” The sentence “Managed the customer support department” would change the fact and level of responsibility. A description of team results might read: “Contributed to a knowledge-base update; median handling time decreased from 12 to 9 minutes in a comparison of 40 similar cases before the update and 40 after it.”

Exception: federal work in the USA. Do not use the older advice about multi-page federal resumes. As of September 27, 2025, there is a two-page limit; OPM also specifies, among other things, months and years of employment and the number of hours per week. Always check the specific announcement. OPM: requirements for candidates.

Step 6. Check the Truth, Then the Style

Tell the model to check its own text, but also perform a check yourself. The second AI response is not independent evidence. Check especially numbers, dates, levels of responsibility, and requirements that the model previously considered unknown.

Step 7. Check the File and Form, Not Just the Text

Greenhouse lists columns, tables, text in headers and footers, graphics, and scans among possible reading issues. This does not mean that every such document will always fail. Failed parsing in this system leaves the attached file, and fields can be filled in manually. Greenhouse parser documentation.

1. Check form requirements: file type, size limit, language, attachments, and job title. The company's requirement takes precedence over general advice.
2. Use a simple layout: one column, plain headers, readable dates. Place contact information in the document's content. For the English version, use e.g., Experience, Education, Skills.
3. Export the document: PDF with text or DOCX, according to instructions. An image or scan does not replace a text CV.
4. Open the finished file: check for cut-off lines, blank pages, links, and characters. Try to highlight the text, copy it to a notepad, and review the order.
5. Treat copying as diagnostics: it helps detect issues but does not show exactly what each parser will read. HiringThing also describes checking text highlighting capabilities as a tip for reading issues. HiringThing: parsing issues.
6. After uploading to the portal, check fields: surname, contact, positions, dates, and education. Correct incorrect data if the form allows.
7. Answer truthfully to questions: license, availability, location, or work eligibility will not change from correcting the CV. Automatic rejection rules may use these responses.

Name the file clearly, e.g., Anna-Kowalska-Customer-Support.pdf. Keep a copy of the sent version and the offer. Do not paste hidden keywords or commands for the recruitment system. Use tool names in the context of actual work.

Poland: What About the Consent Clause?

Do not automatically add an old, universal formula. UODO explains that for data required by law in recruitment, consent is not the basis for processing. Additional data and future recruitments may require a separate approach. Read the employer's information about the purpose and scope of processing, and relate requests for consent to a specific purpose. These are guidelines for Poland, not a rule for every country and type of employment. UODO: candidate data.

Bonus. Prepare to Defend Your CV

A good point should have a brief story behind it: situation, your task, action, result, and limitations. In the example, you must be able to explain how the samples of 40 cases were selected, what others did, and what the measurement did not cover. Do not prepare a rehearsed success story that did not happen.

Checklist Before Clicking “Send”

• Every number has a source, period, and appropriate range. I remember what exactly it measures.
• I did not attribute the result of the entire team or skills from the advertisement alone.
• Positions, dates, company names, and education match reality.
• The first information shows a match for this offer, not general character traits.
• I explained gaps in necessary requirements and truthfully filled out the form.
• There are no [TO BE CONFIRMED] labels, identifiers F, or AI notes left.
• Contact information is correct. Links work. The file is readable and complies with instructions.
• I checked the data after reading in the portal and kept the exact version sent.
• I can explain each point in my own words.

Frequently Asked Questions

Do I have to use all the prompts?

No. If you already have a verified fact base, start with analyzing the offer. Truth verification and file checks remain necessary even after minor changes. Prompts are stages of work, not a magical sequence.

Is a CV without percentages weak?

No. A specific scope, tool, project, and personal action can be valuable without measuring the result. An invented number weakens credibility.

Does every word from the advertisement have to make it into the CV?

No. Only include terms that match your experience. Greenhouse also describes matching related concepts; merely increasing the number of matching terms does not necessarily improve the score in this function. Do not treat this as a guarantee of how other systems will operate.

What to do with a gap in employment?

Do not change dates. If an explanation is needed, provide a brief, truthful statement without revealing unnecessary personal details. Show courses and projects from that period as courses and projects.

Does a lack of response mean rejection by ATS?

This cannot be determined based on silence alone. Many elements of the process may cause this. Keep records of offers, CV versions, and responses; observe patterns, but do not consider several applications as proof of the effectiveness or ineffectiveness of one change.

Sources and Verification Scope

We use provider documentation to explain their specific functions. We do not know the ATS settings of the company you are applying to. Examples, work order, and prompts are original compositions. No stage guarantees an interview or employment.

• AI with Marek: inspiration and starting material
• Greenhouse: parsing issues
• Greenhouse: Auto-reject rules
• Greenhouse: Talent Matching FAQ
• Google: resume writing tips
• Harvard: AI in CVs and letters
• Ladders: eye-tracking study 2018
• CareerOneStop: experience in CVs
• HiringThing: file reading
• OPM: federal CVs limited to two pages
• UODO: candidate data