You send out a hundred applications and receive three automatic rejections and ninety-seven instances of silence. A colleague in the same industry sends ten applications and gets three interviews. The difference rarely lies in qualifications. It lies in whether the CV passed through the Applicant Tracking System (ATS) in a readable form, whether it contains the words the recruiter is searching for in the database, whether someone in the company knows you are applying, and whether what you sent differs from a thousand other documents written by the same ChatGPT. This guide shows how ATS really works and provides nine prompts in one conversation, each in a box to copy, that use AI for research, matching, and preparation, rather than writing the same thing for everyone.
Last verified: September 6, 2026. Informational and educational material. Recruitment systems and AI tools change every few months; rules regarding personal data and discrimination in recruitment vary between countries and states. This guide is not legal or immigration advice.
In Brief
- ATS is a database with a search engine, not a judge. The system breaks down CVs into fields, and the recruiter searches the database by keywords and filters. It mainly rejects poorly parsed documents, lack of required words, and responses to screening questions. A Harvard Business School study: about 99 percent of Fortune 500 companies use such systems, and half of companies in the USA have a filter that rejects candidates with an employment gap longer than six months (Harvard Gazette on the “Hidden Workers” report).
- “Bypassing” ATS is not about tricks, but three things: a format that the parser reads correctly; exact phrases from the job posting supported by evidence from your history; a person in the company who will pass your application out of turn. White text and keyword stuffing work against you.
- Everyone uses the same AI, so the advantage lies in the data that AI does not have: your numbers, projects, names, artifacts. Prompt 0 builds a fact file from them; each subsequent prompt uses it and is prohibited from making things up.
- Recruiters see hundreds of CVs with the same verbs. “Spearheaded,” “leveraged,” “passionate” signal that the text was written by a model. Prompt 3 has a list of banned words and requires specifics instead of adjectives.
- You have the right to know that an algorithm is evaluating you. In New York, an employer using automated tools for recruitment decisions must have a bias audit and notify the candidate (NYC, Local Law 144); the federal EEOC is conducting an initiative on AI in employment (EEOC).
- Your dream job rarely comes from a portal. Prompt 6 builds a map of people and messages, and prompt 8 does a weekly review of what works so you don’t send your hundredth application using the same method that didn’t work ninety-nine times.
How ATS Really Works
The Applicant Tracking System (ATS) is a program where a company collects applications, breaks them down into fields (contact information, positions, dates, education, skills), asks screening questions, and allows the recruiter to search the database. In some systems, candidates are additionally sorted by fit for the posting. The recruiter does not read a hundred CVs in order; they type “Salesforce administrator Chicago” into the search engine and look at the first results. If your CV did not parse correctly or does not contain those words, it does not exist in that search.
The Harvard Business School and Accenture report “Hidden Workers: Untapped Talent” (report page) described the consequences: filters set for employment gaps, lack of a degree, or lack of exact words from the posting reject people who can do the job. The authors estimate this group in the USA at 27 million people. For immigrants, this is particularly important: a gap for relocation, a degree from another country, and different job titles are exactly those filters.
The circulating number that “robots reject 75 percent of CVs” has no verifiable source; what has been documented is more mundane and more fixable: documents must be readable, words must match, screening questions must be passed, and the surest path to an interview still leads through a person.
The system extracts text and guesses what is a position, date, company. Columns, tables, icons, graphic headers, and unusual section names ruin this step. Test: copy the CV into a plain text editor; what you see is what the system sees.
Work eligibility, willingness to relocate, years of experience, education, salary expectations. Answering “no” to a screening question ends the process before anyone opens the CV. You answer truthfully but understanding what they are asking.
The recruiter searches by words and filters, then reads for a few seconds. A referral from an employee goes into a separate queue. That’s why two of the nine prompts are about people, not documents.
Why CVs from ChatGPT Look the Same and How to Be Better
A language model asked to “improve my CV” does the same for everyone: it adds the same verbs, smooths sentences, invents numbers where there were none, and produces a cover letter that the recruiter has already seen forty times today. The advantage does not lie in a better writing prompt. It lies in four things that the model does not have until you give them:
- Data. Numbers, system names, team sizes, amounts, deadlines, artifacts (repository, portfolio, publication, certification). Prompt 0 extracts them from you through an interview before anything is written.
- Matching. One CV for one posting, with exact phrases from that posting, but only where you have evidence for them. Prompt 2 reads the posting, prompt 3 matches.
- People. A person in the company who will pass the application and an informational interview before applying. Prompt 6.
- Loop. Each week you look at which channels are yielding responses and change your method. Prompt 8.
The Right Role of AI – researcher, job posting analyst, interview coach, and editor working on your facts. Not the author of your story.
Before You Paste the First Prompt
One Place for All Job Searching
Job searching takes weeks, and the prompts in this guide use the same fact file. Set up one place: in Claude, it’s a Project where you attach the fact file and enter permanent instructions (Anthropic documentation); in ChatGPT, it’s a Project with files and your own instructions (OpenAI help on projects, on custom instructions). In the permanent instructions, write one sentence that saves you from the biggest mistake: “Never add achievements, numbers, names, or skills that are not in the fact file; if something is missing, ask.” You enable web searching in Claude from the “+” menu (instruction), in ChatGPT it’s in search mode (OpenAI help); Gemini has a Deep Research mode (Google description) useful for company research.
One Calibration Question
Before we start: provide today’s date. I am looking for a job as [POSITION] in [CITY / state / remotely]. List five job titles under which employers advertise this role, with links to three current job postings, and a link to the job description in the O*NET database. If you can’t find something, write “I didn’t find it.”
If the links are dead, the postings are old, or the job titles are synonyms from memory, the model is not searching. Reply: “Search the web and provide current postings with dates.” The O*NET database contains official job descriptions, skills, and alternative job titles; it is the best source of the words recruiters are looking for.
Where the Model Should Search
| What You Need | USA | Poland |
|---|---|---|
| Job titles, skills, keywords | O*NET; MySkillsMyFuture (Department of Labor) for translating skills between jobs | Postings from major portals; job classification on praca.gov.pl |
| Salaries and job outlook | Occupational Outlook Handbook (Bureau of Labor Statistics) | GUS, salary reports from portals (secondary data) |
| CV for federal jobs | USAJOBS: what to include in a CV (different rules than in the private sector) | Civil service recruitment: announcements from offices |
| Recognition of Polish diploma | Evaluation by an organization from the NACES list, e.g., WES | Nostification of a foreign diploma at a university |
| Work eligibility and documents | USCIS: documents for Form I-9 | guide on legalizing work in Poland |
| AI in recruitment and your rights | NYC Local Law 144, EEOC, EEOC for candidates | UODO (personal data in recruitment) |
| Signal “I’m looking for a job” | LinkedIn: Open to Work | The same |
| Recruitment scams | FTC on fake job offers | Our guide on fake job offers |
Brackets in Prompts
In the boxes are brackets: [POSITION], [CITY], [COMPANY], [POSTING]. You replace them once. Each box has a header with information on where to paste it and a “Copy prompt” button. Do not enter your social security number, date of birth, document numbers, or address in the fact file; your name, surname, city, and work history are sufficient.
Scheme: Nine Steps in One Project
- Fact File
The model interviews you about your entire work history and builds an inventory of achievements with numbers, names, and evidence. Without this file, everything else is made up.
- Goal and Market Map
What does the job you are looking for look like, under what names it is advertised, where, for how much, and what it requires. A list of companies and criteria.
- Job Posting Analysis
Hard and soft requirements, exact words, screening questions, hidden signals, assessment of fit without flattery, whether it is even a real offer.
- CV for the Posting
Matching from the fact file in a format that the parser reads. A list of changes so you know what has changed. Parsing test.
- Form and Screening Questions
Work eligibility, experience, salary, relocation: how to answer truthfully and understanding the question.
- Message, Letter, and Profile
A short text that no one else will write, a LinkedIn profile consistent with the words from the postings.
- People
A map of people in the company, a request for a referral, an informational interview. A way out of the queue.
- Interview
Questions arising from the posting, answers from the fact file, a mock interview with critique, company research with sources.
- Weekly Loop
A table of applications, what works, what to change. You read the offer according to a separate guide.
Why in this order. Facts before writing, because the model makes things up without facts. Goal before postings, because without a goal you apply for everything. Job posting analysis before CV, because the CV matches specifics. People after documents, because a request for a referral requires a ready CV. The loop at the end, because it measures the whole.
Prompt 0 – Fact File
The most important step and the only one where you speak, and the model asks. Goal: a document with every achievement, project, tool, number, and name that can be extracted from you. Recruiters look for specifics; models without specifics produce adjectives. Do this once, come back after a week, and add what you remembered.
You are my career editor. Your task: to build a fact file about my work history, which we will use for every CV, message, and interview. You are not writing the CV yet.
RULES
– You ask, I answer. You do not add anything I haven’t said. If the answer is general (“I improved processes”), you ask for numbers, scope, tools, time, and outcome until it is specific or I say “I don’t know.”
– Conduct the interview one position at a time, starting from the most recent. After each position, show me the collected facts and ask if anything is missing.
– Record job titles, tools, and industry terms in both Polish and English so I have both variants.
Start with questions about: current or last position (title, company, industry, size, dates, team, who I reported to); what I was responsible for in numbers (budget, clients, revenue, users, transactions, tickets, projects); three things I changed and what improved because of it; tools and systems with proficiency level; certifications and training with dates; artifacts I can show (portfolio, repository, publications, presentations); languages with proficiency; situations I would talk about in an interview (problem, action, result).
After going through all positions, return the FACT FILE in the format: sections by position, with points “fact: number: evidence,” a separate list of skills with proficiency level, a separate list of stories for interviews. At the end, a list of gaps: things you asked about that I couldn’t provide, so I can check in old documents.
Check in the response: whether there is anything in the file that you haven’t said. Models “fill in” facts in good faith; read the entire file and cross out any number you are not sure about. Save the file outside the chat and attach it to the project; it will be the source for all subsequent prompts.
Prompt 1 – Goal and Market Map
“Dream job” is a set of criteria, not a title. This prompt turns it into a list: what positions, under what names, in what companies, for how much, with what boundary conditions, and what you are missing to be a candidate, not just a willing applicant. You check salaries and outlook in the Occupational Outlook Handbook, job titles and skills in O*NET.
Same conversation, the fact file is in the project. Help me define my goal and market map.
MY GOAL DESCRIPTION
– What I want to do daily: [description]. What I don’t want: [description].
– Location: [CITY / remotely / relocation yes or no]. Mode: [office / hybrid / remote].
– Salary: minimum [AMOUNT], target [AMOUNT]. Benefits I care about: [e.g., visa sponsorship, family insurance].
– Boundary conditions: [e.g., no shift work, no travel over 20 percent].
– Horizon: I want to have an offer within [e.g., 3 months].
RETURN
1. A list of job titles under which such work is advertised, with code and description from O*NET and five alternative titles used by employers. For each title, a link to a current posting with the date.
2. Salaries: median and range for this occupation in my region from a government source with a date, alongside ranges from current postings. Does my target fit within the market.
3. Requirements that repeat in postings: the ten most common skills and tools with the number of occurrences in [e.g., 20] postings you will review. Separately: formal requirements (degree, license, certification, work eligibility).
4. My fit: a table of requirements from point 3 alongside evidence from the fact file or “none.” Honestly. For each lack: can it be closed in a month (course, certification, project), in a quarter, or not at all.
5. A list of twenty companies hiring for these positions in my region, divided: large with developed ATS, medium, small, where the application goes to a person. For each: a link to the careers page.
6. Three strategies with an assessment that fits my horizon: mass applying through portals, targeted applying with a referral, entering through temporary work or contract. No recommendation for one; state how much time each requires weekly.
Check in the response: whether the O*NET code and salary link exist and lead to the correct occupation; the model is eager to provide codes from memory. The matching table in point 4 must contain “none” where you don’t have it; if everything is green, the model is flattering. If your degree is from Poland, check in postings whether employers require evaluation by an organization from the NACES list.
Prompt 2 – Job Posting Analysis
A job posting has three layers: what they really require (usually three to five things), what they pasted from a template, and what they didn’t write (level, visa, on-site, why the position is open). This prompt breaks them down, extracts exact phrases that the recruiter will use in the search engine, and assesses fit without flattery. It also checks whether the offer is real; fraud patterns are described by the FTC.
Same conversation. Below, paste the job posting for [COMPANY], [POSITION]. Read it as a recruiter who will search the candidate database.
RETURN
1. Hard requirements (without them I’m out) and soft (nice to have), each with a quote from the posting. Separately formal requirements: work eligibility, degree, license, location, hours.
2. Keywords in the exact wording from the posting: names of tools, methods, certifications, titles. For each: do I have evidence for it in the fact file (yes, with which fact; partially; no). Do not change the wording, even if I have a synonym in the fact file; note the difference.
3. Screening questions I expect in the application form based on the content of the posting (years of experience, work eligibility, visa sponsorship, relocation, salary, availability). For each: what answer ends the process.
4. Hidden signals: level of the position based on the scope of duties and ranges, whether the position is new or replacing someone, whether the team is being built, whether the posting is a copy of a template (the same sentences in other postings from this company), how long it has been posted, whether ranges are provided, and if the state requires it and they are not provided, note that.
5. Red flags for fraud: contact via messenger, request for bank details or payment, email address from a domain other than the company’s, salary disproportionate to requirements, company not found in registers. Check if the posting is also on the company’s careers page.
6. Fit assessment from 1 to 10 with justification: three strongest matches with evidence and three gaps. Write directly whether to apply, apply with a referral, or skip, and why.
7. Five questions I would ask the recruiter before applying if I had access to them.
POSTING:
[POSTING]
Check in the response: keywords in point 2 must be in the wording from the posting; the parser and search engine do not know synonyms as well as the model. An assessment in point 6 below 6 with hard gaps is a signal not to waste time; a model that gives every posting an 8 needs to be calibrated with one sentence: “Assess strictly, like a recruiter with a hundred candidates.”
Prompt 3 – CV for the Posting
One CV for one posting, from the fact file, with exact phrases from the analysis, but only where you have evidence. Simple format: one column, standard section headers, plain font, no tables, icons, graphics, or text in the header; PDF with text, not a scan. In the USA, no photo, date of birth, marital status, or nationality; these are data that the employer should not consider (EEOC). For federal jobs, the rules are different and longer (USAJOBS).
Same conversation. Based on the fact file and analysis from prompt 2, prepare a CV for this posting.
RULES
– Every sentence in the CV must have a source in the fact file. Do not add achievements, numbers, tools, or titles that are not there. If something is missing for a requirement from the posting, leave a gap and list it separately.
– Use keywords from prompt 2 in the exact wording, in the context of the fact that confirms them. Do not insert a list of words without context and nothing hidden.
– Banned: spearheaded, leveraged, passionate, results-driven, synergy, dynamic, go-getter, team player, detail-oriented, proven track record, and any sentence that someone else could write for this position. Instead of adjectives: numbers, names, outcomes.
– Format: one column; sections in this order and with these names: Summary, Experience, Skills, Education, Certifications; for each position, title, company, city, month and year from and to; bullet points starting with a verb in the past tense; no tables, icons, graphics, page header, columns, or text fields. Length: [1 page / 2 pages].
– The job title in the CV header should match the title from the posting if my actual title means the same; if not, leave mine and add the equivalent in parentheses.
RETURN
1. CV in a ready-to-paste format.
2. LIST OF CHANGES: what and why changed compared to my previous CV, point by point, with reference to the requirement from the posting.
3. GAPS: requirements from the posting that have no evidence in the fact file, and how I can close them or address them honestly in a message to the recruiter.
4. PARSING TEST: a version of the CV in plain text, exactly as the system will see it; check if dates, titles, and companies are recognizable in each line.
5. Three questions for me if any fact needs clarification before I send it.
Check in the response:
- Read the CV sentence by sentence with the fact file next to it. Any number not in the file gets cut. This is the moment when people send CVs with achievements they didn’t have and get caught in the interview.
- Do the parsing test yourself: copy the finished PDF into a plain text editor. If the order of sections falls apart or dates are detached from positions, the format needs fixing.
- Search the CV for words from the banned list and their Polish equivalents (“responsible for,” “dynamic,” “results-oriented”). The model breaks its own ban more often than it seems.
- Save the CV version with the company name and date; during the interview, you need to know exactly what you sent.
Prompt 4 – Form and Screening Questions
The form in ATS asks questions where a “no” answer closes the process, and questions where “too much” or “too little” harms. You answer truthfully; the goal is to understand what they are asking and not to drop out due to misunderstanding. The most common misunderstanding among the Polish diaspora: the question about work eligibility and visa sponsorship are two different questions; documents confirming eligibility are described by USCIS.
Same conversation. Prepare me for the application form in ATS for this posting. My status: [citizen / permanent resident / visa with work rights, which and until when / I need sponsorship]. My salary expectations: [AMOUNT or range].
RETURN
1. A list of screening questions you expect in this form, with an explanation of what they are really asking, and with a truthful answer for my situation. Separately: the question about work eligibility and the question about visa sponsorship, now and in the future; explain the difference and what to answer in my case.
2. Years of experience: how to count for each requirement from the posting based on the fact file, broken down by positions, so I can provide a number I can defend.
3. Salary: what to enter if the field is mandatory, and how it relates to the ranges from prompt 1; if the state requires ranges in the posting, and they are not provided, what does that mean.
4. Fields not to guess: if the form asks for something not in the fact file (license number, graduation date), list it so I can check in documents.
5. Questions about sensitive data (origin, gender, disability, veteran status): explain why the form collects them, that they are voluntary, and that the answer should not affect the decision; source.
6. Attachments: what to send besides the CV, in what format, how to name the files, and whether a cover letter is required or optional.
7. If the employer uses AI tools to assess candidates: does it have to notify me in this state or city, and what can I do; source.
Check in the response: the answers in point 1 must be true for your situation; the model does not know your status better than you, and an untrue answer about work eligibility has consequences far beyond this application. Point 7: in New York, there is an audit and notification requirement (Local Law 144); in other places, the model should say “I didn’t find a regulation,” not guess.
Prompt 5 – Message, Letter, and Profile
A cover letter from AI is today recognizable by the first sentence. If it is optional, a better approach is a short text to the recruiter: three sentences that no one else could write because they refer to a specific fact from your history and a specific problem of that company. The LinkedIn profile should contain the same words that recruiters are looking for, as they search there just like in ATS; the “I’m looking for a job” signal is set in the profile (LinkedIn: Open to Work).
Same conversation. Prepare texts for this posting and the profile.
RULES
– Uniqueness test: each sentence must contain something from the fact file or from research on this company that no one else has in their fact file. General sentences (“I am motivated,” “Your company is a leader”) should be discarded.
– Ban on words from prompt 3. No exclamation marks. No repeating the CV.
– Company research only from sources with links and dates: company website, press releases, postings, public results. Do not invent facts about the company.
RETURN
1. A message to the recruiter or manager (up to 90 words): one specific fact about the company with a source, one fact from my history that addresses the main requirement from the posting, one sentence about what I want. Version in English and Polish.
2. A cover letter (up to 200 words), only if the posting requires it: the same structure, plus one sentence about a gap from prompt 3 and how I am closing it.
3. LinkedIn headline (up to 220 characters) and “About” section (up to 1200 characters) with job titles and keywords from prompt 1, in the context of facts; no list of words.
4. The first three lines of experience in the profile, rewritten to include words from the postings I am targeting, while maintaining truthfulness.
5. A list of ten skills for the Skills section in the profile, in the wording from the postings, only those for which I have evidence.
6. A follow-up message (up to 60 words) to send after 5 to 7 days of silence, and a version after the interview with thanks, referencing one specific topic from the conversation (leave space for input).
Check in the response: click on the source about the company from point 1; models invent “recent implementations” and “awards.” Read the message aloud: if it sounds like a message you would delete, shorten it by half. Do not paste anything into LinkedIn that is not in the CV; recruiters compare both.
Prompt 6 – People
A referral from an employee goes into a separate queue in most companies and is read by a person. An informational interview before applying provides words that are not in the posting and a name to reference. This prompt builds a map of people and writes messages that are not requests for jobs, but for twenty minutes of conversation. The model does not have access to LinkedIn; you create the people map, and the model arranges who to contact, in what order, and what to write.
Same conversation. I want to reach out to people at [COMPANY] and in companies from the list in prompt 1 out of the ATS queue. Below, I paste who I found: [list: name, position, how we are connected: former colleague, same university, Polish community, mutual acquaintance, no connection].
RETURN
1. Order of contact: from people with a real connection to strangers; for each: why this person, what to ask them (referral, informational conversation, information about the team), what not to ask.
2. Three messages (up to 70 words each), in English and Polish: to a former colleague, to someone from the same university or community, to a stranger in the position I am applying for. Each with one specific fact about me from the fact file and one specific request with a deadline. No attaching the CV in the first message.
3. Informational interview: ten questions that give me words and information useful in the CV and interview (what a day really looks like, what the team is missing, what the recruitment process looks like, who decides), without asking about the job directly.
4. Request for a referral: a message (up to 80 words) to someone I have already spoken with, with a link to the posting, one sentence about why I fit, and a ready paragraph they can paste into the referral system so they don’t have to write anything.
Same conversation. I want to reach out to people at [COMPANY] and in companies from the list in prompt 1 out of the ATS queue. Below, I paste who I found: [list: name, position, how we are connected: former colleague, same university, Polish community, mutual acquaintance, no connection].
RETURN
1. Order of contact: from people with a real connection to strangers; for each: why this person, what to ask them (referral, informational conversation, information about the team), what not to ask.
2. Three messages (up to 70 words each), in English and Polish: to a former colleague, to someone from the same university or community, to a stranger in the position I am applying for. Each with one specific fact about me from the fact file and one specific request with a deadline. No attaching the CV in the first message.
3. Informational interview: ten questions that give me words and information useful in the CV and interview (what a day really looks like, what the team is missing, what the recruitment process looks like, who decides), without asking about the job directly.
4. Request for a referral: a message (up to 80 words) to someone I have already spoken with, with a link to the posting, one sentence about why I fit, and a ready paragraph they can paste into the referral system so they don’t have to write anything.
Check in the response: messages must be shorter than the model wants; if they exceed the limit, ask to shorten. Do not send the same message to ten people in one company; people talk to each other. Meetings from point 6 should be checked in the source; the model invents group names.
Prompt 7 – Interview
Interview questions arise from the posting, and good answers come from the fact file. The model can generate a list of questions based on specific requirements, practice answers with you in the “situation, action, result” format, play a tough recruiter, and point out where the answer is general. Company research has sources with dates because the question “what do you know about our company” with an answer based on an invented fact ends the interview.
Same conversation. I have an interview at [COMPANY] for [POSITION], stage: [phone with recruiter / with manager / technical / panel / final]. Prepare me.
RETURN
1. Company research with sources and dates: what it does, what it announced in the last 6 months, who is on the team I am applying to (from public profiles), what problems are visible in postings and communications. Five facts, each with a link. No guesses.
2. Fifteen questions you expect at this stage, derived from the requirements of the posting and gaps from prompt 3. For each: what the recruiter wants to check.
3. For the five hardest: an answer in the situation, action, result format, built solely from the history in the fact file, up to 90 seconds of speaking. Note where I lack specifics so I can fill in.
4. Questions about gaps and difficult topics: employment gap, career change, degree from another country, accent, visa. For each, a short, truthful answer that closes the topic and returns to value for the company.
5. Mock interview: ask me questions one at a time, wait for the answer, after each give a rating from 1 to 5 and one correction. Be strict; do not praise generalities.
6. Eight questions I will ask: about the team, about the first 90 days, about why the position is open, about the decision process, about next steps and timelines.
7. Logistics: what to prepare (copies of the CV I sent, portfolio, questions), how to note the names of the interviewers, when and what to send after the interview.
Check in the response: click on the five facts about the company from point 1 before the interview. Speak the answers from point 3 aloud with a timer; 90 seconds is less than it seems. In the mock interview, do not let the model ask all questions at once; it should wait for your answer.
Prompt 8 – Weekly Loop
Job searching is a process with numbers: how many applications, how many responses, from which channel, after what time. Without a table, you repeat a method that doesn’t work. This prompt reads your table once a week and tells you what to change. When an offer comes, you read the contract according to the guide on analyzing employment contracts with AI.
New conversation in the project. I paste my application table from this week and previous ones (columns: date, company, position, channel: portal, company website, referral, recruiter, direct message; version of CV; response: none, automatic rejection, rejection after interview, interview, offer; response date; notes).
RETURN
1. Numbers: applications, responses, and interviews broken down by channel, and percentage of responses by channel. Median days to response.
2. What works, what doesn’t: which channel and which type of company yields interviews, which yields silence. If a portal gives zero responses from twenty applications, state that directly.
3. Three changes for next week, each with a number (e.g., “five applications with a referral instead of twenty from the portal,” “two informational interviews”), and one thing I will stop doing.
4. Documents: whether the versions of the CV that got interviews differ from those that received silence; what this means for prompt 3.
5. Reminders: applications without responses from 5 to 7 days, to which I send one message; interviews after which I wait for a decision, and when to ask.
6. Status of the fact file: whether a new fact (project, certification, number) appeared this week that needs to be added.
7. One sentence: what is the biggest bottleneck at this moment (lack of responses, lack of interviews after the first stage, lack of offers after interviews) and what it says about where the problem lies.
TABLE:
[paste]
Check in the response: whether the changes in point 3 are numbers, not advice. The bottleneck from point 7 indicates which prompt to repeat: lack of responses means prompts 3 and 6, lack of interviews after the first stage means prompt 7, lack of offers after finals means prompt 7 and, sometimes, the goal from prompt 1.
Appendix: Job Searching in Poland and from Poland
The scheme is the same, with a few differences. Polish portals and companies also use ATS, and recruiters search the database for words in the same way. CVs in Poland are usually with a photo and a clause about personal data; consent clauses are not required for data processed for this recruitment, and the rules are explained by the Personal Data Protection Office. A person returning from the USA has an additional topic: how to describe American positions and companies so that a Polish recruiter understands them, and how to document experience. If you work from Poland for a foreign company, see the guide on remote work from Poland. For prompts 1, 3, and 4, add the block below.
I am looking for a job in Poland. Additional rules:
– Job titles and keywords should be taken from Polish postings; provide both Polish and English versions, as some companies advertise in English.
– CV: prepare a version according to Polish customs (order of sections, length, whether to add a photo, how to formulate information about personal data in accordance with GDPR without unnecessary clauses) and a neutral version without a photo. Indicate what cannot be required from the candidate.
– Experience from the USA: for each position, add one sentence explaining the scale and context (company size, market, Polish equivalent of the position); leave titles in English with the Polish equivalent in parentheses.
– Employment form in the posting: employment contract, B2B, commission; list what each means for me, and what to ask.
– Ranges: whether the posting provides salary; if not, how to ask.
– Form: questions about availability, form of cooperation, expectations in gross monthly; explain how to respond.
– If returning from abroad: documents confirming experience and education that the employer may require, and where to obtain them.
What AI Does Wrong When Job Searching
| Error | How It Looks | How to Catch It |
|---|---|---|
| Invented Achievements | “Increased sales by 35 percent” without that number in the fact file | Reading the CV with the fact file next to it; crossing out every number without a source |
| The Same Verbs as Everyone Else | Spearheaded, leveraged, passionate, results-driven | Search the document; banned list in prompt 3 |
| Synonyms Instead of Words from the Posting | “Project management” where the posting says “Project Management Professional” | Column “exact wording” in prompt 2 |
| Stuffing Words | A list of thirty skills without context, white text | Every word in the context of a fact; nothing hidden |
| Format That Doesn’t Parse | Two columns, icons, a table with dates, a header in graphics | Test copying to a notepad |
| Invented Facts About the Company | “Recent award,” “latest implementation” without a link | Click each source before the interview |
| Invented Codes and Data | O*NET code, median salary, name of the industry group from memory | Link to the source with a date |
| Fit Assessment of 8 out of 10 for Every Posting | Flattery | “Assess strictly, like a recruiter with a hundred candidates” |
| “ATS Score of 92 percent” from online tools | A number unrelated to how a specific system and recruiter search the database | Treat as a hint about words, not as a result |
| False Answers in the Form | The model “optimizes” the answer about work eligibility | Truth; consequences extend beyond this application |
CV Format That Passes Through the Parser
| Element | Yes | No |
|---|---|---|
| Layout | One column, top to bottom | Two columns, text fields, sidebars |
| Section Headers | Summary, Experience, Skills, Education, Certifications | “My Journey,” icons instead of names |
| Positions | Title, company, city, month and year from and to, in separate lines or with a clear separator | Dates in a table, only year, “currently” without start date |
| Font and File | Standard font, PDF with a text layer or DOCX if the form requires it | Scan, image, unusual fonts, text in header and footer |
| Contact Information | Name, city and state, phone, email, LinkedIn link, in the document body | Photo, date of birth, marital status, nationality, full address |
| Keywords | In the wording from the posting, in the context of a fact | List without context, white text, repetitions |
| Length | One page for up to about 10 years of experience, two pages above; federal work according to USAJOBS rules | Three pages in the private sector |
| File Name | Name-Surname-Position.pdf | CV_final_v7_new.pdf |
Safety and Privacy
- Do not enter your social security number, date of birth, document numbers, address, or bank details into the fact file or chat. No legitimate employer needs them before an offer.
- Do not pay for work. Fees for equipment, training, “verification,” or a check to be cashed are a scam pattern described by the FTC; conversation solely through messenger and an offer without an interview are signals of this.
- Do not send a CV with facts you cannot defend. Pre-employment verification and the first weeks of work check them better than ATS.
Check model training settings in your account: Claude (Anthropic article), ChatGPT (OpenAI data settings). If you suspect that a rejection decision was discriminatory, information for candidates is published by the EEOC. We discuss how to find a job in the USA without fluent English in a separate guide, and about the first job in Chicago in a guide for Poles in Chicago.
Common Mistakes
- “Improve my CV” as the first prompt. Without a fact file, the model improves style and adds fabrications.
- One CV for a hundred postings. The recruiter’s search engine will not find words that are not there.
- A nice template. Two columns and icons look good for a human and bad for a parser; a human will see the CV only after the parser.
- Tricks. White text, keyword stuffing, job title you didn’t have. Visible after parsing, costly in the interview.
- Only portals. Referrals and informational interviews go into a separate queue; prompt 6 exists for that.
- Lack of a table. Without numbers, you don’t know what works, and you repeat what doesn’t work.
- Company research from the model’s memory. One invented “recent implementation” in the interview costs more than lack of research.
- Falsehoods in the form. Work eligibility, years of experience, degree: these are checked.
This scheme uses AI for research, matching, and practice, not for invention. Every CV, message, and response you send is yours, and you will defend it in the interview and at work. The model can generate a document that passes through the system but does not survive the first week in a new company. The fact file and the prohibition against making things up are more important in this guide than all the prompts combined.
Frequently Asked Questions
Does ATS really automatically reject CVs?
It rejects applications that did not pass the screening questions, and sometimes sorts the rest by fit. Most often, however, it does not “reject” but makes it so that the recruiter cannot find you because the document parsed poorly or does not contain the words they are looking for. Filters on employment gaps and lack of a degree are documented in the Harvard study linked in the text.
Will the recruiter know that the CV was written by ChatGPT?
Often yes, by the vocabulary and generalities, and this works against you. A CV built from a fact file, with numbers and names, does not sound like a model, even if the model helped arrange it.
Do “check your ATS score” tools work?
They show which words from the posting are missing, and that is useful. The percentage “score” does not correspond to any specific system; each ATS and each recruiter searches differently. Treat it as a list of words, not as a prediction.
I have a diploma from Poland. What should I do with it?
List it in the Education section with the name of the university, field, and American equivalent of the degree. If the posting or professional license requires evaluation, organizations from the NACES list do it; in prompt 1, the model checks whether such a requirement appears in your postings.
How to respond to the question about visa sponsorship?
Truthfully, understanding that the question about work eligibility today and the question about the need for sponsorship in the future are two different questions. Prompt 4 breaks this down for your status; in case of doubt, ask an immigration lawyer, not the model.
How many applications per week?
Fewer, better matched, with one person in the middle. Prompt 8 will show after two weeks what the division between channels yields responses in your case.
Does this work for physical and service jobs?
Yes, with less emphasis on LinkedIn and more on people and places from prompt 6; many such companies recruit through referrals and local networks faster than through portals.
Fact-Check Summary
- Definitely true: according to the Harvard Business School and Accenture report, about 99 percent of Fortune 500 companies use applicant tracking systems, half of companies in the USA filter candidates with an employment gap longer than six months, and the group of “hidden workers” in the USA is estimated at 27 million; in New York, employers using automated decision-making tools in recruitment are required to conduct audits and notify; O*NET and the Occupational Outlook Handbook are official sources of job descriptions and salaries. All with links in the text.
- Probably true: recruiters search ATS databases by words and filters, and documents in columns and with graphics parse worse; this is an observation from practice, consistent with how text parsers work. Referrals from employees are read by a person out of turn in most companies.
- What is uncertain: how a specific system in a specific company sorts candidates; whether a given recruiter reads cover letters; what percentage of CVs the “robot” rejects (the number 75 percent circulates without a verifiable source).
- Common myth: “AI will write me a CV that will pass ATS.” It may pass; the problem starts in the interview if the CV contains things you didn’t do.
Comments (0)
No comments yet. Be the first!