How to Get a Job with AI? Advice from an HR Pro with 20 Years of Experience
What does AI really change in a job interview for the candidate — and how can ChatGPT and Claude be used to prepare better rather than to swap yourself for an avatar, when a person can't be known from one interview anyway?
Anna Volchek says she has advised people on CVs and interviews for twenty years, and for the last couple of years she has sent everyone to language models first — and fewer and fewer come back to her. The foundation of hiring doesn't change: a person can't be known even from a five-hour interview, she puts the probability of a successful hire at 50 to 50, and star investment-bank analysts, according to a Harvard Business School study, began to show successful results after moving to other banks only five years later. The cover letter, according to a Yale University study of the Freelancer platform, stopped working once an AI assistant appeared: the link between a letter tailored to the vacancy and the employer's response weakened by more than half. Anna breaks preparation into six steps — a CV tailored to the role, a company dossier checked against primary sources, your own questions, a rehearsal with the ten hardest questions, terms and law, the rules of the practical task — and warns that in a German test the models advised women to ask for a lower salary. Greenhouse data cited by Anna show that about half of candidates had an AI interview over the year and around 70% of them weren't warned, and in a test by German journalists a video personality assessment changed with a hat, a coat and a headscarf. After the interview she advises going over your answers with ChatGPT and always asking about the next step: by the same Greenhouse data, about 48% of candidates complain that the employer vanishes.
What to watch for
Key takeaways
Today ChatGPT or any other model can prepare a CV, prepare you for an interview and gather information about a company, and fewer and fewer people come back to Anna after the models. AI can even help you get through an interview with prompts, but is unlikely to make you successful in the job if they hire not you but your avatar.
According to a Yale University study of the Freelancer platform, a letter tailored to the vacancy got an employer's response roughly twice as often. When AI that helps write such letters appeared on the platform, the link weakened by more than half: far less effort is needed now, and the letter has lost even its residual value.
An interview lasts an hour, two, sometimes five hours, but people live together for years and don't know how a partner will behave in other circumstances. Anna's husband and a close friend started a business, it didn't work out, and the friendship fell apart: they knew each other well as friends, but not as partners.
To Alexander's question in a Sunday episode about what percentage of sales hires succeed — say, 60 to 40 — Anna answers that the probability of a successful hire is always 50 to 50. It sounds like reading tea leaves, but an interview always carries a right to make a mistake, and not in a negative sense.
Harvard Business School studied star analysts at investment banks: after moving to other banks, they began to show extra results only five years later. Success is also the team, the environment and the circumstances, so a rejection is not necessarily an assessment of the candidate, and the interviewer is not placing a final bet.
Nobody knows how many job interviews take place in the world each day: according to one study, about four hundred and eighty million people are actively looking for work, and in the US alone companies hire more than five million a month. Any study sees one slice, so Anna treats them as signals, not as law.
ChatGPT, Claude and any model, Anna says, perfectly handle tailoring a CV to the role, tasks, company, environment and manager. The filler and descriptive phrases go, results and concrete actions remain, and the model will tell you what is missing and what your strengths and weaknesses are.
Even today people come to interviews without knowing what the company does. With a model it is easy to gather a public company's capitalisation and revenue, growth trend, headcount, offices, leadership and strategy, articles and press releases about the interviewer — but every fact and figure must be checked against the primary source.
A study well known in the recruiting world and repeated a million times: the more questions, including clarifying ones, a candidate asks, the more the other side likes them. Questions should be built on the dossier: if the manager for the position was appointed six months ago, it is a good reason to ask what is expected of a new candidate.
Anna asked ChatGPT and Claude whether there was anything special for preparation, and all the chats named rehearsal as the key path. You can put the vacancy, CV and dossier side by side, ask for the ten hardest and the ten most non-trivial questions, play out scenarios and go over where you sounded strong and where weak.
Studying the European pay transparency directive, Anna came across a test from Germany: models were asked one question — what salary to request at an interview — with the same position, level and city, only the asker's gender changed, and the models advised women a lower sum. So AI's salary advice must be checked.
In the European Union, AI in hiring and human resources falls into a very high-risk zone. Even so, according to Greenhouse, about half of candidates had an interview with AI over the past year, and around 70% of them were not warned in advance — so the question is worth clarifying.
There is no finding, Anna says, that dry, emotionless answers in an AI interview improve the result. Journalists from a German media platform tested a program that assessed personality from a short video: the candidate changed her hat, the colour of her coat, put on or took off a headscarf — and the score changed noticeably.
AI prompts during an online interview, according to the studies Anna found, can't be detected for certain in 70% of cases. Her position: if it hasn't been agreed, we don't use it, and if a candidate does it so masterfully that it can't be seen — perhaps that is the person you need. It may help you get the job, but not always to succeed in it.
While you remember your answers, go over with a model what went well, what was weak and where there was too much description, and if the interview was recorded by mutual consent, upload the transcript. It is a colossal training tool, and career consulting in its old form, Anna believes, already seems to be dying.
According to Greenhouse, about half of candidates — 48%, as Anna recalls — complain that the employer simply disappears without an answer. In her teams one of the first rules was always to answer the person personally; a candidate should always be told what the next step is and how long to wait.
Anna's main advice: use AI to strengthen your competencies and opinions, not to replace them. As in preparing for a podcast, materials and studies can be gathered in concentrated form, but AI won't replace your words and your line.
What this episode is about
A solo ToTheMoon episode with Anna Volchek, who has interviewed people and advised on CVs and interviews for twenty years. For the last couple of years she has sent everyone to language models first, and people have come back to her less and less: with ChatGPT or any other model a person prepares a CV, prepares for an interview and gathers information about a company on their own. AI can even help you get through an interview, but is unlikely to make you successful in the job if they hire not you but your avatar; a person can't be assessed in one or two interviews, so an interview is not the final bet.
AI changes a lot in interviews, but the foundation stays the same. Anna never treated the cover letter as a criterion, and a Yale University study of the Freelancer platform showed that a letter tailored to the vacancy got a response twice as often, but once an AI assistant appeared that link weakened by more than half. Personal contact always mattered more to her than credentials — how a person communicates, what soft skills they have and whether she is ready to do this work with them.
An interview lasts an hour, two, sometimes five hours, but people live together for years and don't know each other in other circumstances: Anna's husband and his close friend started a business, and the friendship fell apart because they didn't know each other as partners. Anna puts the probability of a successful hire at 50 to 50, and according to a Harvard Business School study star investment-bank analysts showed successful results after moving only five years later: success is also the team, the environment and the circumstances. So a rejection is not necessarily an assessment of the candidate as a person, and the interviewer is not placing a final bet.
There are a million studies of interviews, Anna says, but each sees one slice: according to one of them about four hundred and eighty million people are actively looking for work, and in the US alone companies hire more than five million a month. So for her studies are signals, not law. A candidate's preparation, however, can be made better and more effective today, and Anna breaks it down into steps.
Models help tailor a CV to the role, company and manager, cut the filler and show results. A company dossier — a public company's capitalisation and revenue, growth trend, headcount, offices, leadership, strategy, articles about the manager who will interview you — is easy to put together, but the facts must be checked against the primary source. From the dossier you prepare your own questions: the more questions a candidate asks, the more the other side likes them. All the chats name rehearsal as the key path — the ten hardest and the ten most non-trivial questions and a review of where you sounded strong and where weak.
Terms can also be explored with models, but here AI must be checked: in a German test, asked the same question with the same position and city, the models advised women to name a lower salary than men — Anna came across it while studying the European pay transparency directive. It is worth sorting out in advance the types of contracts and employment agreements in different countries and checking with the employer whether AI is allowed in the practical task.
AI may be on the other side of the screen. In the European Union hiring falls into a very high-risk zone, but according to Greenhouse about half of candidates had an interview with AI over the past year, and around 70% of them weren't warned. There is no finding, Anna says, that dry answers without a smile improve the result, and a program tested by German journalists changed its personality assessment of a candidate with a hat, the colour of a coat and a headscarf.
Candidates also use AI prompts during online interviews, and according to the studies Anna refers to, in 70% of cases it can't be detected for certain. Her position: if it hasn't been agreed, we don't use it, and if a candidate does it invisibly and masterfully, perhaps they will work the same way; but you can't fool yourself. After the interview she advises going over your answers with ChatGPT and, with mutual consent, the transcript of the recording; career consulting in its old form, she believes, is dying. And always ask about the next step: about 48% of candidates, according to Greenhouse, complain that the employer vanishes. The main advice is to use AI to strengthen yourself, not to replace yourself.
The episode separates what AI changes in hiring from what stays: a candidate's preparation has become more accessible and stronger, but a person still can't be known from one interview, and the models and assessment systems themselves need checking — from salary advice to video assessment. Anna Volchek's six steps turn ChatGPT and Claude into a training tool that strengthens the candidate rather than replacing them.
Episode transcript
The episode is in Russian; below is an English reading guide to the transcript (the full EN transcript is a machine translation). Voice matching applied to 45 segments: 45 identified, 0 mixed, 0 probable, and 0 unresolved.
Read transcript on a separate page
Loading…