How AI is influencing recruitment processes

Does AI really “choose” candidates on its own, and is the human role no longer needed? I’m Viktoriia Kopeikina, HR Manager, Recruiter, and Digital Marketing Specialist at 111Techs. Since 2019, I’ve been involved in hiring from the moment a manager says, “we need a specialist,” all the way through to successful years of collaboration with that person. Over the years, I’ve gone from manually reviewing thousands of CVs to using AI-assisted screening, and I want to share why automation has become necessary and exactly what AI does in our work. Why automation became a necessity Previously, our process was simple: Google Drive, spreadsheets, and manual review. It worked for us until we faced a real challenge. At the beginning of 2024, we opened a manual QA engineer position and received 500+ applications in one week. With that volume of CVs, there was a real risk that a strong candidate could go unnoticed. Even when I reviewed applications carefully, it was physically very difficult to process hundreds of CVs with the same level of quality. Unfortunately, that is exactly what happened: although we did find the right person, it took about two months. That happened for several reasons: A large number of applications that did not decrease over time Many irrelevant candidates We waited just over half a month for our future employee to join us Still, the main issue was the high number of applications, and this is not a unique situation today. According to Djinni analytics, as of June 2026, 89,097 candidates are competing for 7,731 vacancies in the Ukrainian tech sector alone. The market has not become quieter, and the risk of “losing” a talented candidate in such a volume is only increasing. That was when we first started seriously thinking about how to automate the initial screening stage, so we would not potentially miss strong candidates simply because of the volume of applications. We did not have time for a long search for a new system, so we started with what we had: we created a Google Form so applications would automatically flow into a spreadsheet with the required criteria filled in. It was not perfect, but it was our first real step toward automation. The Google Form helped us structure the application flow, but two difficult tasks remained: Dozens of CVs come in for a single vacancy, and among them are both candidates who truly match the requirements and have carefully read the job description, and those who send applications in bulk to everyone, without reading the requirements. Today, there are already various services and AI tools that allow candidates to adapt a CV to a vacancy in just a few clicks or automatically send it to 100+ vacancies. Below is a real example of one candidate (name changed) who applied to our company for seven vacancies at the same time. Or mid-level and junior candidates apply for senior positions, while only frontend developers apply for full-stack vacancies. I have very little time to analyze hundreds of CVs manually or hold introductory calls, and I do not want to spend it on candidates who clearly do not meet the requirements. I want to find and speak with the people who match the profile as closely as possible. Analyze CVs that arrive in different formats. Usually, this is a DOCX or PDF file prepared in a way that is convenient for the candidate: one person writes “3 years of Python,” another simply says “development experience,” and someone else hides the key information in a five-column table. Ideally, candidates would enter their data into a structured form themselves or submit CVs in a standardized format, such as HR-XML (https://schemas.liquid-technologies.com/hr-xml/2007-04-15/) or Europass XML. But write in the comments: how many of you have a CV in that format? So, it became clear what AI should take on and why it is becoming a necessity. Where we use AI the most At 111Techs, AI helps me the most as a recruiter with CV screening. I tested several popular ATS platforms with built-in AI, including Manatal and Zoho Recruit. That was when I noticed a problem: a candidate with strong experience but an unconventional CV or career path could receive a low score from the system. When we reviewed those profiles manually, it turned out that the person was an excellent fit for the role. In other cases, the system could evaluate the same CV differently each time. I found a great article about another ATS service describing a similar case: https://danunparsed.com/p/hackerrank-open-source-ats This became an important conclusion for us: we do not want AI to evaluate candidates. We want it to structure data: employment start and end dates, job titles, company names, technologies, education level, and English proficiency level. Today, we use Airtable as the foundation for collecting and structuring applications, with AI connected to help structure and populate the data. What the formula calculates, not AI Once the facts have been extracted and organized into fields, the math begins and AI has nothing to do with that part. For selection, we use ranking: a formula that calculates a CV score based on the key parameters for the vacancy. For example, for a Senior Python Software Engineer role, the following criteria are important to us: overall experience in the specialization (hard skills relevant to the role and how experienced the candidate is in the profession overall); relevant education (computer engineering, computer science, software engineering, informatics); English proficiency level. To extract and classify the required data, we use an LLM, specifically the OpenAI GPT-5.5 model. To illustrate the difference between "evaluating a candidate" and "structuring data," I’ll show one of our working prompts. Extract the candidate’s employment history from the attached PDF or DOCX resume/CV. Return only valid JSON. Do not include markdown, explanations, comments, or any text outside the JSON. Output must be a JSON array. Each element represents one employment position and must contain exactly these fields: { “date_start”: “YYYY-MM or YYYY-MM-DD”, “date_finish”: “YYYY-MM or YYYY-MM-DD”, “job_title”: “string”, “company_name”: “string”, “software_development_relevance_score”: 0.0, “software_development_relevance_reasoning”: “string” } Extraction rules: Extract only actual employment positions. Ignore education, certifications, skills, projects, publications, volunteer work, and awards unless explicitly listed as paid employment. Preserve the order from the resume, preferably most recent position first. Do not invent information. Date normalization rules: Convert all dates to ISO format. If only year and month are known, use YYYY-MM. If a full date is available, use YYYY-MM-DD. If only a year is available, use YYYY. For current positions (“Present”, “Current”, “Now”, etc.), replace the end date with the current year and month in YYYY-MM format. If a date cannot be determined, use null. Software development relevance scoring: Assign a score between 0.0 and 1.0 indicating how strongly the company and position appear related to software development, software engineering, IT services, SaaS, technology products, software consulting, platform engineering, DevOps, data engineering, AI/ML, cloud services, or similar technical software domains. Guidelines: 1.0 = clearly a software company and/or software engineering role. 0.8–0.9 = strongly software-related organization or role. 0.5–0.7 = partially software-related, mixed technology/business environment. 0.2–0.4 = limited evidence of software development relevance. 0.0–0.1 = unrelated to software development. Reasoning rules: Provide a concise explanation for the assigned score. Reference evidence found in the company name, job title, or resume context. Do not speculate beyond available evidence. Example output: [ { “date_start”: “2022-01”, “date_finish”: “2026-06”, “job_title”: “Senior Software Engineer”, “company_name”: “Acme Cloud Technologies”, “software_development_relevance_score”: 1.0, “software_development_relevance_reasoning”: “Software Engineer role at a cloud technology company developing software products.” } ] Once all the facts have been extracted, each criterion receives its own weight: overall experience in the specialization — multiplied by 2; relevant education — coefficient of 1; English proficiency level: B2/C1 = 2 points, B1 = 1 point. The final formula mathematically calculates the overall rating—a ranked list where the highest scores go to those who best match the key requirements of the vacancy. I then personally review these candidates and invite them to an HR interview. So, at 111Techs, AI does not make decisions, it only restructures the first stage of screening. What this changed in practice The manual ranking stage disappeared. Previously, I read every CV myself; now the formula handles prioritization, and I can immediately call the most relevant candidates. Almost all the time is spent on the right candidates. Previously, every second to fourth candidate was irrelevant, and a quarter of a 40-hour workweek was spent on calls that went nowhere. Now, the focus is on people who are truly a fit. Top candidates become visible from the first days. The strongest profile always rises to the top of the ranking, which sharply reduces the risk of “not getting around to a great candidate.” Speed. From CV submission to a Slack notification with a TOP-10 report, about a minute passes. There are no notifications for irrelevant candidates. How to make your CV clear for both AI and the recruiter Now that it is clear what AI actually does during screening, a logical question arises: how can you prepare your CV, so the system correctly reads all the important information? Here are a few tips that can help. 1.Describe not only the role, but the full stack. “Python developer” is one field. “Python, Django, PostgreSQL, Docker” are four facts that will actually be captured in the structure. 2.Include your education and language level. These are separate criteria in the formula—without them, the score will be lower, even if you actually have them. 3.Do not try to “guess the keywords.” Simply copying wording from the vacancy will not work: AI does not evaluate matches; it extracts facts. Yes, inaccurate data can temporarily move you higher in the ranking, but it will become obvious during a real interview. Write what is true. 4.Choose a simple format: tables, columns, graphics, and non-standard fonts make reading harder—AI may simply fail to extract a fact from a beautifully designed cell. The most reliable option is clean text in one column, clear section headings, and a PDF or DOCX format. The simpler the format, the more accurately the model will extract your data. 5.Keep your LinkedIn and CV consistent: recruiters always check profiles, and discrepancies between a CV and LinkedIn immediately raise questions. The initial background check will be done through social media. 6.Do not overload your CV: a large amount of information does not increase your chances—it reduces readability. Two years of experience described clearly and to the point makes a better impression than five pages listing everything you have ever done. As an experienced recruiter, I spend 30 to 60 seconds reviewing a CV before deciding whether the profile is potentially interesting, so a strong CV should create a clear impression right away. Aspects of hiring that cannot be automated I have read many LinkedIn posts and articles about AI-led interviews. Personally, I am skeptical of this approach. As soon as AI starts automatically evaluating answers, candidates can just as easily generate polished responses with AI, and the interaction risks becoming performative on both sides. Just let your AIs talk to each other and call it a day 🙂 Or take an example from our own practice: a candidate received a high score from the algorithm—everything matched the criteria. However, during the interview, it turned out that part of the experience had been exaggerated: the person listed technologies they had only worked with superficially, not as their main stack. No algorithm would have noticed that. Something essential disappears in that kind of exchange. Of course, everyone uses AI for some things, but a live conversation reveals qualities that structured analysis often misses: hesitation, curiosity, confidence, uncertainty, emotional intelligence, sincerity, or genuine interest in the candidate or the role. These moments are difficult to measure, but they shape the recruiter’s understanding of whether the candidate truly has relevant experience, what kind of experience it is, and whether they will fit into the team or organization. For this reason, I believe recruitment work is becoming hybrid, as it is happening in every specialization. Automation takes over part of the operational workload, while specialists focus more and more on interpretation, communication, and decision-making. Conclusion A few years ago, I spent hours reviewing CVs, manually sorting candidates, and working with endless spreadsheets. Today, AI handles a significant part of that work. Does this mean AI has replaced the recruiter and “chooses” candidates on its own? No. It has simply taken over the work that a person should not have to do manually. That is why I see the future of recruitment not as “AI instead of a recruiter,” but as “AI + recruiter.” By the way, this article is about AI, but it was written by me. There is no such thing as a perfect prompt, and it is important to me that my voice, my thoughts, and my experience are visible here. This is exactly how I see recruitment: automation can help at the first stage, but the human always comes back into the process. We regularly share team life, events, and how our work is organized from the inside. Join our LinkedIn page: www.linkedin.com/company/111techs (the link is also on my page), and let’s stay connected 🙌