How AI is Reshaping Recruiting
The amount of information we are subjected to daily is simply brutal, making it increasingly difficult to distinguish signal from noise.
The impact of AI on recruitment processes is already making waves. According to the 2024 Global Workforce Report by Remote, 3 out of 4 companies are encountering AI-generated CVs that include false information about skills and/or experience.
Excerpt from - Three out of every four companies in Spain receive resumes created with AI with false information - RRHH Press
AI Helps You Overcome the First Interview Hurdle
Many candidates struggle to pass the initial screening in recruitment processes. Stand out with a CV that captures attention. This often doesn’t happen because candidates use templates that are not ATS-friendly, make mistakes in their writing, or fail to tailor their CVs to specific job offers. The reasons vary.
However, thanks to AI, more and more candidates are breaking through this barrier.
AI enables the creation of optimized CVs designed to pass ATS filters. By simply providing a job description, it can generate a tailored CV that successfully navigates ATS screening.
I put it to the test. I asked Claude.ai to create an ATS-optimized CV for a .Net professional.
Here’s the prompt I gave:
"Generate a CV for me with fictional data, but similar to that of a person residing in Spain, experienced as a Software Engineer specializing in .Net (invent the name, experience, etc.), and optimize it for ATS for someone with over 10 years of experience."
This is the CV it generated—a far better one than many candidates I review daily. Without a doubt, this CV would successfully navigate 90% of the ATS systems in the market for a Senior .Net Developer role.
It even included an explanation of the optimizations made and how they were implemented:
Additionally, I provided it with real context—a typical job posting from LinkedIn (omitting only the company name). I then asked it to generate a CV tailored to that specific offer, ensuring it met 95% of the requirements while also optimizing it for AI-powered ATS systems.
Here’s the result: a CV that successfully breaks through the screening barrier. A CV that enables candidates to bypass the filter and proceed to an initial HR call or even a technical test directly.
For reference, this was the prompt I used:
"I’m going to give you a real job offer. Please adapt this CV to meet 95% of the necessary requirements for the candidate to be considered for the position. Additionally, optimize the CV for ATS systems that use Artificial Intelligence for filtering."
The LLM even provided a detailed explanation of how it achieved this:
In a time when the volume of applications far exceeds the capacity to manage them, recruiters face an increasing burden of tasks that add little value to the company. Distinguishing genuine CVs from fabricated or "enhanced" ones becomes a growing challenge—hunting the proverbial catfish.
When overwhelmed by CVs, what’s the best way to filter? Keywords in ATS searches are proving ineffective, as CVs are now crafted to exploit these systems.
Some companies are falling into the trap of filtering by education, prestigious universities, or well-known employers. However, this approach doesn’t guarantee better candidates—it’s merely a way to reduce the pile, often at the cost of missing out on talent.
IA vs IA
This is where the battle begins—AI vs AI. ATS systems using AI to filter CVs, which are also generated by AI. Who will come out on top?
Without a doubt, the real loser here is the company, as it will spend significantly more time filtering CVs and verifying the authenticity of candidates’ experience. What happens when, instead of 20 candidates passing the filter, there are 200?
ATS systems have been using keyword filtering algorithms for years, and now, with AI, have been “supercharged”. However, the real challenge with AI lies in the data samples we provide to train these algorithms. Effective AI filtering requires “megatons” of data—data that many companies don't even track or have access to, in order to personalize the process. In the end, these algorithms are only as good as the quality and volume of their data.
*Excerpt from viterbit.com/blog
But... Who watches the Watchmen? Who ensures that AI doesn’t engage in illegal practices like discrimination?
In Europe, within 18 months, AI models used in sensitive or “high-risk” processes—such as filtering job applications—will require registration. This measure aims to prevent biases in candidate screening, an issue that can easily arise when AI is trained on biased datasets.
Real Superpowers: A Tool to Boost Your Chances of Success
Interviewing.io—an online platform for technical testing and interview preparation—conducted an experiment this year with candidates and companies to see how many interviewers would notice if candidates were using LLMs during the interview process.
SPOILER: None of the interviewers realized that candidates were "cheating" by using LLMs. Moreover, 72% of the interviewers stated they would hire the candidates who passed the tests.
But is using AI really “cheating”?
It depends on the rules set by each company for their technical tests. However, there’s no denying that AI has become a tool anyone can use in their daily work.
So, what’s the solution? Should we prohibit or encourage the use of AI in technical testing?
Common sense suggests that banning a tool that will likely be used daily in the job itself—and is readily available to everyone—shouldn’t be a detriment to the hiring process. Instead, it could be seen as an aid.
That said, this creates new challenges for interviewers. It becomes increasingly difficult to assess a developer’s true experience and skills in areas like programming fundamentals, system design, or coding proficiency. The lines between different levels of seniority—such as an L4 and an L5—are becoming blurred in interviews.
The devil AI is in the details
Nobody likes being deceived, and feeling like you’re not competing on a level playing field is inherently frustrating. As an interviewer, realizing that the answers you’re getting might not be from the candidate but from an AI creates a sense of unfairness.
Here are some red flags to watch for during interviews that could suggest candidates are using AI:
- Pauses After Questions: Candidates take unusually long to respond—longer than what’s typical for thinking or organizing thoughts.
- Looking at Another Screen: If you notice them typing while you’re speaking or glancing at another screen, they might be inputting your question into an LLM. Alternatively, they could just be replying to a work chat.
- Vague, Overly Wordy Answers: If responses consistently start with lengthy, convoluted explanations or take unnecessary detours, much like a generic LLM prompt response, it’s possible (though not certain) they’re using AI.
Disclaimer: Let’s not forget that we’re not detectives. These behaviours can happen naturally for legitimate reasons. People often pause to think or organize their thoughts. Many interviews occur during work hours, where candidates might be multitasking—glancing at emails or Slack while keeping the video call open on another screen.
And this isn’t entirely new. Long before LLMs, candidates were looking up answers on StackOverflow or Googling during interviews. Technology evolves, but the challenges it brings to the hiring process are familiar.
Renew or Die: AI Forces Us to Redefine Processes
It’s a fact, and it’s happening. AI is making life harder for everyone involved in the recruitment process. So, what can we do?
As much as it may sound like classic McKinsey bullshit, the answer is simple: adapt to change. There’s no alternative. At present, there are no tools that can reliably differentiate between candidates using AI and those who aren’t.
AI compels us to rethink and reshape our selection processes as we know them. It’s time to innovate, evolve, and embrace new strategies to stay ahead.
Redefining CV screening or Job Offers
Option 1: Redefine CV Screening
When we delegate CV reviews to an ATS because we don’t have time to go through the volume of applications, we risk missing out on candidates who haven’t optimized their CVs for that specific ATS.
If we enrich the ATS AI with our job description requirements and data from previous processes or successful candidates, we must be mindful of the bias in the data we feed it. This is a complex issue, especially when buried under a pile of CVs.
Our solution? We do it manually. We review, prioritize, and respond to each candidate, step by step, one by one.
Option 2: Redefine the Job Listing
As my colleague Anxo Pérez once said, "English is taught poorly. Period." The same could be said about job offers. In most cases, job descriptions are poorly written. A quick search for "Software Engineer" on LinkedIn Jobs will reveal countless listings that are nothing but endless lists of requirements—and in recent times, an even longer list of perks or benefits.
Letting hiring managers write the job offers is like asking children to write the letter to Santa Claus. They’ll end up searching for the perfect candidate—the unicorn, the one who ticks every box. But the reality is, that person doesn’t exist. This leads to either few applications or overly long hiring processes for a role.
Tweet translation:
«Companies take themselves so seriously in job offers that we reach the height of absurdity.
In a startup with fewer than 60 people in the tech team, it's rare that you need such a high level of specialization as we usually demand in job offers.
We've probably drunk too much from what those who really know tell us.
The reality is that most of the job offers I post should start like this:
«Read the offer and if this is more or less what you want to do, apply»
The verb 'want' is much more important than it seems.»
Redefine Technical Questions
Many of the technical questions used in interviews are taken from the Internet. And the LLM's have been trained with Internet data. Therefore, they also have the answers. So, if your candidates have access to the Internet, they have the answers to your questions. To yours and to those of thousands and thousands of companies around the world.
That's why, basic and generic technical questions no longer make sense in the processes. It's going to be very difficult to distinguish who has real knowledge from who has copied the LLM's answer.
Redefining Technical Testing
Does it still make sense to use the same asynchronous technical tests when you can't tell how much of the candidate's solution was written by them, and how much was generated by an LLM?
Should we prohibit the use of LLMs in selection processes, or is it more logical to encourage candidates to use them?
Asynchronous technical tests are starting to face this issue. By the time candidates present their code for review, they may not be able to fully explain their solution. Or, on the other hand, you might find that every candidate you interview “passes” the technical test, making it hard to gauge their actual abilities.
As a result, companies are rethinking their internal processes and taking positions on whether to allow:
- The use of LLMs in technical tests
- The exclusion of LLMs, opting instead for live coding or pair programming, which not everyone is a fan of
In a conversation on X/Twitter, Javi and Jorge discuss the use of LLMs in asynchronous tests. They both suggest a practical approach: if a candidate uses an LLM, they should provide a .txt file detailing the prompts they used, along with an additional section explaining how they would integrate an LLM into a production environment for a similar task.
Other companies are adopting real business cases as technical assessments, encouraging AI use. Since these are real-world scenarios, they often pay for the technical tests, turning them into something like a hackathon within the hiring process.
New Technical Tests: The Skill of Writing Effective Prompts
Some companies are starting to include tests focused on a candidate’s ability to craft effective prompts for LLMs. This trend is gradually gaining traction and is likely to become a standard part of many recruitment processes.
As LLMs become a common tool in daily workflows, the ability to maximize their potential—through well-crafted instructions—will increasingly be seen as an essential skill. Writing effective prompts, which requires a mix of clarity, precision, and creativity, will soon be a requirement in many job descriptions.
In the Near Future, Technical Skills Will Be Less Important in the Tech Job Market
As AI-powered agents like Copilot, Cursor, and others continue to evolve, they will enable developers to contribute to teams much more quickly, regardless of their familiarity with specific languages, stacks, or tools. This means that technical skills, once central to the hiring process, will become less relevant over time.
Instead, soft skills such as written communication, problem-solving, and the ability to learn and adapt will take precedence. These are qualities that AI cannot replicate or replace.
In the job offers of the near future, soft skills will be prioritized over technical expertise, as they remain essential for thriving in a dynamic, AI-driven environment.
The Growing Importance of Soft Skills
In a time when technical skills increasingly depend on AI, soft skills are becoming even more critical.
AI will enhance the abilities of developers at all levels, from juniors to seniors, and the distinctions in mastering languages and technologies will blur more and more.
As a result, we’re likely to see job offers placing greater emphasis on cultural fit—focusing on alignment between the candidate's current situation and the company’s stage of growth, experience in similar organizations, adaptability to comparable work models, and integration into new environments. Soft skills will become the deciding factor in who moves forward in the hiring process.
Key soft skills include:
- Teamwork
- Problem-solving
- Adaptability to change
- Learning ability
- Autonomy
- Written and verbal communication
Candidates who may have slightly less technical expertise but excel in soft skills will often outshine technically superior candidates with weaker interpersonal abilities. AI will help level the playing field in terms of technical skills, making soft skills the true differentiator.
How to Integrate AI into Your Recruitment Processes
One of the first recommended steps is to run all your current technical questions through LLMs to see how they respond. This will give you a baseline of answers, helping you better distinguish between candidate responses and AI-generated ones. If you know what LLM-generated answers typically look like, you’ll be better equipped to spot similarities.
Another key piece of advice is to start evolving your selection process:
- Write better job descriptions, considering how AI-generated CVs will pass through your filters. Focus more on soft skills.
- Improve CV review processes, whether done manually or with ATS tools.
- Shift away from generic questions and include more specific scenarios or questions based on the real context of your company.
- Decide how to handle asynchronous technical tests and determine your stance on candidates using LLMs for solving challenges.
- Consider evaluating technical skills through live coding or pair programming sessions. Decide whether to permit the use of AI for code challenges during these evaluations.
By adapting your process thoughtfully, you can leverage AI as an ally while ensuring a fair and effective assessment of candidates.
Conclusions
The use of LLMs in recruitment processes is already driving drastic changes—and this shift will only continue. The same goes for technical interviews and tests, which are evolving rapidly.
The main challenge lies at the top of the funnel: the growing number of applications containing false or “optimized” information that bypass ATS systems—and recruiters like me—makes daily work increasingly complex.
This issue persists if the screening process lacks depth or if AI tools are employed for filtering. Outsmarting an AI-driven tool can be just as easy as bypassing an ATS.
The challenge extends to interviews and technical tests:
- Will we start seeing a rise in live coding tests?
- Or will companies embrace the use of AI during the technical assessment phase?
- How can we determine whether a candidate truly understands their responses to technical tests, or if they’ve primarily relied on copying?
- Will there be an increase in requests for references and direct calls to former employers?
What’s clear is that, for now, candidates hold the upper hand in the selection process. This makes it more critical than ever to distinguish the signal from the noise.
* Although I often use the first-person plural to refer to us recruiters, at Manfred, we don’t rely on CV review software or AI. Each scout (recruiter) personally reviews every CV submitted for an application one by one.
**This article draws inspiration from Gergely Orosz's blog, published on The Pragmatic Engineer.