Headhunters in the AI industry can earn nearly 3 million yuan per deal?
Recently, a screenshot went viral in various groups: a certain AI company is recruiting an AI algorithm expert with a total annual salary package of as high as 10.5 million yuan; the headhunter fee is 26%, which is 2.73 million yuan.
As the news broke, some people envied the high value of AI core talent, while others lamented that the AI wave not only raised the salaries of practitioners, but also allowed related roles in the industry chain to find new opportunities to make money. The scarcer the talent, the more intense the competition, and headhunters standing between companies and candidates have also been pushed to the forefront.
This competition is not just hearsay. According to Euronews, giants such as OpenAI, Meta, Google, and Anthropic are vying for a very limited number of top AI researchers. Meta even offered a four-year compensation package worth approximately $250 million to recruit 24-year-old AI researcher Matt Deitke. Domestic competition is also driving up the prices of a few core positions, with the annual salary of some core positions in ByteDance's Top Seed exceeding 6 million yuan.
And in this escalating round of high-price competition, a intriguing paradox has also emerged.
From a technical perspective, headhunters seem to be a profession that can be easily replaced by AI. Companies can use AI to search for resumes, screen candidates, and match job openings, with much higher efficiency than traditional manual recommendations. At the same time, top AI talent is in high demand and often has multiple job offers to choose from, making it seem unnecessary for them to rely on headhunters to find employment.
But reality has taken a different turn: the headhunting business in the AI industry is booming, with professions that can be made more efficient or even replaced by AI earning more money in the AI talent war.
With many questions, we found Peter, an employee of another subsidiary of the headhunting group mentioned in the online screenshot. He confirmed that the order in the screenshot does exist, but the 2.73 million yuan is a company-level service fee and does not mean that a single headhunter can take it all.
In between, there are multiple layers of distribution among companies and teams, so the amount that actually ends up in an individual's hands is probably around 400,000 to 500,000 yuan.
What's even more worth exploring is that when AI can already screen resumes and match job openings, why do top talent recruitments rely more heavily on headhunters. What do headhunters truly provide in the negotiations between companies and candidates who have multiple job offers?
The answer lies in the real workflow of headhunters in the AI industry, as told by Peter -
AI Headhunters on the Rise, But Not as Successful as They Seem
I switched to being a headhunter because I didn't want to stay on the client side anymore.
I started my internship in 2022 doing recruitment work on the client side. For those few years, I was almost always working on finding talent in the same industry, and the positions and candidates I encountered were very specialized. But being a headhunter is different - it's like pushing open one wall of the office: I'm facing different companies and different fields at the same time, and my options have increased exponentially.
So in 2025, I decided to switch to the other side and start helping companies find people in the market.
After joining, I discovered that being a headhunter isn't as simple as receiving a job description and passing it on to candidates. The work is divided into two main parts: one is development, which involves proactively finding companies with recruitment needs and negotiating cooperation with founders, HR, or other responsible personnel; the other is delivery, which involves receiving job openings and presenting suitable candidates to companies. I handle both aspects, not only needing to clarify what kind of talent clients are looking for, but also bearing the pressure of ultimately delivering or not delivering the right candidates.
Although I am based in Ningbo, most of my AI clients are concentrated in Beijing, Shanghai, Shenzhen, and Hangzhou. The majority of requirements can be discussed and clarified over the screen, and when face-to-face communication is necessary, I can buy a ticket and arrive the next day. After doing this for a while, I've found that for headhunters, the distance between cities is not a major issue.
What's truly difficult to overcome is the distance between different AI businesses.
Similarly, among AI companies, some develop software while others produce hardware; some serve enterprises, while others target consumers directly. Even when it comes to products, some sell efficiency, while others provide emotional value. With different businesses come entirely different talent requirements. Therefore, every time I encounter a new client, I must first clarify what the company is currently doing, the founder's background, and the current stage of the enterprise, among other things.
Only by answering these questions can we determine where to look for people.
As these demands become clearer, the market's hot and cold spots will gradually emerge. World models, embodied intelligence, and robotics have received relatively more funding, with hardware roughly in the second tier, and software varying depending on the specific application.
Where the money goes, recruitment demand naturally follows.
In terms of specific positions, the demand for front-end jobs is relatively low, while back-end, full-stack, algorithm, and product positions are extremely popular. Once the product is actually developed, companies still need to figure out how to sell it, making business development an indispensable part of the recruitment chain.
That said, more openings don't mean everyone can command a fat paycheck. Most job switchers still see raises of just 10% to 20% over their previous salary, with compensation largely cash-based. Whether stock options are included, and how much they account for, depends on the role and the candidate's title.
Headhunters' income also follows the same reality. The common service fee rate in the industry is around 20%-25%, and the annual performance of traditional headhunters is usually around 700,000 to 1 million yuan. Including base salary and commission, personal annual income is typically around 200,000 to 400,000 yuan. However, when they have access to talent that companies cannot find and other headhunters have difficulty reaching, they also have the opportunity to negotiate higher fee rates.
There are many AI positions and high talent mobility, resulting in more deal opportunities than some traditional industries. However, a hot market does not mean every headhunter can earn more money, as income ultimately depends on delivery: finding the right person is one aspect, but they must also be hired and successfully onboarded by the company to generate performance.
Moreover, even after an employee has been hired, the money is still not considered to be truly earned.
Companies typically set a probationary period of three months or six months, and headhunting fees are paid in two or even three installments. If a candidate leaves during the probationary period due to a mismatch in abilities, the headhunter must replenish the position within the agreed-upon timeframe; if they fail to do so, they will refund the fees already collected.
So, what headhunters earn is not the money for "finding someone," but the money for "finding the right person and making them stay."
Having transitioned from the client side to headhunting, and then from traditional recruitment to AI, I initially thought it was just a switch to a popular new track. It wasn't until later that I realized the trend may bring opportunities, but whether you can seize them depends on finding the right person.
The Best Time to Poach Talent Is After Hours
Many candidates' phones can only be reached in the evening.
During the day, they are busy with meetings and projects, and won't stop work to answer calls from unfamiliar numbers, so my work schedule gets pushed back. While others work seven or eight hours a day, I may work an additional three or four hours; sometimes I continue chatting with them on weekends, and sometimes I rush to their city just to have coffee with one person.
The public's perception of headhunters poaching talent often carries a sense of drama, as if it's a high-stakes pursuit. In reality, it's not that exaggerated. I won't stalk someone's social media to manufacture a chance encounter, that's too forced. If I want to meet someone, I'll reach out first and schedule a formal meeting. The first time we sit down, we don't necessarily have to discuss job openings. Instead, I'll get to know them, listen to what they're working on, and then gauge whether they're interested in changing environments.
Before this, there was a period of invisible preparation.
When a company comes to me with a job posting, I won't immediately dive into the resume database. Instead, I'll ask: Why are you hiring now? How long have you been looking? What stage is the team at? What kind of person are you looking for? What salary and benefits can you offer? Only after the company has clearly stated their needs will the real work of recruiting begin.
I will start by drawing a "talent map": where the core talents in this field are concentrated in terms of cities and companies, and how to get in touch with them through various channels. Recruitment platforms and LinkedIn are the most commonly used entry points, and offline hackathons, forums, salons, and exhibitions cannot be neglected. When suitable resumes are found but the individuals cannot be contacted, it is possible to continue searching through their alma mater and classmate relationships.
However, even if the direction is correct, it's not necessarily easy to find the right person immediately. The AI industry is updating too quickly, with new concepts emerging constantly, and the same capability may be packaged into completely different job titles at different companies.
FDE is a typical example.
Different companies have inconsistent understandings of FDE, and the candidate profiles they provide are sometimes vague. Searching solely by job title can easily lead to biased results. In such cases, the only option is to first select a batch of candidates who roughly meet the requirements and recommend them, then continuously adjust based on the company's interview feedback. After several rounds, the initially unclear requirements will gradually reveal their true contours.
Company names may change, but the standards for selecting talent have many commonalities. Being intelligent, able to think independently, and having experienced the process of going from 0 to 1 are the most common requirements. Startup companies will also often ask one more question: is this person willing to grow with the company?
Apart from the soft requirements, there is a more direct screening criterion.
Some companies explicitly require candidates to come from 985, 211, or double-first-class universities, and even specify the school, academic level, and experience at top companies. Age is also a barrier: some companies want to limit it to 30 years old, while others may relax it to 35 years old. However, being over 35 does not mean there are no opportunities at all - if the project experience is outstanding and there is a background at top companies or in entrepreneurship, it is still possible to try for product positions, although opportunities for technical positions will be much fewer.
After finding a candidate that fits the profile, the next hurdle is negotiating salary.
Outstanding candidates often have more than one job offer in hand, with some proposing a 30%-40% increase in salary. If the original salary and job title were mismatched and the interview results prove that the candidate's abilities are indeed a good match, I can understand such a request. From the perspective of a headhunter, I naturally also hope that the candidate can secure a higher salary, because the higher their annual salary, the higher the headhunter's service fee will be.
However, companies on the other side of the negotiating table are also calculating their own personnel costs. Most companies can accept a 10%-20% increase, but only when the position is urgent enough, the field is new enough, or there are truly few people in the market who can do the job will the budget be further increased.
However, what truly makes people nervous is often not the salary negotiation itself, but the timing.
The same position may be assigned to multiple headhunting firms, and the company's HR department is also building its own list of candidates. Whoever completes a valid referral first usually gets the opportunity. If a person is already in the company's talent pool or has been recommended by another headhunter, the later recruiter will not get credit for the achievement even if they find the candidate. Therefore, before each communication, I will ask: "Have you been in contact with this company before?"
Within our group, we share candidates and do not poach from each other, nor do we solicit candidates from Client B for a position at Client A. However, outside of the group, all headhunters often end up facing the same pool of names.
So, the so-called "AI talent war" that outsiders see may not be as intense as it seems. More often, it's like a quiet race: when others stop at an unresponded resume, I try to find one more connection; when others have gone home after work, I make one more phone call.
Acting one step ahead may mean securing an order, but being one step behind means the person has already been poached by AI and is now in someone else's talent pool, having been fished out from being "hidden underwater".
AI Can Screen Resumes, But Not Hidden Talent
As the competition for talent is often decided by a hair's breadth, AI has naturally become the most convenient accelerator. Tasks that previously required a long time to search and organize can now be done by letting AI run them first.
For instance, after receiving resumes, I will unify them in my own resume database. When a new position becomes available, I first use AI to conduct a round of matching based on the job description, and identify potential suitable candidates.
Nowadays, there are also many Agents on the market that specialize in finding talent for companies. They can search for candidates based on job requirements, and some can even complete the initial contact with candidates through voice conversations. For positions with annual salaries below 1 million yuan and relatively clear requirements, these tools can indeed take over a large amount of repetitive labor.
However, after efficiency has improved, the differences between headhunters have actually become more pronounced.
On one occasion, the same candidate had been contacted by another headhunter beforehand. The other headhunter had simply copied and pasted the job description and company introduction, and quickly completed the referral. When it was my turn to communicate with the candidate, I didn't rush to make a referral. Instead, I first took the time to thoroughly explain the company: its basic situation, its current stage, the team's structure, the backgrounds of the founder and co-founder, the candidate's past experience, and their reasons for leaving their previous position.
The candidate, although he didn't pass the company's initial resume screening, remembered the difference between the two headhunters and thought that the latter's communication was more reliable.
This experience has made it clearer to me that the threshold for headhunters is not about having "AI" on their business cards, nor is it about copying a job description faster than others, but rather about being able to clearly tell a specific person where they are going.
AI can quickly identify a batch of "suitable-looking" candidates, but may not necessarily find the one the company truly wants.
Especially core AI talent, who rarely make their resumes publicly available on recruitment platforms. They are not short of job opportunities and have no shortage of offers, so their resumes are not "floating on the surface". Moreover, these individuals have interacted with numerous founders and CEOs and can even move directly between companies, eliminating the need for headhunters.
AI can screen out 100 suitable resumes from existing materials, but it's difficult to find the person who hasn't left any clues in public channels. Finding these "underwater" individuals relies on talent maps, internal referrals, and relationships accumulated through in-depth communication.
Even if they finally get in touch, the task is only half done.
Top AI talent face a dilemma where the problem isn't a lack of opportunities, but rather too many choices. A company's reputation, background, new business direction, offered salary and benefits, and even the founder's personal charm can all be enticing; however, the stage at which the team is at, the actual scope of the position, and the underlying risks are not necessarily proactively disclosed by the enterprise.
At this point, the headhunter's task is not to simply present another offer, but to break down several opportunities and help the candidate compare them one by one: where the company is currently at, whether the team is a good fit, what the position can offer, and where the choice may lead in the future.
For those who have multiple offers, what they really need is not opportunities, but judgment.
Top talent is spoiled for choice with numerous opportunities, while the majority of job seekers face a different kind of competition: how to get noticed by enterprises in the first place. As a result, some resumes are being repeatedly polished by AI, with project descriptions becoming increasingly elaborate and technical terms more densely packed. At a glance, it seems like every experience is highly relevant to the position.
However, the more impressive the resume, the more headhunters want to continue asking questions.
After getting the call connected, I won't just ask "what have you done", but also follow up on the extent of their involvement in the project, what decisions they made personally, and how they solved problems when they arose. Only by dissecting the details layer by layer can we distinguish between genuine experience and a set of prepared answers for the interview.
I've seen instances where the same person's old and new resumes don't match, with discrepancies in education, age, and project experience, and even a missing company in their work timeline. I've also come across people who simplify their education history, such as changing a part-time or junior college degree to a "bachelor's degree", or omitting a work experience they deem unimpressive. While it's understandable to want to highlight one's strengths, once the formal process begins, failing a background check can result in the loss of an already secured opportunity.
Typically, background checks are outsourced by companies to third-party providers, and sometimes they are directly handled by headhunters. However, whether it's an open or discreet investigation, prior authorization from the candidate is required. After the candidate provides their contacts, the background check often doesn't stop at just that level, and may continue to find other colleagues for secondary or tertiary cross-validation.
Candidates' packaging experience and companies also retain some information, with some companies citing confidential recruitment as a reason to only provide headhunters with a job title, and being unwilling to disclose more background information.
Therefore, headhunters stand in between and need to piece together the unspoken parts as much as possible.
It is here that the boundaries between AI and headhunters are gradually becoming clear.
AI can increasingly handle tasks such as job searches, initial screenings, basic communication, and resume reporting more quickly. However, it cannot guarantee that a person without a public resume will be willing to appear, and it is also difficult to determine whether a project experience is genuine or not. Furthermore, AI cannot replace candidates in weighing their options or help companies identify risks.
AI can screen candidates on a list more quickly, but the value of headhunters lies in finding people outside the list and truly understanding those on it.
Don't Ask Me About Five-Year Plans—AI Changes Faster Than My Career
Someone asked me if I would continue to work as a headhunter in the future.
This sounds like a job interview question. The interviewer sits across from you, waiting to hear a clear career plan: what position you'll reach in three years, where you'll be in five. But in the AI industry, I find it hard to give such an answer.
Because the changes are happening too quickly.
At the end of last year, when I first came into contact with my first AI client, I hadn't yet defined myself as an AI headhunter. By the first half of this year, AI positions had gradually become the focus of my work. Many things hadn't had time to develop into stable methods before new demands emerged.
This uncertainty has forced me to keep learning new things.
I'm not from a technical background, and when I encounter an unfamiliar position, it's impossible to pretend that I understand everything. A more practical approach is to hand over the job requirements and candidate information to AI, first to figure out where the two sides match and where the gaps are, and then to supplement my technical stack and business knowledge based on the results. AI is a bit like a teacher that I can ask questions to at any time, it doesn't mind that my questions are basic, and it can provide explanations quickly.
However, it can only help me push the door open, and I still have to walk the rest of the way by myself.
After AI tools entered the headhunting industry, the gap that was initially widened was not between humans and machines, but between headhunters themselves.
Those who do not use AI at all will likely see their efficiency and income decline; those who only use AI to write reports or screen resumes may see a slight increase in work efficiency, but for the most part, they will only be able to maintain their current level; only those who truly integrate tools into their workflow, participate in AI communities and activities, and continually update their understanding of the industry will likely see significant changes in their performance.
So I don't think AI will completely replace the headhunting industry, it's more like it will re-screen practitioners: repetitive and basic work will decrease, and the advantages that originally relied on information asymmetry and manpower will also become thinner.
However, this does not mean that newcomers have lost their opportunity to enter the field. The traditional growth path for headhunters still exists, with the use of AI and continuous learning becoming fundamental skills that span career development.
People without relevant experience who enter the industry typically start as interns or assistants. At this stage, they do not need to develop clients, but rather take on some positions handed over by their leaders and search for candidates as required. Once they become familiar with the entire process from demand to delivery and achieve certain performance, they may become independent project consultants, directly facing clients and understanding the needs of both enterprises and candidates.
Further up, the job shifts from "finding people on your own" to "leading a team to find people". The number of clients in hand may increase to several dozen, with more positions to fill, and new job openings may emerge every day, or the job requirements may be adjusted. Assistants start to help with delivery. Some people continue to move up and eventually start their own companies; others incubate new teams from their original headhunting companies.
But not everyone can get there.
Many people get stuck at the initial stage. Headhunters have clear KPIs for performance, and if they're unlucky in the first few months, they may not be able to send out a single offer. Sometimes, after a busy day, they can't even recommend a single truly suitable resume. Effort doesn't immediately translate into results, and income won't grow steadily with the time invested.
This process is a test of one's motivation to learn, ability to think, and capacity to endure a period of time with no human connection, no referrals, no interviews, and no performance.
With a bit of luck, and coupled with a quick learning pace and a knack for thinking, it's possible to recommend several decent resumes in a day, and companies are also willing to arrange interviews. However, if results are not forthcoming, some people will choose to leave their jobs voluntarily, while others will be eliminated by the company. After leaving headhunting companies, some people will return to the client side to do high-end recruitment or other human resources work, while others will transition to consulting, and some will leave the industry altogether.
Therefore, when someone asks me if it's still suitable to enter the AI headhunting track now, I won't directly answer "suitable" or "not suitable".
The market is not yet completely saturated, and there is still room for newcomers. However, before entering, it's essential to answer another question: why do you want to become a headhunter?
If one only sees the high salaries and high headhunter fees in the AI industry and thinks it's easier to make quick money, they will likely be disappointed once they start working. This industry requires patience and the ability to accept a large amount of effort that yields no results. Just because a job posting is released, it doesn't mean the right person will be found; even if the right person is found, it doesn't mean they will pass the interview; and even if they pass the interview, it doesn't mean they will ultimately be hired.
I haven't thought about where I'll be in the future. For now, the only thing I can be sure of is to do a good job with what's in front of me.
AI is still advancing, and all I can do is keep up with it.