Picture a small business owner in Nashville, or Noida, who needs a logo for her new bakery. She posts the job on a freelancing platform before her morning coffee. By lunchtime, forty proposals have landed in her inbox. Each one is beautifully written. Each one says the freelancer is “passionate about bringing your vision to life.” Each one promises “attention to detail” and “timely delivery.” She reads ten of them and realises, with growing despair, that she cannot tell a single applicant apart. It is a bit like speed dating where everyone has rehearsed the same script from the same self-help book.
This little scene sits at the heart of a study presented at the International Conference on Information Systems (ICIS) 2025 by Jie Ren of the University of Melbourne and Li Ding and Jiayu Yao of Nanyang Technological University. Their question is deceptively simple: when freelancers use generative AI to write their job proposals, does it help employers find the right person, or does it make everyone look the same?
Why matching matters more than we think
Online labour platforms such as Upwork and Fiverr are, at their core, matchmakers. Their job is not to do the work themselves but to connect someone who needs a task done with someone who can do it well. Economists call this “matching efficiency.” A good match is one where the employer gets quality work, the worker gets fair pay and recognition, and neither of them would have been better off choosing someone else. When platforms match well, more people join on both sides, and the whole marketplace thrives. When they match badly, employers get burned, good workers get overlooked, and trust leaks away. The proposal is the first and arguably most important step in this dance. Before an employer looks at ratings or prices, she reads the pitch. A good proposal shows that the freelancer actually understood the job, has the right skills, and has a plan. In the language of economics, it is a “signal.” The effort and thought that go into a customised proposal tell the employer something real about the person behind it.
Now enter generative AI. Tools that write polished proposals in seconds are spreading across these platforms at remarkable speed. In fact, the authors note that generative AI has been adopted faster than personal computers or the internet were in their early years. A freelancer who struggles with English, or who simply hates writing about themselves, can now produce something that reads like it came from a professional copywriter.
The double-edged sword
Here is where the AI literature in information systems becomes genuinely interesting. The literature argue that AI-assisted proposals could push matching in two opposite directions at once. On the bright side, hiring has always suffered from what economists call information asymmetry. The worker knows how good she is; the employer does not. Many talented people are poor at describing their own abilities. A brilliant coder in a small town may write clumsy English, and lose out to a smooth talker with mediocre skills. If AI helps such workers explain their qualifications clearly, employers can judge actual ability more accurately. In this rosy scenario, AI acts like a translator that lets real talent shine through. On the darker side, if everyone uses the same tool, everyone starts to sound alike. The unique quirks that once revealed a worker’s personality, effort, and genuine fit begin to disappear under a layer of generic polish. The signal gets weaker. When every proposal is excellent, excellence stops telling you anything. The employer is back to square one, possibly worse off, because now the bad candidates look just as impressive as the good ones on paper. Think of it as a job market where everyone shows up wearing the same perfectly tailored suit. Lovely to look at, useless for telling people apart.
How the researchers plan to find out
Because this is work in progress, the literature lays out a carefully designed experiment rather than final results. The setup is clever. The researchers will recruit people to play the roles of freelancers and employers in a controlled lab setting. Some freelancers will write their proposals with access to generative AI, while others will write them on their own. Employers will then read the proposals and pick the candidates they believe will do the best job, without knowing who used AI. The twist comes next. The chosen freelancers will actually complete the task, and employers will evaluate the finished work. This lets the researchers check whether the person who looked best on paper turned out to be the person who actually delivered the best work. If the two line up, matching is efficient. If the shiny proposal belongs to someone whose real work disappoints, the signal has failed.
The study also adds real money to keep everyone honest. Freelancers earn bonuses for being selected and for producing top-quality work, and employers earn bonuses for correctly spotting the strongest candidates. The researchers will look separately at creative jobs and routine jobs, since AI might blur differences far more in one than the other. They will also track whether AI concentrates hiring on a few people or spreads opportunity more widely.
What it means for the rest of us Even before the results arrive, the study raises questions that matter well beyond freelancing. Anyone who has reviewed job applications, grant proposals, or college essays in the last two years has probably felt the same fatigue as our bakery owner. When AI can make everyone sound competent, the way we evaluate people has to change.
The researchers suggest several practical responses. Employers could design proposal formats that ask for concrete, verifiable details rather than open-ended prose, making it harder for generic text to pass as genuine insight. Platforms could invest in training so that workers learn to use AI to sharpen their own voice rather than replace it, and could build tools that strip out repetitive filler so reviewers see what is truly distinctive. Policymakers, meanwhile, should watch for fairness. AI could level the playing field for people who were held back by language barriers, but it could also widen gaps if only some workers know how to use it well. There is a lesson here for students and professionals too. If a machine can write your cover letter, then the cover letter is no longer where you stand out. Your portfolio, your track record, and the specific, honest details only you can provide become your real competitive edge. AI can polish the window, but employers still want to see what is inside the shop.
In the end, this study reminds us that technology rarely solves a problem without creating a new one. Generative AI may well make the freelance marketplace a better matchmaker. Or it may turn it into a room full of identical, eloquent strangers. Thanks to careful experiments like this one, we may soon know which.
Reference
Ren, J., Ding, L., & Yao, J. (2025). A better matchmaker? The impact of GenAI on matching efficiency in online labor markets. In ICIS 2025 Proceedings (Paper 42). Association for Information Systems.
