Did ChatGPT actually shrink freelancers' paychecks, or did it just change what work looked like?

This explores whether ChatGPT measurably cut freelancers' work and pay, or whether that's a story that sounds right but lacks evidence. For online freelancers, the corpus says yes. A study of Upwork compared occupations heavily exposed to generative AI with less-exposed ones, before and after ChatGPT's November 2022 release. In the exposed occupations, freelancers landed about 2% fewer jobs and earned 5.2% less, and writing work took the biggest early hit Did ChatGPT's release reduce freelance writing work and pay?. One caveat about wording: the study measures jobs won and earnings, not hours worked. The 'hours' part of the question is never measured directly for freelancers here.

The surprising part is who got hurt. You might expect strong reviews and a long track record to protect a freelancer. They didn't. The same Upwork data found that past performance offered no protection, and the top freelancers may have been hit harder Does a strong track record protect freelancers from AI?. This cuts against a popular finding from lab experiments, where AI mostly helps weaker workers catch up. In a real market, clients don't need to hire the best writer for a task ChatGPT now handles well enough. The top of the market loses the most because it had the most to lose.

The picture changes once you leave gig platforms. Danish administrative records follow regular employees whose employers rolled out chatbots. Two years after ChatGPT's launch, their earnings and hours stayed flat within a 2% margin, even though they had clearly taken on new AI-related tasks Does AI chatbot adoption change worker pay and hours?. A good way to reconcile the two results: freelancers sell finished pieces of work one at a time, so a client can swap them for a chatbot overnight. Employees sit inside firms that reorganize their tasks first, and any change in pay comes later, if it comes at all.

Why writing first? Other studies point to the same weak spots. When researchers sorted real ChatGPT conversations, writing turned out to be the main work use, more than looking up information How is ChatGPT actually being used by real people?. And the same kind of loss shows up beyond paid work. Stack Overflow posts fell 16% after ChatGPT compared with similar sites, reaching about 25% within six months, with the steepest drops in programming languages ChatGPT knew best Did ChatGPT cause Stack Overflow posting to decline?. The pattern is consistent: ChatGPT replaces the work where it's strongest, and the people who did that work as one-off tasks feel it first.

So the answer: yes for writing-heavy freelancing on platforms, with modest but real drops in jobs and earnings. For employees, no wage or hours effect has shown up yet. The corpus can't tell you whether the employee figures are a lasting null result or just a delay before effects arrive.