Once AI made polished job proposals cheap, did a freelancer's effort still prove they were good at the work?

This explores why a well-written job proposal used to tell employers something real about a freelancer's ability, and what happened once AI made polished writing cheap. The short answer: ability and effort cost ran in opposite directions. On Freelancer.com, a carefully tailored proposal took real effort, and that effort cost less for more capable workers. Because strong writing was cheaper for them to produce, employers could read the proposal as a rough measure of the person Why did AI tools break the effort signal in hiring?. This is the classic logic of costly signaling. A signal is worth something only when it is harder to fake for the people it is meant to screen out.

AI writing tools broke that relationship. In proposals written with the platform's built-in AI tools, the effort a worker put in no longer tracked the quality of the signal. The relationship even flipped, so the polish of a proposal stopped predicting whether the worker would finish the job. Employers seem to have noticed. Their willingness to pay more for workers who sent strong signals dropped sharply after the tools arrived Why did AI tools break the effort signal in hiring?. A related simulation estimates the cost of losing the signal. Without written signals, hiring becomes about 19% less merit-based: the strongest fifth of workers get hired 19% less often, and the weakest fifth get hired 14% more often Does cheap writing weaken hiring based on worker ability?.

This offers one possible explanation for a separate Upwork result that is otherwise puzzling. A strong track record did not protect freelancers from ChatGPT's harm, and top performers may have been hit hardest Does a strong track record protect freelancers from AI?. That finding cuts against experiments in which AI mostly helps weaker workers catch up. It is consistent with signal collapse: if skilled workers' edge partly came from being able to show their ability cheaply, flattening the signal takes that edge away. The corpus does not directly test this link, though, so treat it as a hypothesis worth following up.

The question also points somewhere less obvious: where might a new costly signal come from? A recruiter study gives a warning. Listing AI skills raised interview invitations by 8 to 15 percentage points, yet certificates added little over simply claiming the skill. In other words, recruiters are rewarding a claim that costs almost nothing to make and that they rarely check Do AI skills help candidates get more job interviews?. A more promising lead is process data. When someone hands writing or coding work over to AI wholesale, the AI's contributions arrive in sudden bursts that break the author's normal rhythm, which leaves a recognizable signature. Ordinary back-and-forth use of AI does not leave that signature Can process data distinguish AI delegation from ordinary collaboration?. In principle, how a proposal was written could become the new signal, even if the finished text no longer is.

One caution: almost all of the corpus's material on this question comes from a single study of Freelancer.com. The effort-cost mechanism is well described there, but the collection has nothing yet on whether the pattern holds in other labor markets or on what new signals employers actually adopt.