When AI helps write the work, workers guard their own voice but let their effort quietly vanish from what they hand in.

This explores what happens to the ways workers show how hard they tried and whose thinking a piece of work reflects once AI helps produce it. The short answer from the corpus: workers handle these two signals differently. They protect voice. They let effort quietly vanish. An analysis of 1,250 worker interviews found that people guard identity-bearing cues, like their own voice and where the work came from, but let effort, attention and uncertainty disappear into the finished deliverable Which workplace cues survive AI mediation and which disappear?. The reason is mundane. Most workplaces judge the output, and a finished task is treated as proof that work happened, so nobody asks how much labor went into it.

Once effort disappears, the signal stops meaning anything. A study of Freelancer.com shows what that looks like in a real market. Before AI writing tools, a carefully written proposal was costly to produce, and it cost less for skilled workers, which is why it told employers something real about ability. After the platform added AI writing tools, polished proposals cost almost nothing. Signal quality no longer predicted whether a worker would finish the job, and employers' willingness to pay for workers with strong-looking proposals fell sharply Why did AI tools break the effort signal in hiring?. When everyone can look diligent, looking diligent stops paying.

Workers seem to sense this and manage it on purpose. In four experiments with over 4,400 people, AI users expected colleagues to rate them as less competent and less diligent, and they were less willing to tell managers they had used the tool Do people fear judgment when they use AI at work?. That explains the asymmetry above: hiding effort cues partly protects people from an expected penalty. The workers can be fooled too. The 'LLM fallacy' research finds that when AI-assisted output is smooth and fluent, people start counting it as evidence of their own skill Do AI-assisted outputs fool users about their own skills?. So the effort signal can blur for the people sending it, not only for the people reading it.

Two results suggest that effort was never as easy to read as we assumed. In one interface study, AI-assisted chat cut clicks, page navigation and scrolling, but tasks took no less time Does chat delegation actually save time on task completion?. Visible effort and actual time spent moved independently. Separately, heavy generative AI users shifted their activity toward solo documentation and away from communication and coordination Does generative AI shift knowledge workers away from communication?. That matters for voice, because conversations with colleagues are one of the main places people show their judgment and reasoning live.

What you might not have expected: the real risk isn't that AI erases workers' voices, since people actively defend those. The risk is that effort, the quieter signal, stops being visible. Employers, platforms and even workers themselves then lose a cue they relied on without noticing. The corpus is strong on diagnosing this but thin on fixes. It has little on new ways to make effort or uncertainty visible again, so that remains an open question.