When AI makes every cover letter sound great, do reviews and track records replace them, or do the strongest candidates get lost?

This explores whether track records (ratings, reviews, work history) step in when written applications stop being informative. The corpus has a clear answer from one marketplace: employers do switch, but the switch is only partial. On Freelancer.com, an AI tool made tailored cover letters nearly free to write. After that, letter quality became much weaker at predicting who got interviews and offers, and employers leaned more on work history and reputation instead Does AI-generated cover letter access weaken hiring signals?. So the substitution is real. Employers stop trusting a signal once anyone can produce it, and they look for one that's harder to fake.

The records don't fully replace what the writing did, though. A simulation of the same market with written signals removed found that the most able workers were hired 19% less often and the least able 14% more often Does cheap writing weaken hiring based on worker ability?. A good cover letter used to be costly to write, and that cost was the information: it showed who was willing and able to put in effort. Reviews and history apparently don't capture all of that. Some of the old signal also survives. Workers who spent more time editing their AI drafts were hired more often, even though most people submitted drafts with little revision Does editing time on AI drafts predict hiring success?. The effort signal didn't disappear; it moved from writing the letter to editing it.

The pressure behind the switch builds up from several directions. Recruiters report being flooded with applications, and applicants report using prompt injections to get past filters, which looks like an arms race between applicants and employers Are job applicants and employers locked in an escalating AI arms race?. When an AI does the screening, writing gets distorted further. LLM evaluators favor resumes written in their own style and shortlist applicants who used the same model 23 to 60% more often Do LLM evaluators favor resumes written by their own model?, Do language models favor resumes they rewrote themselves?. Human evaluators can also mistake polish for merit Does polished writing actually signal better quality work?. In graduate admissions the reaction runs the other way: officers often spot AI essays and rate them lower, and AI users were admitted less often despite writing better essays Do admissions officers penalize essays they suspect are AI-written?, Does AI essay use hurt admissions chances despite quality gains?. In both settings, the writing has stopped meaning what it used to.

A paper from a different area explains why records are the natural place to retreat to. It proves that a checker reading only generated text can't reliably judge quality when the truth depends on facts outside that text; a checker that can see those outside facts succeeds Can transcript alone tell whether a reflection helps?. A job application is that kind of text, and a track record is that outside evidence. There's a warning too: records are only safe while they stay grounded in real history. LLM judges are easily fooled by fake references and authoritative formatting Can LLM judges be fooled by fake credentials and formatting?, so a record that is only described in the application, rather than checked against the source, can be gamed like the letter.

Where the corpus is thin: the substitution evidence comes almost entirely from one freelance platform. Nothing here measures who loses when reputation replaces writing. The obvious candidates are newcomers with no reviews yet, but that remains an open question rather than a finding.