Automated filtering of applications has existed for years, matching keywords. Language models changed it from matching to interpreting, and the practical effects are worth knowing from both sides.
What it does now
Reads applications and summarises them against the role requirements. Ranks candidates. Sometimes conducts a first-round interview through video or chat and produces an assessment.
The stated benefit is handling volume, which for remote roles is genuinely enormous – hundreds of applications per listing is normal.
What candidates should know
Keyword stuffing is obsolete and now counterproductive, since a model reading for meaning notices text that does not make sense.
What helps is being specific and concrete. A model summarising your application produces a better summary if the source contains actual achievements with actual numbers.
Standard formats still help. Unusual layouts, images and tables can be parsed badly by whatever extracts the text.
The bias problem
Models trained on past hiring decisions learn the patterns in those decisions, including the discriminatory ones. This has been demonstrated repeatedly.
Regulation is arriving – several jurisdictions now require disclosure and auditing of automated hiring tools – and enforcement is early.
Your rights
In some jurisdictions you can ask whether an automated system was involved in a decision and request human review. Whether that is useful in practice varies.
The arms race
Candidates use AI to write applications, employers use AI to read them. The result is more volume on both sides and no obvious improvement in matching, which is where this currently sits.