The people building the AI don't trust it to hire people like them.
The Summary
- Google's own AI research team is warning job applicants that the company's HR AI filters are unreliable — the same tools Google sells to enterprises for recruitment
- The irony: Google markets these AI screening tools as efficiency multipliers for sorting candidates at scale
- Signal: When the builders won't eat their own dog food, pay attention
The Signal
Google's AI researchers are telling job seekers not to trust the automated resume filters their employer sells as enterprise software. This isn't a leak or a whistleblower situation. This is the AI team publicly acknowledging that algorithmic screening misses good candidates and rewards the wrong signals.
The context matters here. Google positions its AI hiring tools as solutions to the "resume problem" — the thousands of applications that pour in for every open role. The pitch to HR departments is simple: let the machine handle the first cut. Save your recruiters for the finalists. Scale your hiring without scaling your team.
"The people who understand how the sausage gets made won't eat the sausage."
But here's what that pitch leaves out. AI resume screeners optimize for patterns in past hires. If your company historically hired people with certain degrees, from certain schools, with certain keywords on their resumes, the AI learns to filter for exactly those things. It doesn't screen for talent. It screens for conformity to historical accident.
The AI team's warning suggests they know this. They know the tools reward resume formatting over problem-solving ability. They know the filters catch people who are good at SEO-optimizing their career history, not necessarily people who are good at building things. And they know that if you're trying to hire researchers, engineers, or anyone doing novel work, filtering by pattern-matching to the past is how you miss the future.
Key problems with AI hiring filters:
- They optimize for keywords and formatting, not capability
- They encode historical bias as "efficiency"
- They penalize unconventional backgrounds — exactly the profiles that drive innovation
This creates a split-screen moment for Google. Externally: "Our AI makes hiring better." Internally: "Don't let our AI filter you out." That gap between the sales deck and the engineering team's actual beliefs is the story. It's not that the technology doesn't work. It's that it works exactly as designed, and the design has a flaw baked in. It selects for sameness.
The Implication
If you're applying to jobs at companies using AI screening, assume the filter is broken in predictable ways. Use the keywords. Format conventionally. Game the system, because the system is gameable by design. But if you're hiring, especially for roles where you need people who think differently, don't outsource judgment to a pattern-matcher. The best candidates for hard problems often have resumes that look nothing like your last ten hires. That's not a bug in their background. That's the signal you're filtering out.