How to screen candidates when every resume is AI-written

A few years ago, a sloppy resume told you something. A missing skill, a vague job title, an obvious typo: these were small signals about how someone worked. Today most of those signals are gone. Candidates paste the job description into a chatbot, ask it to tailor their resume, and send back something clean, keyword-matched, and confident.

That's not cheating. It's the smart move, and good candidates are doing it too. But it changes what you're actually reading. When every resume is polished to the same shine, the polish stops being information. The job now is to screen for signal that survives the rewrite, and to stop rewarding the things a language model can fake in thirty seconds.

What AI writing actually changes

It helps to be precise about what's different. AI-assisted resumes don't invent entire careers for most people. What they do is smooth over the gap between what someone did and how well they describe it. That gap used to be a filter. Now it's closed for nearly everyone.

• Keyword matching is now trivial. Anyone can mirror your job posting's exact language, so a resume that hits every term you listed tells you very little.
• Writing quality is no longer a proxy for the candidate. A crisp bullet point might reflect the person, or it might reflect the tool. You can't tell from the page.
• Formatting and structure are commoditized. Clean layouts are one prompt away. A tidy resume is table stakes, not a differentiator.
• Vague accomplishments get dressed up. "Improved team efficiency" becomes "drove a 30% lift in operational throughput," and the number is often unverified.

None of this means resumes are worthless. It means you have to read them differently, weighting the parts that are hard to fabricate and discounting the parts that aren't.

Screen for specifics, not adjectives

The most reliable signal in an AI-heavy pile is concrete, checkable detail. A model can generate fluent language, but it can't invent a candidate's real project without the candidate feeding it real facts. So look for the facts.

• Named tools, systems, and technologies used in context, not just listed in a skills bar.
• Specific scope: team size, budget, number of customers, region, product area.
• Decisions the candidate made, and what they'd have done differently.
• Numbers that come with a denominator ("grew from 200 to 900 accounts" beats "grew accounts 350%").
• Timelines that make sense against the rest of the resume.

A resume heavy on adjectives ("results-driven," "passionate," "strategic") and light on nouns is a weak signal regardless of how well it's written. The reverse, a plain resume full of specific nouns, is usually the stronger candidate.

Move the real screen into the conversation

If the resume can't be trusted to separate candidates, the interview has to carry more of the load. That's not a burden, it's a correction. Resumes were always a rough proxy. The fix is to get to a structured, skill-based conversation faster, on more candidates, earlier in the process.

Practical ways to do that:

• Ask candidates to walk through one real project in detail. People who did the work can go five layers deep. People who described the work can't. This is the single most reliable way to tell them apart.
• Use a short, role-relevant work sample. Not a take-home that eats a weekend, but a focused task that mirrors the actual job. Real skill shows up under real constraints.
• Follow up on any number they claimed. "How did you measure that?" is a fair question, and the answer is diagnostic.
• Score against a rubric, not a gut feeling. When resumes converge, your interview structure is what keeps decisions honest.

Don't punish candidates for using the tools

It's tempting to try to detect AI-written resumes and screen them out. Resist that. AI writing detectors are unreliable, they produce false positives, and they tend to flag non-native English speakers and people who write plainly. You would be filtering for the wrong thing and quietly introducing bias while you do it.

A candidate who used a tool to write a clearer resume is not a worse hire. In many roles, using available tools well is exactly the skill you want. The goal is not to catch people using AI. The goal is to stop treating resume polish as evidence and to test the underlying ability directly.

Where AI screening helps, and where it doesn't

There's a useful irony here: AI is a good way to handle a pile of AI-written resumes, as long as you're clear about what it's doing. Aster reads every resume as it arrives, pulls out structured skills and experience, and ranks applicants against the specific role with the reasons attached. That does two helpful things in an AI-heavy world.

• It normalizes the pile. Instead of being swayed by which candidate wrote the slickest bullet points, you get a consistent read of what each person actually has against what the role needs.
• It surfaces the reasoning, not just a number. A match score with its reasons lets you sanity-check the substance instead of trusting the presentation.

What it can't do is verify that a claim is true. No screening tool, human or AI, knows whether "led a team of eight" really happened. That verification lives in references, work samples, and the interview. The right division of labor is clear: let AI compress the pile and rank on structured signal so you get to a shortlist in an afternoon, then spend your human time verifying the specifics on the people who made the cut.

A simple screening checklist

If you want a starting point for the AI-resume era, screen roughly in this order:

• Substance over style. Reward specific, checkable detail. Discount adjectives and keyword mirroring.
• Consistency. Do the timelines, scope, and claims hang together across the whole resume?
• Role fit on the things that matter. Rank against the actual requirements, not against who wrote the best prose.
• Verify in conversation. Get strong-looking candidates into a structured, skills-first interview quickly, and follow up on every number.
• Ignore the AI question entirely. Whether they used a tool is not your concern. Whether they can do the job is.

The shift is real, but it's not bad news. When polish stops carrying information, you're forced back to the things that actually predict performance: concrete evidence and a fair, structured test of skill. That was always the better way to hire. The AI-written resume just made it non-optional.