How Do You Do AI Candidate Screening When Every Resume Says AI?

What does AI experience on a resume look like?

Real AI experience on a resume names a system that went into production and a measured result: volume handled, accuracy reached, time or money saved. Everything else is a claim, and claims no longer separate anyone.

In a sample of more than 300 AI Engineering leaders in our network, 96% describe at least one achievement aimed at enabling a new capability. Ninety percent describe one aimed at improving quality or accuracy, and C-suite and VP leaders make both claims at almost identical rates. When nearly every resume on your desk says the person "enabled new AI capabilities" and "improved accuracy," those phrases carry no information. Their presence tells you nothing; only their absence would.

So the screen has to move. Stop asking whether the claim is there and ask whether it carries three things: scale, a baseline and a result. How many transactions, users or decisions did the system touch, what was the number before it existed and what was it after?

One healthcare director in our network shows the standard. They built AI customer engagement and decision support systems handling about 220,000 interactions a day, reaching 75% containment, 94% customer satisfaction and more than $8 million in annual savings. Volume, a quality measure, a customer measure and money, all checkable in a reference call.

The practical rule is simple. Strip the adjectives from an achievement, and if nothing measurable is left, treat it as unproven until the candidate fills it in.

How is AI candidate screening different from keyword filtering?

Keyword filtering checks whether AI terms appear. Screening checks whether the person built and ran something with them, and the two produce very different shortlists because the vocabulary is shared across backgrounds that have little in common.

In our sample of more than 300 AI Engineering leaders, 63% have also worked in data science, while only 7% have worked in ML engineering. ML engineering history is rarer still at the top: 4% of C-suite leaders in the sample have it, against 9% of VPs. Data science and ML engineering share the words (models, training, features, accuracy, pipelines). But one background is mostly analysis and the other is mostly putting models into production and keeping them running.

A keyword filter cannot see that difference. It rewards whoever uses the most terms, which is why relabeling beats it so easily. Anyone who has worked near a model can describe their work in model language, and a filter has no way to ask what shipped.

A good screen looks instead for the operating detail that only comes from running something: tool use, evaluation, failure rates, accuracy on real outputs. The tool names on a line like that are common. How the output was evaluated and the accuracy it reached are what make it credible, and those are the parts a keyword match ignores.

Use keyword filters to clear out resumes with no AI content at all. Do not use them to rank the rest. Credential rules have the same weakness when one rule is applied to every level.

What do employers weight in AI job requirements?

Employers weight credentials differently by level. In the AI job postings we analyzed, 32% of C-suite postings mention a PhD against 10% of director postings, while certifications run the other way: named in 9% of director postings and 1% of C-suite ones. A screen that applies one credential rule to every level drifts away from what the market is asking for.

A degree filters nothing. It is required in 99% of the postings we analyzed and in every VP posting, so screening on "has a degree" removes almost no one and tells you almost nothing.

The signals that matter sit in the level-specific asks. The PhD climbs with level, with VP postings in between at 20%, while certifications nearly disappear at the top.

So do not penalize a director candidate for lacking a PhD, and do not give a C-suite candidate credit for a stack of certifications. If you are assembling a slate, our guide to building a defensible shortlist for AI leadership roles covers how to set the bar per level before the first resume arrives.

How does AI candidate evaluation change from VP to C-suite?

What changes from VP to C-suite is the business outcome attached to the work, not how technical the candidate is. In our sample of more than 300 AI Engineering leaders, hands-on technical building is a main emphasis on both VP and C-suite resumes, so a technical C-suite candidate is normal and not a sign that someone is pitching themselves too senior.

One C-suite technology leader at an IT services firm in our network implemented a vector database for semantic search, indexing multimodal video in real time with query latency under 50 milliseconds. That is deep, hands-on work from someone at the top of the org chart, and in this sample it is not unusual. One caveat: our C-suite group includes founders and chiefs of small companies, which pulls hands-on work up at that level.

Do not screen a C-suite candidate out for being technical. Screen for whether the technical work is tied to an outcome the business felt. The levels separate a little here: 80% of C-suite leaders in the sample describe an achievement aimed at customer or employee experience, against 74% of VPs. The gap is small, but it points toward the outcome question.

At VP level, ask what they built and how well it works. At C-suite level, ask the same, then ask who noticed and what changed for them. If the candidate cannot answer the second question, the technical depth is not doing the job you are hiring for.

What are the AI resume red flags?

The main red flags are an outcome with no number, a tool list with no system that shipped and a certification offered in place of deployed work. Each one is a claim standing where evidence should be.

Take the unquantified outcome first, and the revenue claim in particular. In our sample of more than 300 AI Engineering leaders, 63% of leaders describe at least one achievement aimed at growing revenue, with C-suite and VP leaders close together. A revenue claim is easy to write and hard to check, so it needs a before and after and a named system behind it. "Drove revenue growth through AI" is a red flag; a specific product, a specific lift and a specific period is evidence.

One VP in financial services in our network built an LLM recommendation engine that automated dispute reviews, cutting cycle time from over 50 days to 23-25 days and lifting profitability from 72% to 84%. Named system, baseline, result and a business measure. Nothing on that line needs to be taken on trust.

The tool list is the second flag. A resume that names a dozen frameworks and platforms but no system that went live has described an environment, not an accomplishment, so ask which of those tools ran in production and what they produced.

The third flag is a credential doing the work a deployment should. In the sample, 1% or fewer hold the AIGP governance certification, while the PMP, a project management credential, is held by 10% of C-suite leaders and 7% of VPs. Almost none of the leaders in the sample hold an AI-specific certification, so a credential cannot stand in for shipped work. All three flags come down to evidence on paper, and the last screen tests the one thing paper cannot.

How should non-technical executives approach screening AI candidates?

Ask the candidate to explain one system they shipped in plain terms: what it did, how they measured it and what broke. In the AI job postings we analyzed, explaining AI plainly is a main emphasis at director, VP and C-suite level, and in our sample of more than 300 AI Engineering leaders, resumes stress it just as heavily. When employers and candidates agree that completely, the paper cannot settle the question. A short call can.

That works in favor of a CEO, COO or CHRO running the screen. You do not need to judge the model, only the explanation, and that is a skill you already have. Is it plain and specific? Is it tied to a measured result, and can the person describe what went wrong without getting defensive or vague?

The wider labor market supports this approach. Only 6% of AI workers have AI degrees, and only 2 of the top 10 AI job skills are AI-related; the other 8 are human skills like communication and leadership.¹ Communication is part of the job, so it belongs in the screen.

Structure the call around one achievement from the resume and ask the candidate to walk through it as if you were a new board member. Then ask what the number was before, who used the system and what they would build differently. Our list of AI interview questions that separate capability from hype gives you follow-ups for each of those.

Screening works when it asks for proof the resume cannot fake

When every resume says AI, the word stops meaning anything. Nearly every leader in our sample describes new capabilities and better accuracy, model vocabulary sits on top of very different backgrounds and the credential rules employers use change sharply from director to C-suite.

The screen that works has three parts. Look for a shipped system with scale, a baseline and a measured result. Apply credential expectations by level, not as a blanket rule, then put the candidate on a short call and have them explain one system in plain language.

None of this needs a technical interviewer at the first stage. It needs a screener who refuses to accept a claim without its number. If you would rather hand that work to a search team, see how our AI executive search practice runs it.