Keyword screening was always a proxy. It never measured whether a candidate could do the work. It measured whether the language on the resume overlapped with the language in the job description, and it assumed that overlap correlated with capability. For a long time that assumption was weak but usable, because most candidates wrote their own resumes in their own words and the overlap carried at least some signal.
That assumption has now collapsed. When a large share of your applicant pool runs the job description through an AI tool and asks it to rewrite their resume to match, the overlap stops being a signal. Every qualified applicant and every unqualified applicant can produce a resume that mirrors your posting almost perfectly. The filter still ranks them. It is just no longer ranking on anything real.
Read that number against how a keyword filter works and the problem is obvious. The screening method that is supposed to narrow the field is being optimized against by most of the field, using tools that are free and take a few minutes to use.
What the Filter Now Selects For
A keyword filter under these conditions does not select for the most capable candidate. It selects for the candidate who did the best job of pointing an AI tool at the job description. Those are not the same person, and they are often not even correlated.
The strongest practitioner in your pool may submit an honest resume that describes real work in their own plain language and never mentions three exact phrases from your posting, because that is not how they talk about their job. The filter scores them below an applicant who has never done the work but generated a flawless keyword echo of the requirements. The screening step is now actively inverting the result you wanted.
Keyword matching used to reward resume writers over qualified candidates. AI did not fix that. It industrialized it. The advantage no longer goes to the person who learned to write for the algorithm. It goes to anyone willing to paste a job description into a chatbot.
Why "Detect the AI" Is Not the Answer
The instinct is to fight back by detecting and rejecting AI-written resumes. This does not work, for two reasons.
First, detection is unreliable. AI-writing detectors produce false positives at rates no responsible hiring process should accept, and rejecting real candidates because a tool guessed wrong is its own legal and reputational exposure.
Second, and more important, an AI-assisted resume is not evidence of a bad candidate. A genuinely strong professional who used a tool to tighten their bullet points is still a genuinely strong professional. Penalizing the use of a writing tool screens on a signal that has nothing to do with whether the person can do the job. You would be replacing one broken proxy with another.
The Real Fix Is to Stop Screening on Words
If the problem is that everyone can now manufacture the right words, then the only durable answer is to evaluate something words alone cannot fake: what the experience actually demonstrates.
A resume that claims fluency in a skill but describes no work that would require it reads very differently from a resume where the same skill is implied by the scope, the progression, and the responsibility described. A keyword filter cannot tell those two apart, because both contain the keyword. An evaluation that reads the full document for demonstrated capability can, because the evidence either supports the claim or it does not.
This is the line that separates screening that survives AI from screening that does not. Counting or matching terms is trivially gameable and now routinely gamed. Assessing whether the described experience supports the claimed capability is not, because the candidate would have to fabricate a coherent career history rather than a phrase, which is a far harder and far more detectable thing to do.
Practical step: Pull ten resumes your system ranked at the top this month and ask one question of each: does the work history described actually demonstrate the skills being credited, or does the resume simply contain the words? If you cannot tell the difference from the ranking, your screening is measuring vocabulary, not fit.
Where This Leaves Employers
The era when a keyword threshold was a defensible first filter is over. It was always a proxy, and the proxy has been commoditized. Continuing to rank candidates on language overlap now means ranking them on who used AI most effectively, while the people who can actually do the work sit unread below them.
This is the problem TrueScan HR was built to solve. It reads the full resume the way an experienced hiring manager would and evaluates whether the described experience supports the capability being claimed, rather than whether the right words appear. When everyone can generate the words, the only thing left worth measuring is whether the work behind them is real.
Thabiti Adams is a CISSP and CCSP certified cybersecurity professional and founder of Adams Cloud & Cybersecurity.