Human-written CVs outperform AI-written ones by a factor of three to one, according to new research examining AI-written CV performance across a series of sales and marketing roles, raising questions about the growing trend of using artificial intelligence to automate job applications.
Job resource site Kickresume set up a controlled experiment comparing AI-written CVs against those written entirely by people, sending out 200 applications in total. The human-written CV was the most successful, receiving a positive response for 31% of applications. AI-only CVs were the worst performers, with just 10% receiving a positive initial response.
Tailoring Closes the Gap in AI-Written CV Performance
The results were not entirely damning for AI. When applicants tailored AI-written CVs to fit the specific job description, the response rate rose to 18%, nearly doubling the untailored AI figure.
Peter Duris, CEO of Kickresume, said: ‘While the strongest CV was human-written, a little tailoring of your AI-written CV to fit the job description can boost its performance significantly. In our test, tailoring nearly doubled the AI resume’s response rate.’
Duris added: ‘Pulling in the language from the posting shows recruiters and hiring managers you’re a genuine fit for the role and helps your CV read cleanly to an applicant tracking system along the way.’
The experiment focused on a single sector and did not include additional materials such as cover letters, meaning results may differ across other employment areas. The core finding, however, is clear: unmodified, automated applications fared worst of all.
Wider Data Points to a Nuanced Picture
The experiment sits alongside broader data from Kickresume’s own usage figures. According to Kickresume’s analysis of anonymised internal data, the company examined usage from over 1.2 million people who used AI-powered features in 2025. Of those, around 43,000, approximately 3.5%, used AI specifically to tailor their CVs to a particular job description rather than generating a document wholesale.
That figure underlines how rare targeted AI use remains, even as the broader adoption of AI tools in job searches accelerates. The majority of users, it appears, are not applying the kind of human editorial judgement that Kickresume’s experiment suggests makes the difference.
Separate research from MIT Sloan offers a more favourable view of algorithmic assistance. Applicants who received AI-assisted support with their CVs secured 7.8% more job offers and were more likely to be hired in their first month on the platform than an unassisted control group. The same research found that applicants with more than 99% of the words in their CV spelled correctly were hired nearly three times more often in their first month than those with fewer than 90% spelled correctly, pointing to accuracy, however achieved, as a key factor.
The two sets of findings are not necessarily contradictory. Kickresume’s experiment tested fully automated, untailored AI output against human-written applications, while the MIT Sloan data looked at AI as an assistive tool rather than a wholesale replacement for human input.
The practical takeaway from Kickresume’s research is that mass-sending identical, AI-generated CVs without review is the least effective strategy available to jobseekers. Recruiters and employers, the data suggests, respond to differentiation, and a jobseeker’s own knowledge of their strengths and experience is not something an AI model can fully replicate without guidance.
‘It was really interesting carrying out this experiment, and we hope it inspires people to keep pushing with their job search and experimenting with their own CVs,’ Duris said.
Kickresume has not indicated when further research into AI-written CV performance across other sectors or application types will follow.
