Ever since AI entered hiring processes, one question has dominated candidates' minds: "What criteria is this machine using to screen me out?" Lacking technical knowledge, candidates develop their own intuitive assumptions about how the system works. Known in research as "Folk Theories," this phenomenon leads candidates to view AI as a black box — and to devise sometimes creative, sometimes completely misguided strategies to crack it open. Recent studies reveal striking data on how candidates try to "hack" AI and how these theories affect hiring quality.

Folk theories: Viewing AI as a "superpower"

Even though candidates don't fully understand how AI makes decisions, they construct elaborate mental models about it. The research titled "Experience and Adaptation in AI-mediated Hiring Systems" by Md Nazmus Sakib and colleagues (2018/2024) finds that individuals rely on heuristic reasoning — "folk theories" — to explain AI behavior. According to this study, people variously describe AI as a simple tool, an assistant, or a mysterious "superpower" whose actions are unpredictable.

These theories directly shape how candidates behave during interviews. For instance, when candidates realize the system isn't as advanced as advertised (an expectation violation), they begin speculating about how decisions are made — and this uncertainty significantly raises stress levels. Sakib and team (2018/2024) emphasize that when candidates can't find clear guidance, they generate their own theories, which creates emotional tension.

Strategic manipulation: Attempts to game the algorithm

When candidates believe they've cracked the AI's "ideal employee" profile, they start tailoring their answers accordingly. The field experiment titled "Behavioral Measures Improve AI Hiring" by Marie-Pierre Dargnies and colleagues (2025) proves that candidates strategically modify their answers based on what they think the company expects. However, these predictions don't always turn out to be accurate.

One of the study's most interesting findings concerns the "patience" variable. According to Dargnies and team's (2025) data, candidates actually present themselves as more "impatient" than they really are — because they assume impatience will be read as a sign of high motivation and ambition. In reality, the algorithm scores patient candidates higher for long-term productivity. Similarly, candidates try to impress the AI by underreporting their neuroticism scores and significantly inflating their locus of control scores.

Performance art: Interviewing against silence

One of the most prevalent folk theories concerns how AI scores body language and accent. Many candidates believe the AI will code even the slightest eye movement as "negative." In their research, Sakib and colleagues (2018/2024) note that candidates describe this experience as "performing to silence."

This belief drives the following behaviors:

• Accent masking: Research data shows that non-native speakers suppress their natural accents and speak in a robotic tone to avoid being misunderstood by the AI.
• Forced smiling: Candidates assume the system measures "positivity" and try to hold an artificial smile throughout the interview.
• Keyword hunting: Nicole Jurado's (2025) study reports that candidates stuff their resumes and interview answers with keywords they believe the system "likes."

Candidates say these performative actions are emotionally draining and make them feel "dehumanized." Sakib and team (2018/2024) highlight that candidates view these processes more like a "hackathon" than an interview.

How transparency eliminates folk theories

The only way to prevent candidates from building strategies on false assumptions is to establish a transparent communication framework. Research by Poenaru and Diaconescu (2025) shows that transparent and fair AI usage increases candidates' trust in the technology. When candidates know what the system evaluates (for example: "we only analyze word content, not facial expressions"), they stop feeling like they're playing a game and focus on their actual performance.

Md Nazmus Sakib and colleagues (2018/2024) recommend the following design interventions to reduce candidate stress:

• Transcript editing: Showing candidates how the AI heard them and giving them the right to correct errors.
• Real-time feedback: AI responses like "I understand, that was a great example" eliminate the feeling of talking to a wall.
• Guidance: Providing a short informational video before the interview explaining how the system works.

The "folk theories" candidates develop are ultimately a cry of uncertainty and distrust. As long as companies use AI as a "screening shield," candidates will keep looking for ways to pierce it.

The successful hiring systems of the future won't be the ones that force candidates to guess — they'll be the ones that explain the process with full transparency and offer a human-centered interaction. As emphasized in the research by Brian Jabarian and Luca Henkel (2026), when the balance between AI and human involvement is struck correctly, job offer rates increase. The best candidates aren't those who game algorithms, but those who fearlessly showcase their true potential in a transparent environment.

References

• Dargnies, M. P., Hakimov, R., & Kübler, D. (2025). Behavioral Measures Improve AI Hiring: A Field Experiment. Discussion Paper No. 532, CRC TRR 190.
• Jabarian, B., & Henkel, L. (2026). Voice AI in Firms: A Natural Field Experiment on Automated Job Interviews. Booth School of Business, University of Chicago.
• Jurado, N. (2025). The effects of artificial intelligence on shaping employer brand perception: insights from entry-level hiring practices. Master Thesis, Universidad Carlos III de Madrid.
• Poenaru, L. F., & Diaconescu, V. (2025). Bridging Technology and Talent: Gen Z's Take on AI in Recruiting and Hiring. Bucharest University of Economic Studies.
• Sakib, M. N., Rayasam, N. M., & Dey, S. (2018/2024). Experience and Adaptation in AI-mediated Hiring Systems: A Combined Analysis of Online Discourse and Interface Design. University of Maryland.
• Gartner. (2026). Gartner Survey Shows Just 26% of Job Applicants Trust AI Will Fairly Evaluate Them.
• Chopra, F., & Haaland, I. (2024). Conducting Qualitative Interviews with AI. CESifo Working Papers, No. 10666.