"If You're Very Clever, No One Knows You've Used It": The Social Dynamics of Developing Generative AI Literacy in the Workplace

Based on interviews with 19 knowledge workers, this paper reveals that while workplace social dynamics support GenAI learning, the perception that concealing AI use validates expertise leads to reduced knowledge sharing and transparency, prompting a call for more open dialogue and collaborative learning to advance AI literacy.

The Gist

The Big Picture: The "Secret Sauce" of the Office

Imagine a new, magical kitchen gadget has appeared in every office. It can write emails, fix code, and draft reports in seconds. This is Generative AI (GenAI).

The researchers wanted to know: How do office workers actually learn to use this gadget? Do they take a class? Do they ask their boss? Or do they figure it out on their own?

They interviewed 19 different knowledge workers (like lawyers, coders, marketers, and researchers) to find out. What they discovered is a bit like a game of "Secret Agent" mixed with a "Cooking Class."

The Two Main Skills Workers Value

The study found that workers are trying to master two very different skills at the same time:

1. The "Explorer" Skill (Knowing what the gadget can do)
Workers want to know all the cool things the AI can do. They want to find new ways to use it to make their jobs easier.

• How they learn this: They mostly learn by talking to each other. It's like a potluck where everyone brings a new recipe. One person says, "Hey, I used AI to write a poem," and another says, "Oh, I used it to fix a spreadsheet!"
• The result: This sharing helps everyone get smarter about what the tool is capable of.

2. The "Invisible Magician" Skill (Hiding that you used the gadget)
This is the surprising part. Workers also value the ability to use the AI so well that no one can tell they used it.

• Why do they do this?
  • Fear of Judgment: Some people worry their boss or colleagues will think, "Oh, you didn't do the work yourself; you just asked the robot." They feel ashamed or fear being seen as lazy.
  • Proving Expertise: Others do it to show off how smart they are. If you use AI but edit the output so perfectly that it looks 100% human, you get credit for being a "genius." As one participant put it: "If you're very clever, no one knows you've used it."
• The Analogy: Imagine a chef who uses a high-tech sous-chef robot to chop vegetables. If the chef hides the robot and serves the dish as if they chopped it by hand, they get a standing ovation for their knife skills. If they admit they used the robot, people might say, "Well, that wasn't really your cooking."

The Problem: The "Silent Kitchen"

Here is the catch. When everyone tries to be the "Invisible Magician," something bad happens to the learning process.

• The Paradox: To learn how to use the tool best, you need to share your mistakes and tricks with others (like the "Explorer" skill). But, to look smart and avoid judgment, you hide the fact that you used the tool at all (the "Invisible Magician" skill).
• The Result: The kitchen becomes silent. No one admits they are using the gadget. No one shares their "secret recipes" because they are afraid of being judged or because they want to keep their "genius" status.
• The Consequence: Even though everyone is using the tool, the company doesn't know how it's being used, what mistakes are being made, or how to teach new employees. It creates a culture of secrecy rather than teamwork.

The "Workslop" vs. The "Masterpiece"

The paper also notes that sometimes people hide the AI because the raw output is actually bad.

• The "Workslop": If you just copy-paste what the AI says, it often sounds robotic, uses weird words (like "delve" or "tapestry"), and lacks real human insight. This is like serving a frozen dinner and calling it a gourmet meal.
• The Fix: Skilled workers "critique" the AI. They fix the weird words and add their own human touch. They hide the AI not just to look cool, but to ensure the final product is actually good. However, by hiding this process, they miss the chance to teach others how they fixed it.

What the Authors Suggest

The researchers don't say "stop using AI." Instead, they suggest a few changes to make the office a better place to learn:

1. Stop the Shame: Managers need to create a safe space where admitting you used AI isn't a crime. It's okay to say, "I used the robot to draft this, and here is how I improved it."
2. Celebrate the "How," Not Just the "What": Instead of just looking at the final report, talk about the process. Share the "prompts" (the instructions given to the AI) and the edits made.
3. Make Sharing Rewarding: Just like people get "likes" on social media, companies could reward people for sharing their AI tips and tricks, making it cool to be transparent rather than cool to be secretive.

Summary

The paper argues that while office workers are getting very good at using AI, they are also getting very good at hiding it. They do this to look smart and avoid judgment. But this secrecy stops the whole team from learning together. To fix this, workplaces need to stop treating AI use as a secret and start treating it as a shared skill that everyone can learn from openly.

Technical Summary

1. Problem Statement

While Generative AI (GenAI) is rapidly transforming knowledge work, existing research on AI literacy lacks empirical insight into how knowledge workers develop competencies within the complex social dynamics of the workplace. Current frameworks often focus on technical skills or formal education, overlooking the unstructured, informal learning processes used by professionals.

Key gaps identified include:

• The "Hiding" Phenomenon: Workers often conceal GenAI usage due to social stigma or fear of judgment, yet the impact of this behavior on organizational learning and transparency is poorly understood.
• Competency Definition: It is unclear which specific competencies (technical vs. socio-technical) workers value most in practice.
• Learning Strategies: There is limited understanding of how workers navigate the tension between learning from colleagues and maintaining a professional identity that may require obscuring AI assistance.

2. Methodology

The study employed a qualitative approach to investigate the beliefs, strategies, and social dynamics of GenAI adoption.

• Participants: 19 knowledge workers from diverse sectors (IT, Law, Creative Industries, Academia, HR, Finance, etc.).
• Recruitment: Participants were recruited via Prolific, screened for specific criteria:
  • Employed in sectors adopting GenAI rapidly.
  • Use GenAI at least weekly for work-related purposes.
  • Have >1 year of experience in their current role.
  • Work in environments without explicit prohibitions on GenAI (or uncertain policies).
• Data Collection: Semi-structured interviews (50–60 minutes) conducted online via Microsoft Teams between December 2024 and January 2025.
• Analysis:
  • Positionality: The research team adopted a critical realism stance, acknowledging that interpretations reflect subjective experiences and social contexts rather than objective reality.
  • Thematic Analysis: A template approach was used, scaffolding initial coding with two existing AI literacy frameworks (Pinski & Benlian; Annapureddy et al.) and the COM-B model (Capability, Opportunity, Motivation) to analyze learning behaviors.
  • Rigor: Multiple authors reviewed codes and themes to ensure reflexivity and critical reflection.

3. Key Contributions

The paper makes three primary contributions to the field of Human-Computer Interaction (HCI) and AI literacy:

1. Empirical Grounding: Provides a detailed account of how GenAI literacy is practiced and negotiated by 19 diverse knowledge workers, moving beyond theoretical frameworks to real-world application.
2. Reframing "AI Hiding": Challenges the prevailing view that hiding AI use is solely rooted in shame. The study finds that successfully erasing cues of GenAI usage is often viewed as a positive validation of domain expertise. However, this behavior creates a paradox: it signals competence to peers but simultaneously reduces opportunities for knowledge sharing and transparency.
3. Framework Extension: Extends existing AI literacy frameworks by highlighting the critical role of socio-technical competencies. It argues that human stakeholders are not just resources for learning but also sources of social influence that shape usage patterns, necessitating new design considerations for open dialogue and disclosure.

4. Key Results

The study identified two core competencies and two corresponding learning strategies, revealing a central tension between individual identity and collective learning.

A. Core Competencies

1. Technological Competency (Awareness of Use Cases):
   • Workers value the ability to identify novel applications of GenAI to gain a competitive advantage.
   • This awareness is often driven by the fear of job displacement or the desire to stay ahead of industry trends.
2. Socio-Technical Competency (Cue Removal):
   • Workers develop the ability to detect and remove "tells" (e.g., specific phrasing like "delve," excessive comments in code, lack of context) that indicate GenAI usage.
   • Motivations for Removal:
     • Critique/Quality Control: Removing generic or low-quality outputs to ensure work meets professional standards.
     • Expert Identity Validation: Successfully masking AI use signals that the worker possesses deep domain knowledge, distinguishing them from those who rely blindly on AI.
     • Concealment/Stigma: Avoiding negative judgment from older generations or conservative management.

B. Learning Strategies

1. Learning from Others:
   • Informal knowledge sharing (colleagues, LinkedIn, social media) is vital for discovering new use cases and addressing knowledge gaps.
   • However, this is often strategic; workers may share knowledge to normalize AI use or seek validation, but they may withhold specific details to maintain a competitive edge.
2. Hands-On Experience:
   • Competence is largely implicit and developed through trial-and-error.
   • Workers must learn to evaluate context-specific appropriateness, understand model limitations (hallucinations), and develop metacognitive skills to articulate prompts effectively.

C. The Central Tension

The study reveals a paradox: While learning from others is essential for developing technical awareness, the drive to validate one's expertise through "cue removal" actively suppresses transparency.

• Workers who successfully hide AI use are praised as experts.
• This behavior creates "organizational silences" where AI usage is widespread but unacknowledged, preventing the sharing of best practices, error detection, and collective sense-making.
• Disclosure is often avoided due to perceived indifference from colleagues ("just get the job done") or fear of social stigma.

5. Significance and Implications

The findings have significant implications for HCI design, organizational management, and AI literacy frameworks.

• For Organizations:

  • Culture of Openness: Leaders must foster psychological safety to reduce the fear of judgment associated with AI use. Passive allowance of AI is insufficient; active discourse is required to break "organizational silences."
  • Incentivize Sharing: Organizations should reward the sharing of GenAI workflows, prompts, and lessons learned, treating these as valuable intellectual assets rather than secrets.
  • Policy Clarity: Clear policies on disclosure are needed to reduce uncertainty and encourage responsible usage.

• For Designers (HCI):

  • Support Disclosure: Design tools that make it easier to share "user-generated knowledge" (e.g., prompts, workflows) without exposing the user to stigma.
  • Appreciation Systems: Implement mechanisms similar to open-source "kudos" or "hugs" to positively reinforce the sharing of GenAI artifacts.
  • Metacognitive Focus: Move beyond simple prompt sharing to help users understand the strategies behind successful interactions, fostering deeper learning.

• For AI Literacy Frameworks:

  • Literacy models must evolve to include socio-technical competencies, recognizing that the social context (reputation, identity, peer judgment) is as critical as technical skill in determining how AI is adopted and used.
  • The definition of "literacy" must account for the tension between individual identity work and collective learning.

In conclusion, the paper argues that true GenAI literacy in the workplace cannot be achieved through technical training alone. It requires addressing the social dynamics that drive workers to hide their tool usage, thereby undermining the transparency and collaborative learning necessary for sustainable adaptation to rapid technological change.