Something is happening in your organization right now. An employee submits a polished memo. A manager sends a tightly structured performance summary. A candidate delivers a sharp cover letter. The writing is clean, organized, and strangely uniform. Then you ask one follow-up question. The silence is deafening.
AI overreliance in the workplace has a fingerprint. Most organizations are not trained to read it.
This post is not an argument against AI. Tools that reduce administrative noise and accelerate output are genuinely useful, and any HR professional who has spent hours wrestling with a spreadsheet-based merit process knows exactly what that relief feels like. The problem is not AI use. The problem is AI use without comprehension. When employees can’t describe, defend, or build on what they’ve submitted, the output is noise dressed as signal.
The Tells Are Hiding in Plain Sight
AI-generated work has patterns. Once you know them, you can’t unsee them.
The em dash epidemic is the most recognized signal. Tools like ChatGPT and Claude default to em dashes frequently and structurally, in ways most humans simply don’t write. A single em dash in a paragraph isn’t evidence of anything. Three em dashes in a 200-word update from someone who has never punctuated that way before? That’s a tell.
Vocabulary drift is subtler but equally revealing. Words like “keen,” “leverage,” “delve,” “nuanced,” and “multifaceted” appear with unusual frequency in AI-assisted writing. These are not wrong words. They’re just not words most people reach for in workplace prose. When an analyst who typically writes in plain, direct sentences suddenly produces copy dense with “comprehensive frameworks” and “robust methodologies,” the drift is worth noting.
Structural over-formality is another pattern. AI-generated responses frequently default to three-point structures, numbered lists, and headers, even when the question was conversational. If someone asked for a quick opinion on a job description and received a five-section formatted document with an executive summary, that’s not thoroughness. That’s AI output shape-shifting into a deliverable.
None of these signals, standing alone, proves anything. Together, they raise a reasonable question: Does the person behind this output actually understand what it says?
The Real Test Is What Happens Next
Here’s the part most managers miss. AI overreliance in the workplace is not primarily a problem of writing quality. It’s a comprehension problem, and comprehension only surfaces when you ask.
The diagnosis is straightforward. After receiving AI-adjacent work, ask the employee to walk you through their reasoning. Not “did you use AI?” because that question produces defensiveness and minimal information. Instead, try:
- “Walk me through how you got to this recommendation.”
- “What would you change about this if the situation shifted slightly?”
- “What’s the assumption this analysis is most dependent on?”
A fluent AI user can answer these questions. They built the output by directing the tool, reviewing the results critically, and making judgment calls along the way. Someone who pasted a prompt and submitted the first result will stall, paraphrase what’s already on the page, or confess they’re not sure.
The inability to explain the output is the signal. Everything else (the em dashes, the vocabulary, the formatting) is circumstantial. Comprehension is the standard.
This distinction matters enormously for compensation and performance decisions. Merit increases, promotions, and role expansions are increasingly evaluated on the quality of someone’s analytical judgment, not just their deliverable count. If that judgment is outsourced wholesale to an AI without critical engagement, the compensation system rewards an intermediary, not the employee.
Why HR and Managers Are Poorly Equipped to Catch It
AI overreliance in the workplace thrives in organizations where output volume is the primary performance signal. When the review system rewards deliverables rather than reasoning, employees have little incentive to deeply engage with what they produce. The AI becomes a shortcut that no one has explicitly sanctioned, and no one is actively catching.
Most managers are also not trained to distinguish AI dependency from AI fluency. These are genuinely different things. AI fluency means using tools selectively, directing them with precision, and taking ownership of the output. AI dependency means using tools as a substitute for thinking, producing work that looks complete but isn’t backed by understanding.
HR professionals are in a unique position here. They see patterns across teams, departments, and levels that individual managers cannot. An uptick in polished but shallow documentation across a department is a data point. Sudden uniformity in how different employees structure performance self-assessments is a data point. Compensation systems that track output quality over time, rather than just completion, can surface these trends earlier than any single manager’s observation.
Organizations that build AI fluency expectations into role definitions now will not have to retrofit them later. That recalibration will be significantly harder to execute after compensation structures and performance standards are already set around AI-augmented output norms.
A Decision Framework for HR and Managers
When you suspect AI overreliance in the workplace, resist the impulse to confront. Instead, verify through conversation. Here is a simple framework:
Step 1: Observe patterns over time, not isolated incidents. One polished document proves nothing. Consistent output style drift, sudden capability jumps, and recurring vocabulary patterns together build a picture.
Step 2: Ask process questions, not product questions. “How did you approach this?” surfaces comprehension. “Did you use AI?” surfaces defensiveness.
Step 3: Differentiate AI-assisted from AI-dependent. Employees who use AI as a tool will walk you through their reasoning clearly. Employees who used it as a crutch will struggle to explain what the output actually means.
Step 4: Calibrate compensation and performance standards accordingly. If a role requires analytical judgment, the evaluation standard should assess judgment, not just deliverable quality. Work with your compensation team to ensure job descriptions and competency models reflect the skills that matter in an AI-augmented environment.
Step 5: Set clear, written expectations. Organizations that establish AI use policies, including what constitutes appropriate AI use for a given role, are better positioned to evaluate performance fairly. Ambiguity benefits no one.
What This Means for Compensation
AI overreliance in the workplace has a direct compensation implication that most organizations haven’t confronted yet. If employees in the same role are performing at meaningfully different levels of genuine comprehension and judgment, and merit decisions are based solely on output quality, the compensation system is rewarding the wrong inputs.
This isn’t an abstract concern. Consider two analysts who submit similarly polished quarterly compensation analyses. One built the analysis, reviewed the assumptions, and can defend every number. The other prompted an AI tool, formatted the output, and submitted it. If the evaluation standard is the document itself, both receive the same merit consideration. If the evaluation standard includes comprehension and judgment, the results diverge.
Tools like SimplyMerit help streamline merit administration by removing the spreadsheet friction that distracts managers from the evaluative work that actually matters. But no tool substitutes for the manager conversation that reveals whether the performance behind a merit recommendation reflects real competency or well-formatted AI output.
The compensation question is not whether employees use AI. It’s whether they understand what they’re recommending, building, and deciding. And it’s whether your evaluation process is designed to find out.
This Post Almost Proved Its Own Point
A note from Laura Morgan, CEO, MorganHR Inc.
Here’s something Alex did not put in his original draft: I almost couldn’t publish it.
Alex submitted this post the way many employees submit work right now: polished, structured, complete on the surface. When I read it, something felt off. The argument was sound. The flow was clean. Yet when I asked him to walk me through his sourcing on a few key points, the conversation got uncomfortable fast. He could explain the framework. He could not explain the evidence behind it as someone who had genuinely wrestled with the research would.
To his credit, he owned it immediately. His draft had been heavily AI-assisted, more than he’d disclosed, and he hadn’t stress-tested the underlying claims the way the topic demands. So we rebuilt the post together. His framework stands. Every research reference and verification pass is now actually verified.
I’m sharing this because it would be a little too convenient to publish a post about AI overreliance without acknowledging that it nearly suffered from exactly that. The irony is instructive. AI dependency doesn’t announce itself. It produces work that looks finished, reads well, and passes a casual review. The gap only surfaces when someone asks the follow-up question.
That’s the whole point of this post. And apparently, it applies to the people writing it, too.
If you’re a manager reading this, the lesson isn’t that AI use is disqualifying. Alex is a strong analyst. The lesson is that even skilled people can slip into dependency mode when the tool makes output feel easy. Build the follow-up question into your process. Not to catch anyone. The work actually gets better when people have to stand behind it.
One more thing.
Then I asked him one more question. I looked at him and said, “Okay. I’m dead. How would you walk our client through this?”
Not as a trap. As a real question. The kind I ask when I want to know if someone actually owns what’s in front of them.
Because that’s the line AI cannot cross. It can produce the output. It cannot sit across from a client who is skeptical, or quietly confused, or too polite to say they disagree, and feel the room shift. It cannot move eyes on eyes, mouth to ear, and know when to stop talking and when the silence means something. It cannot recalibrate mid-sentence because a client’s expression changed.
That is the skill. Not the document. The judgment behind it, and the presence to deliver it.
Alex knew the answer. He just needed to be reminded that the answer had to come from him. His instincts are sharp. His craft is real. What the tool had temporarily replaced wasn’t his ability. It was his ownership of it. Those are not the same thing, and the difference matters enormously when you’re sitting across from a client who is paying for the person, not the output.
Key Takeaways
- AI overreliance in the workplace leaves detectable patterns: em dashes, vocabulary drift, overly formalized structure, and an inability to explain outputs.
- The comprehension test (asking employees to walk through their reasoning) is the most reliable diagnostic.
- HR professionals are positioned to detect patterns across teams that individual managers miss.
- Compensation and performance systems must evaluate reasoning and judgment, not just deliverable quality.
- Organizations that define AI fluency expectations in job roles now will avoid a much harder recalibration later.
- When you find AI overreliance, the BEER framework (Behavior, Expectation, Effect, Resolution) converts a charged moment into a productive, documented conversation.
WHEN YOU FIND IT: USE THE BEER FRAMEWORK
Another note from Laura, because the story didn’t end with the draft.
Here is what actually happened next, and why I’m telling you in a blog post rather than in a closed-door conversation.
After Alex and I worked through the draft together, I shared it with a few members of our team for a final review. The work was strong. The research was verified. I was proud of what we had rebuilt. Then we got to slide 2 of the supporting presentation, and someone on my team said, casually, without any apparent awareness of what they were implying: “That’s what you did with your presentation.”
They meant it as a compliment to the AI. What it communicated to me was something else entirely: that my own team had unknowingly assumed I couldn’t tell the difference between my craft and a machine’s output. Thirty years of compensation expertise, client work, and deliberate communication choices were, in that moment, indistinguishable from a first-pass AI draft.
The sting of that is real. And it is exactly the kind of moment that, if you don’t handle it with structure, becomes either an overreaction or a silence that calcifies into resentment.
So I used the BEER framework. The same one we published on this blog.
The Conversation, Word for Word
Behavior: “In the presentation review, you said ‘that’s what you did with your presentation’ when referring to the AI-generated slides, without distinguishing between the work I’ve built over the years and what the tool produced.”
Expectation: “When we review work that involves AI assistance, I expect the team to be precise about attribution. Not because I need credit, but because sloppy attribution erodes the professional standards we ask our clients to hold their own teams to.”
Effect: “When my own work is conflated with AI output, it signals that the assumption in the room is that I don’t know my craft, or that it doesn’t matter whether I do. That undermines my credibility, and it quietly lowers the bar for everyone.”
Resolution: “Going forward, when AI contributes to something, name it specifically. And when a human’s expertise shaped it, name that too. Both things can be true, and the distinction is exactly what this entire post is about.”
The conversation was uncomfortable for about four minutes. After that, it was one of the most clarifying exchanges our team has had in months.
That’s what BEER does. A moment that could become a festering assumption becomes a documented, forward-facing standard instead. Behavior gets named without attacking the person. The expectation connects to a professional principle rather than a personal grievance. Everyone walks away with a resolution they can actually act on.
If you’ve caught AI overreliance on your team, or had your own work casually conflated with machine output, you don’t have to choose between letting it go and blowing it up. BEER gives you the structure to do neither.
Read the full BEER framework guide: Grab a BEER: The Framework That Fixes Workplace Relationships
Quick Implementation Checklist
- Review recent employee deliverables for recurring AI-style markers (em dashes, vocabulary uniformity, structural over-formality).
- Train managers on the difference between AI fluency and AI dependency.
- Build process questions into performance conversations, not just output reviews.
- Audit current job descriptions and competency models to align with AI fluency expectations.
- Establish a written AI use policy that defines appropriate use by role and function.
- Confirm that your merit evaluation process includes behavioral and judgment indicators, not only output quality metrics.
Frequently Asked Questions
For HR Professionals and People Managers
Q: How do I raise AI overreliance concerns without accusing an employee of cheating?
A: Frame it as a competency conversation, not an accusation. Ask process-oriented questions, such as “walk me through your thinking,” rather than tool-oriented ones. Most employees will either demonstrate genuine comprehension or reveal the gap themselves through their explanation.
Q: Is there a reliable way to detect AI-generated writing?
A: Detection tools exist, but they are imperfect. Rather than relying on technology to detect AI use, build evaluation practices that emphasize transparency in reasoning. If an employee can credibly explain and defend their work, the origin of the draft matters far less.
Q: What if the employee’s role genuinely permits heavy AI use?
A: That’s a legitimate design choice, but it still requires a competency standard. Even in roles where AI drafts are expected, employees should be able to review critically, edit accurately, and take accountability for what they submit. AI fluency, not just AI access, is the standard.
For Executives and HR Leaders
Q: How does AI overreliance in the workplace affect compensation strategy?
A: When merit decisions are based on deliverable quality rather than the judgment behind them, compensation systems risk rewarding AI output rather than employee competency. Performance standards need to evolve alongside the tools employees are using, and that recalibration starts with what organizations choose to evaluate.
Q: Should we prohibit AI use in performance documentation?
A: Prohibition is rarely the right answer, and it’s nearly impossible to enforce consistently. A clearer approach is to define what good AI-assisted work looks like in your context, then evaluate accordingly. Employees who engage critically with AI output are demonstrating a skill. Those who submit AI output without review are not.
Q: What does an AI fluency standard look like in a job description?
A: It looks like a behavioral competency, not a technology requirement. For example: “Applies analytical judgment to synthesize information from multiple sources, including AI tools, and communicates findings with clarity and accountability.” The emphasis is on judgment, not the tool.
Regulatory and Compliance Considerations
Q: Are there regulatory requirements around AI use in HR processes?
A: Regulatory frameworks are evolving rapidly. In the United States, the EEOC has issued guidance on AI use in hiring and employment decisions, and several states have passed or proposed laws governing automated decision-making in employment contexts. Organizations should consult legal counsel to ensure compliance. Always verify current requirements with qualified counsel.
Q: Does AI use in performance documentation create legal exposure?
A: Potentially, yes, particularly if AI-generated documentation is used to support adverse employment actions without human review and accountability. Establishing clear documentation standards that require employee sign-off and manager verification helps create an auditable record of human accountability.
Ready to Take Action?
AI overreliance in the workplace is not a future problem. It is already shaping the work on your managers’ desks and in your compensation reviews. Start with one simple step: in your next performance conversation, ask the employee to walk you through their reasoning. What you learn will tell you everything.
Have questions about how to build AI fluency standards into your compensation or performance framework? Reach out to the MorganHR team and explore how structured merit administration can give managers more time for the evaluative work that actually requires human judgment.