The New Reading Experience
Certain writing breaks the fourth wall, pulling you from immersion into meta-awareness—much like this sentence just did. When reviewing a client presentation or proposal and spotting a typo or formatting error, the illusion fractures. For those attuned to these details, the spell is instantly broken. These mistakes undermine the polished professionalism expected in high-stakes deliverables.
Reading and writing through large language models creates a similar disruption. Certain patterns now serve as tells: parallel structures in groups of three, liberal use of em dashes, and the contrarian “it’s not x, it’s y” construction. When I encounter an em dash mid-paragraph, I’m transported from reader to skeptic, asking: can I trust this content?
The statistics suggest this is no longer a marginal concern. Recent analysis found that as of November 2024, AI-generated articles briefly comprised over 50% of newly published web content, though the ratio has since stabilized around an even split between human and AI authorship. More striking still, approximately 74% of web pages created in April 2025 contained some AI-generated content.1 The more ubiquitous AI-assisted writing becomes—from reports to Outlook emails to Teams instant messages—the more normalized it will be. Perhaps this is simply the evolution of workplace communication.
But the question remains: does it matter if content appears human-written when it communicates effectively, perhaps even articulating our thoughts with greater eloquence? The emergence of “human-generated music” labels on YouTube (not on anyone’s 2025 bingo card) suggests we’re entering an era where authenticity itself becomes a differentiator, even marketing copy.
Perhaps we need clearer boundaries: AI for brainstorming and first drafts, human refinement for the final product. Or we must deliberately re-edit outputs to embed our distinctive voice. My concern centers on homogenization—that we’ll gradually converge toward a uniform communication style. I also worry about those still developing their professional voice, growing up in this environment from the start. Is it too pessimistic to fear this will flatten their uniqueness before they discover it?
There are silver linings worth considering. One positive outcome could be the resurgence of underused stylistic choices like the em dash. And if professionals are relying on AI tools to generate content, they’re likely using punctuation and grammar more accurately than they would otherwise. I’ve debated reducing my own em dash usage—I employ them more than most—simply to signal this is a human-created, trustworthy space. But that would mean fighting my authentic voice and abandoning perfectly appropriate punctuation conventions.
Preserving Critical Thinking Skills
We risk becoming performers who lip-sync rather than sing live. This comparison matters because workplace conversations aren’t scripted performances—they’re constant, impromptu, and unpredictable. When we delegate too much cognitive load to AI tools, we may accelerate task completion, but we leave ourselves vulnerable in unplanned situations. Many roles, particularly in consulting, depend on real-time thinking and adapting to shifting conversational dynamics.
While 75% of knowledge workers now use AI at work (with 46% having started within the past six months), and these users report saving approximately 5.4% of their work hours weekly, the quality of implementation varies significantly.2 Analysis of workplace discussions reveals that 42% of critical comments cite a gap between leadership expectations and actual productivity gains.3 Success appears less about the technology itself and more about how thoughtfully it’s integrated into workflows.
How do we maintain our competitive edge—the ability to think sharply, communicate clearly, and operate without a script—while still benefiting from AI capabilities? Here are practices I’ve begun implementing:
-
Use AI to Sharpen, Not Substitute. Language models can help refine thinking, but shouldn’t replace it. I draft the slide deck, memo, or outline independently before consulting AI tools for reframing, stress-testing logic, or formatting improvements. I’m concerned about outsourcing my point of view formation entirely if I start with AI from the outset.
-
Practice Unscripted Thinking. My role involves impromptu client meetings, leadership discussions, and spontaneous problem-solving—letting those capabilities atrophy creates genuine professional risk. I’m building “live response” exercises into my routine: answering complex questions aloud in one minute, converting abstract requests into concrete action items, presenting without slides. These drills maintain readiness for unprepared moments.
-
Question AI Output Actively. Early in my AI adoption, I treated model responses as authoritative, assuming they drew from comprehensive, high-quality sources. I’ve since shifted to interrogating and pushing back on initial outputs rather than accepting them as finished products—a healthier dynamic that improves results.
-
Do the Hard Work. It’s tempting to bypass the messy middle of strategic thinking. It’s easy to immediately consult AI tools for answers before forming independent perspectives. This extends the first principle but emphasizes purposefully inhabiting strategic thinking space—it requires time in an era that only values speed. Resist that pressure where it matters. In a workplace where more tasks can be delegated, real value comes from knowing what not to delegate.
-
Journal Regularly. Write freely and uncensored. Write as if no one will read it. Map connections across different aspects of your professional and personal life. Sit in reflection. Let your thoughts remain your own, without AI input shaping them from the start.
-
Read Deliberately. ‘m increasingly conscious of what I consume—watching for AI tells, expanding my vocabulary, and diversifying my informational diet across different content types and sources. Don’t let AI-generated content become your primary intellectual input. Writers and readers existed long before AI-generated material—producing excellent work that these models trained on to begin with.
This framework applies to any language model—ChatGPT, Gemini, Claude, or emerging alternatives. If you’ve developed practices for maintaining your edge while leveraging these tools effectively, I’d welcome hearing about them.