Making AI Use Visible in Professional Work

In the first post of this series, I argued that AI is likely to change tasks before it eliminates entire jobs. As AI becomes embedded in professional workflows, graduates will need to decide which tasks to complete themselves, which to perform with AI assistance, and which to delegate more fully to AI systems. But those choices introduce another responsibility: How will other people know what role AI played in the work? In education, we often discuss this as a disclosure problem. We ask students to indicate whether they used AI to brainstorm, outline, draft, revise, summarize, or generate content. That is an important starting point. But the workplace version of transparency is more complicated. A statement such as “AI was used in the preparation of this document” may technically disclose AI involvement, but it tells us almost nothing about what actually happened. Did AI correct a few sentences? Did it summarize source material? Did it generate the analysis? Did it recommend a course of action? Did it make decisions about what information to include? Did the employee independently verify the output? Did confidential information enter the system? Who reviewed the final result? A generic disclosure can reveal that AI was present while leaving its actual influence invisible. That distinction will matter as AI-assisted work becomes more common. Imagine that an employee submits a market analysis containing a recommendation about where an organization should invest its resources. If AI was used only to improve the formatting, that probably has little effect on how the recommendation should be evaluated. But what if AI selected the data, interpreted the trends, or generated the recommendation itself? Those are very different forms of assistance. They introduce different questions about accuracy, expertise, bias, confidentiality, and responsibility. The same is true in other professions. If a nurse uses AI to improve the wording of general educational material, that is different from using it to summarize a patient’s medical history. If a human resources professional uses AI to organize interview notes, that is different from using it to rank applicants. If a technician uses AI to rewrite a service report, that is different from using it to diagnose the problem. If an instructor uses AI to create a sample activity, that is different from using it to evaluate student work or generate individualized feedback. The question is not simply whether AI was used. The more important question is what role it played in producing the outcome. This is why I think we need to help students move beyond disclosure and toward documentation. Disclosure says that AI contributed. Documentation explains how. That difference may appear small, but it changes what transparency makes possible. When AI use is documented, a colleague can identify which parts of the work require closer review. A supervisor can determine whether the use complied with organizational policy. A client can understand how a recommendation was developed. If something goes wrong, the people involved can reconstruct the process instead of relying on memory or assumptions. Documentation also makes it possible to distinguish between responsible AI assistance and unexamined dependence. The National Institute of Standards and Technology emphasizes documentation throughout its AI Risk Management Framework. It describes documentation as a way to improve transparency, support human review, clarify roles, and strengthen accountability. Its generative AI guidance goes further by recommending systems that can track when content is generated, modified, and shared. (NIST, 2023; NIST, 2024) That guidance is primarily written for organizations managing AI systems, but the underlying principle also applies to individual professional practice. If AI meaningfully contributes to a decision, analysis, communication, design, or recommendation, there should be enough information to understand that contribution. This does not mean professionals must save and share every prompt they type. Complete transcripts can create their own problems. They may contain sensitive information, irrelevant experimentation, proprietary material, or so much detail that meaningful review becomes difficult. Transparency is not the same as producing the largest possible record. Useful documentation should make the important parts of the process visible. That might include identifying the system used, describing the task it performed, explaining what information it received, recording how the output was evaluated, and clarifying what the human changed or decided afterward. Most importantly, it should identify who accepts responsibility for the final result. That last part matters because AI disclosure can easily become a way of distancing ourselves from our own work. “The AI suggested it.” “The system generated the recommendation.” “That is what the model said.” These statements may describe what happened, but they do not resolve responsibility. If a professional chooses to rely on an AI-generated recommendation, that choice remains part of the professional’s work. Making AI visible should not provide an escape from accountability. It should make accountability clearer. This is also why transparency cannot be reduced to labeling something “AI generated.” AI involvement exists along a spectrum. A person might use it to brainstorm possibilities, retrieve information, summarize documents, write code, generate an image, recommend a decision, or complete an entire workflow. A binary label treats all of these uses as equivalent when they are not. Emerging research suggests that people already make these distinctions. A 2025 study of disclosure, ownership, and accountability in human–AI co-created work found that people’s expectations about disclosure depend partly on how substantially AI contributed. Questions about ownership and responsibility become more complicated as AI moves from providing limited assistance to making a major creative or intellectual contribution. (Draxler et al., 2025) More recent studies also point to a transparency dilemma. People frequently say they want to know when AI contributed to something, but disclosure can lower their trust in the work. The effect appears to depend on the context, the extent of AI involvement, and how much detail is provided. In one study of AI-assisted news, approximately two-thirds of participants preferred detailed disclosure, while many who preferred a brief label still wanted the option to access additional information. (Guan et al., 2026) That finding raises an important concern. If disclosing AI use can cause others to view work less favorably, people may have an incentive to hide it. Students already encounter versions of this problem. They may receive general encouragement to learn about AI while also believing that admitting to its use will cause instructors to distrust their work. Graduates may enter workplaces with similarly conflicting messages: employees are expected to use AI to improve productivity, but the organization has not established clear expectations for when or how that use should be reported. Transparency becomes difficult when the culture surrounding it is punitive or unclear. If every disclosed use is treated as evidence that the work is less valuable, people will learn not to disclose. If no disclosure is expected, organizations may have little idea how AI is actually being used. If policies prohibit tools without addressing the AI features already embedded in common workplace software, employees may not even realize when meaningful AI use has occurred. This is not only an individual responsibility. Organizations must create conditions in which useful transparency is possible. They need to distinguish between low-risk and high-risk uses. They need to explain what kinds of AI involvement require documentation. They need to identify which information may never be entered into particular systems. And they need to respond to honest disclosure in ways that encourage responsible practice rather than drive AI use underground. But graduates also need to be prepared to enter workplaces where that clarity does not yet exist. They may need to ask questions that no one else has asked. Does our organization have an approved AI system? Can I use it with internal information? Do I need to disclose AI assistance to a client? What records should I retain? Who reviews an AI-assisted recommendation? Who is responsible if the output causes harm? These are not questions about prompting. They are questions about professional judgment. That is why disclosure activities in college should do more than ask students to add a sentence to the end of an assignment. A required statement can become another box to check without producing much understanding. Instead, students can practice describing the role AI played in their process. They can explain what they asked the system to do, why they chose to use it for that task, how they evaluated its contribution, what they changed, and what parts of the work remained their responsibility. This creates a more accurate picture of human–AI collaboration. It also helps students examine their own use. When students must describe what AI contributed, they may recognize that the tool influenced more of the work than they initially realized. A brainstorming session may have framed the direction of the project. An early summary may have shaped which evidence they considered important. A generated outline may have determined the structure of their argument before they began writing. AI influence is not always limited to the sentences it generates. It can shape attention, frame possibilities, establish assumptions, and affect which paths seem worth pursuing. Documenting AI use helps make those less visible influences available for reflection. That may be one of the most important reasons to teach transparency. The goal is not merely to help an instructor, employer, or client inspect the work. It is also to help the person using AI understand the process well enough to take ownership of it. Ultimately, that is what professional transparency should preserve: ownership. Not ownership in the narrow sense of copyright, but ownership of the choices, interpretations, and consequences associated with the work. A professional should be able to say: This is where AI contributed. This is how I evaluated that contribution. This is what I decided. And I remain responsible for the result. As AI becomes part of ordinary work, we may no longer be able to judge professional capability by asking whether someone used it. The more meaningful question will be whether they used it intentionally, whether they made its role visible when it mattered, and whether they can stand behind the work that resulted. But standing behind AI-assisted work requires more than transparency. It requires accepting responsibility for the decisions made with it. That is where the next post will turn. Continuing the Conversation Series 1: AI Is Exposing Existing Problems ✓ Completed
Series 2: What We Do About It ✓ Completed
Series 3: Cultivating Human Thinking in an AI World ✓ Completed
Series 4: Learning Alongside AI ✓ Completed
Series 5: Preparing Students for an AI World
Current Post (2 of 8): Making AI Use Visible in Professional Work
Next Up: Taking Responsibility for AI-Assisted Decisions‍