Issue 155  October 9th 2026

The rigor behind the result - making the human contribution visible in AI-assisted work.

A UX researcher noticed something missing from an AI-assisted synthesis. The summaries had flattened differences in participants’ views and standardized very different workflows. He remembered quotes that carried more emotional weight than the synthesis conveyed, so he went back to the transcripts.

Recovering those quotes brought back some of the intensity of participants’ experiences. The report could help readers understand more than the broad themes. When his manager dug into how he had done the work, he appreciated the researcher’s rigor. The manager looked beyond the finished result and asked about the process that produced it. He could see a contribution that the report alone might have concealed.

We talk a lot about what AI saves us. We talk less about the work that follows: checking original evidence, recovering distinctions, deciding which details matter, and correcting an interpretation that sounds reasonable.

Some of that is rework. Some is the professional judgment the task always required. Either way, it takes time, and a polished output rarely tells you how much.

IBM’s recent CHRO study puts numbers around this concern. Eighty percent of surveyed CHROs said AI adoption creates invisible work, including validating recommendations, fixing mistakes, supplying context, and managing exceptions. Forty-two percent of employees said AI increases their work or their work goes unrecognized. These are survey responses, not proof that AI causes these outcomes, but they raise a useful question: are we counting the work required to make the output dependable?

For the researcher, empathy involved preserving differences between people’s experiences and recovering emotional context the synthesis had compressed. That required attention to the source material. A vivid quote still needs context; its emotional force doesn’t establish how widely an experience is shared.

Advocacy enters when we make that contribution visible. We need language for explaining what our involvement changed. “Human reviewed” gives the reader very little to go on. What was checked? What was corrected? What remains uncertain?

This also matters for managers. If people are responsible for the result, they need time to examine it and authority to challenge it. A review has limited value if the reviewer is expected to approve whatever the system produces.

Transparency gives others something on which to base their confidence. It may reveal careful work, or it may expose a gap that needs attention. Both are useful. We should want confidence that survives questions about how the work was done.

THE TELL

You can explain what the AI produced, but struggle to explain what your involvement changed.

Or your team celebrates the speed of the first draft while the time spent checking and correcting it disappears from the conversation.

Watch for “I reviewed it” becoming the whole explanation. Ask what that review involved. The answer should connect to evidence, a decision, or an acknowledged limitation.

YOUR PRACTICE THIS FORTNIGHT

Choose one AI-assisted deliverable that someone else will rely on. When you share it, add a short account of your contribution:

• AI contributed: What did the tool actually do?
• I checked or changed: What did you examine, preserve, correct, or reject?
• This matters because: How did that intervention affect the result, and what uncertainty remains?

Keep it proportionate to the stakes: a few sentences will often do - you’re giving the recipient a reason to assess the work with confidence.

If you manage the person producing it, ask: “What required your judgment here?” Then consider whether your expectations leave room for that work. Recognizing rigor means allowing time for it, too.

SPARRING PROMPT

Use this after writing your own account of the work:

I’m sharing an AI-assisted deliverable. Below is my account of what AI contributed, what I checked or changed, and what remains uncertain.

Challenge the account. Where am I claiming more review than I can substantiate? What would a recipient still need to know to assess the result? Which claims need evidence?

Ask me questions before suggesting wording. Don’t invent checks, findings, or human contributions.

The prompt can help you spot gaps in your explanation. You still need to supply the evidence and make the account truthful.

ARTICLES

IBM: AI puts critical thinking at the center of workforce priorities
The survey highlights invisible work and accountability gaps. Read it with a question in mind: what does your team expect people to take responsibility for, and can they challenge it?

NN/g: The three roles of context for AI agents
A study of Claude power users distinguishes global, local, and ambient context. It makes another contribution worth examining visible: the information and guidance that shape what an agent produces. Deciding what belongs in that context is part of the work.

NN/g: When should you disclose AI use? The PACED framework
Audience, context, expectations, policy, and the degree of AI involvement all matter. Disclosure doesn’t consistently increase trust. Use it to explain the contribution accurately, while recognizing that people may respond differently to the same explanation.

ONE THING TO TRY

At your next review, ask someone to explain one consequential change they made to an AI-assisted output and why they made it.

Listen for the reasoning behind the change. In the researcher’s case, returning to the transcripts restored distinctions and emotional context. Understanding that intervention gave his manager a clearer view of his rigor.