When "AI-Assisted" Becomes Meaningless

Amazon now requires authors to declare whether their work is “AI-assisted” or “AI-generated.”

Academic journals demand disclosure statements. Publishers are building entire attribution frameworks. LinkedIn is having daily debates about where the line sits.

I keep thinking about that line.

Because the more I look at it, the less convinced I am it exists.

The Taxonomy Problem

Picture your Marketing Manager drafting a campaign brief.

They prompt Claude with the core requirements. Claude returns a structure. They rewrite the introduction entirely, keep most of the middle section, change three examples to better fit the brand voice, and add two paragraphs of market context they know from experience.

Is that assisted or generated?

Now picture the same scenario in reverse. They write the brief themselves, every word, then run it through Grammarly for punctuation, ask Claude to “make this more concise,” and accept 80% of those suggestions.

Which category now?

The distinction feels increasingly arbitrary. We’re trying to draw precise boundaries around something inherently fluid.

And whilst we’re all arguing about percentages and thresholds, 55% of US workers are using unapproved AI tools (Netguru). They’re not waiting for our definitions or taxonomy. They’re not checking attribution boxes. They’re just working.

What We’re Actually Measuring

The “assisted versus generated” debate was always a proxy for something else.

We’re trying to measure quality. Accuracy. Originality. Accountability.

The problem is we’ve attached those concerns to methodology rather than outcome.

Your Head of Content should be able to look at any piece of work (regardless of how it was created) and tell you whether it meets your standards. If they can’t, you have a quality problem, and AI is exposing it rather than creating it.

The UK’s five core AI principles are Safety, Transparency, Fairness, Accountability, and Contestability. They don’t mention percentages. They mention responsibility. There’s probably something in that.

The Accountability Question

When something goes wrong, nobody asks about your taxonomy.

When the client calls with concerns, when the regulator asks questions, when the mistake makes the trade press, they ask one thing: who signed off on this?

Your Finance Director using AI to draft board papers isn’t trying to game the system. They’re trying to manage an unreasonable workload with the tools available to them (that’s a separate conversation about resourcing.)

The question isn’t whether they used AI. The question is whether they understood they owned that output. Every word. Every claim. Every strategic recommendation.

If they thought the tool was responsible, you have a problem. If they knew they were responsible, the methodology becomes secondary.

The Pragmatic View

The UK ICO released its internal AI policy (and it’s not a bad template if you’re looking for one, though it does still miss delivering the point in my article here) in August 2025, emphasizing that any AI-assisted decision must include a mechanism for individuals to contest outcomes.

Notice what they’re measuring. Not percentages of human versus machine contribution. Accountability and redress.

Perhaps that’s the actual standard.

Can your team explain their decisions? Can they defend their outputs? Do they understand that “Claude suggested it” isn’t a valid excuse when something goes sideways?

Because the technology will keep evolving. Claude Sonnet 4.5 is vastly different from ChatGPT-3.5. Your percentage-based attribution framework will be obsolete before you finish writing it.

What won’t change is the need for someone to own the output.

Reflection

There’s something slightly uncomfortable about this realisation.

It means you can’t hide behind process. You can’t create an attribution form that transfers accountability to the tool. You can’t build a compliance framework that protects you from poor judgment.

You’re back to trusting your people. Their expertise. Their understanding of quality. Their willingness to put their name on something.

That’s either terrifying or liberating, depending on how much you trust your hiring process (and outsourcing this responsibility to AI is an equally hotly-debated topic).

The craft hasn’t changed. Good work still requires judgment, context, and accountability. The tools have just made it cheaper and faster to produce bad work that looks professional.

Which means the standards matter more, not less.

Make them explicit. Make them enforceable. Make sure your team knows they own what they output.

The machine is just a tool. The accountability is still human.