Society's New Purity Test

Last week, a client asked me a question I couldn't stop thinking about: “How much of this was put together with AI?”

So I wrote about it. That issue went out Monday morning.

Monday afternoon, news broke that Anthropic was working on ways to make AI-generated text easier to identify.

Apparently, we're all thinking about the same thing.

And I understand why. AI is now good enough, common enough, and embedded deeply enough in the way people work that we're trying to figure out what its involvement means. Schools are wrestling with it. Publishers are wrestling with it. Businesses are wrestling with it. Your customers probably are too.

But I think we're reaching for the wrong answer.

We want a clean test: Was AI involved? Yes or no.

The problem is that knowing AI was involved doesn't tell you what you actually want to know.

The New Purity Test

Publishing has already run headfirst into this problem. Books have faced backlash and scrutiny because readers or publishers suspected AI had a hand in creating them. The conversation quickly stops being about whether the work is good and becomes about whether it's pure — whether a machine touched it.

Now the technology is starting to catch up with that instinct. Companies are developing ways to identify or mark AI-generated content so that, in theory, we can know when AI was involved.

There's an obvious appeal to that. A yes-or-no answer feels useful.

But imagine two people writing an email to their customers.

The first types a prompt into AI, copies whatever comes back and hits send without giving it much thought.

The second has an idea, uses AI to help research it, challenge the argument and rough out a draft, then spends an hour deciding what's true, what's useful, what sounds like them and what needs to go.

AI was involved in both.

A test designed to identify AI involvement may tell you that. What it can't tell you is what happened after AI showed up.

And that's the part that matters.

“Is This Human?” Is the Wrong Question

I think we're starting to confuse two very different questions:

Was this made without AI?

And:

Was there a human thinking?

The first one is relatively easy to understand. The second is much harder, because it isn't yes or no. It's a long gray line.

On one end, someone ran a prompt and published whatever came back.

On the other, someone used AI to help gather information, explore ideas, challenge assumptions or build a first draft, then brought their own experience and judgment to everything that followed.

Most AI-assisted work lands somewhere in between.

A purity test can't tell you where something falls on that line. At best, it can tell you that

AI entered the process somewhere along the way.

And I don't think that's the part we should be worried about.

Your Customers Already Have Their Own Test

Your customers aren't going to run a watermark detector on your next email, proposal or Facebook post.

But they're doing their own version of the test.

They're reading what you send and deciding whether it sounds like you understand them. They're deciding whether the answer actually fits their situation. They're noticing when something feels generic, when a claim doesn't quite make sense or when a business they've trusted for years suddenly sounds like every other business on the internet.

They probably won't think, AI wrote this.

They'll just think, this isn't very useful.

That's why I keep coming back to the question my client asked last week. “How much of this was AI?” was a completely fair question. But the percentage wouldn't have told him much.

AI helped me get there faster. It helped with research, organization and drafting. But the value wasn't in whether AI touched the document. The value was in the decisions about what belonged in it.

That's the distinction I think we're going to have to get better at making.

The goal isn't to prove that AI never touched your work.

The goal is to make sure your judgment did.

Put It to Work

Pull up the last three things AI helped you create — a quote, a post, an email, a plan. Don't ask whether AI wrote them. Ask what you added after AI got involved.

What did you question? What did you change because you know the customer better than AI does? What did you cut because it sounded right but wasn't true? What decision did you make that the machine couldn't make for you?

If you can't point to much, that's useful information. The answer isn't to stop using AI. It's to move yourself further into the process.

Try giving AI one of those pieces again and ask:

“Don't rewrite this yet. Help me evaluate it. Ask me five questions that would reveal where this is generic, where I'm making assumptions, and where my own experience or knowledge of the customer should change what I've written. Ask the questions one at a time and use each answer to inform the next.”

Then answer them.

Once you've worked through all five, ask:

“Based on my answers, show me the three places where my judgment should have the biggest effect on this piece. Explain why before you suggest any changes.”

That's a very different way to use AI. You're not asking it to replace the thinking. You're using it to make sure you actually do some.

AI can help you research, draft, organize, challenge and refine.

Your job is still to decide.

One Question for You

I have a feeling we're going to be talking about this distinction a lot more over the next few years.

So I'm curious: Does it matter to you if something you read was created with the help of AI? Or does it only matter when you can feel that nobody thought very hard about it?