AI Can Help Me Write. But I Still Have to Live the Story.

I use AI as a thinking partner every day.

I bring it unfinished ideas, real experiences, and questions I am still trying to answer. It helps me explore, challenge, organize, and sometimes find the words for something I already know I want to say.

Sometimes that means bringing a real experience from my life or work and asking for help shaping it into a story. Sometimes it means asking for a first draft that gives me something to react to.

But AI has not lived my experiences or decided what matters to me.

I add the experience. I decide what is true. I reject sentences that do not sound like me. I remove ideas that are clever but not honest. I keep working until I can stand behind the piece.

That is a healthy balance for me.

I am not interested in generic AI writing where someone enters a broad prompt, accepts the first response, and publishes words that could belong to anyone. That is not why I write.

I could have written this article without AI.

It would have taken me longer. I would have spent more time staring at a blank page, organizing my thoughts by myself, and trying to find the right first sentence. I also would not have had an immediate thinking partner to challenge an idea, help me see another angle, or give me a first draft to react to.

AI does not replace the writer in me. It helps me get to the writing faster.

That matters because a quicker first draft gives me more time for the part that matters most. I can decide what is true, make the message clearer, add the experience that only I have, and keep working until the piece feels useful and honest.

For me, this is not only about efficiency. It is a fulfilling way to think. I can bring a rough idea into a conversation, test it, disagree with it, and discover what I actually want to say.

It also helps me finish and improve more of the thoughts I already have. Not to fill LinkedIn with more noise, but to share useful ideas that might help more people think clearly about AI, work, and the decisions in front of them.

That is why the recent conversation about AI text watermarking caught my attention.

My first reaction was personal. I wondered whether a watermark might become a shortcut for judging a piece of writing.

AI was involved.

Therefore the writer did not really write it.

Therefore the idea, experience, or effort behind it might not be real.

That is too simple.

At the same time, I do not think watermarking itself is a bad idea. AI can produce spam, impersonation, fake reviews, fraudulent messages, and an enormous amount of AI-generated material designed to mislead people. We need better ways to understand where AI may have entered the picture.

The important question is what a watermark can actually tell us.

A watermark cannot tell us who owns the work

Anthropic has announced that future Claude models will include an invisible, machine-readable watermark in generated text. It says the watermark can help determine the likelihood that Claude was involved in producing the text. It does not identify a user, organization, or conversation. Anthropic explains the approach in detail.

That distinction matters.

A watermark can answer a narrow question.

Was a particular AI model likely involved in producing some of these words?

It cannot answer the much larger questions people often care about.

• Who had the idea?
• Whose experience is being described?
• Who decided what was true?
• Who reviewed the work?
• Who is responsible for the final message?
• Who is willing to put their name behind it?

Those are questions of ownership, responsibility, and human judgment. A text watermark is not designed to answer them.

How an AI text watermark works

The basic idea becomes easier to understand when we start with how a language model writes.

A model produces text one word or piece of a word at a time. At many points, there may be several sensible ways to continue a sentence.

Imagine a model has written:

The meeting was productive, and the team left with a clearer…

Several next words may work. It could choose “plan,” “direction,” or “understanding.” The meaning may remain very similar.

Watermarking uses choices like these. Instead of allowing the system to choose among acceptable words in its usual way, a secret mathematical key subtly influences those low-stakes choices. Across a long enough piece of writing, those choices create a pattern that a detector with the right key can recognize.

The reader does not see a stamp. No extra characters are added. Anthropic says the intended result should not change the meaning, quality, or readability of the text.

I think of it as a pattern in how a tool made many small choices, not as a signature from the person who used the tool.

That is useful information. It is also limited information.

A watermark cannot see the work behind the words

Anthropic is clear about this limitation. Its watermark can indicate that Claude was likely involved, but it cannot distinguish between Claude writing a passage and Claude heavily editing a passage. It also says the watermark does not determine ownership or legal responsibility.

That means a watermark cannot see the creative process behind the words.

Consider two people creating a LinkedIn post.

The first person enters a generic request, copies the response, and publishes it without much thought.

The second person begins with a real experience. They explain what happened, why it bothered them, what they learned, and where they are still uncertain. They review several drafts. They remove sentences that do not feel true. They add details that only they know. They take responsibility for the final piece.

AI may have been involved in both processes.

But they are not the same process.

The watermark may not be able to tell the difference. That does not make the difference unimportant.

We need a healthier way to talk about AI-assisted work

The cultural conversation often gives us only two categories.

• Human-written
• AI-generated

I think we need another category.

Human-owned, AI-assisted

This does not mean pretending AI had no role. In many cases, it may have helped shape the structure or produced a meaningful share of the sentences.

It means recognizing that ownership involves more than physically typing every word.

Ownership can include supplying the lived experience, directing the thinking, making editorial choices, checking what is true, rejecting what does not fit, and accepting responsibility for what is published.

The person using AI still has work to do. In fact, as AI becomes more capable, the work that remains may become more important.

The question I ask myself

I do not think the most useful question is:

Did AI touch this sentence?

The more useful question is:

Did I do the thinking behind this sentence?

That is the boundary I want to protect.

If I cannot explain an idea without reopening the AI conversation, if I do not believe the conclusion, or if I would be uncomfortable defending the piece in front of another person, I should not publish it under my name.

AI can help with craft. It should not become a substitute for judgment.

This is also why I do not believe good AI use means hiding the tool. It means using the tool deliberately while remaining honest about what it can and cannot do.

What this means for business owners

For a local business, a watermark should not become a lie detector or a shortcut for deciding whether a message can be trusted.

A more useful approach is to ask practical questions.

• What information did the AI receive?
• Is the final message accurate?
• Has someone with the right knowledge reviewed it?
• Does it make promises the business can keep?
• Does it reveal information that should remain private?
• Is disclosure appropriate for this audience or use case?
• Who is accountable if the message is wrong?

These questions matter whether the work is a marketing post, a customer email, a proposal, an internal summary, or a policy.

In higher-stakes situations, the business may need stronger review, clearer documentation, and more direct disclosure. In lower-stakes situations, a person may simply use AI as a drafting partner and review the result carefully before sending it.

The right standard depends on the work. But responsibility should not disappear because a useful tool was involved.

Transparency should make us more thoughtful, not more cynical

I support efforts to make AI involvement easier to understand. The scale of synthetic content makes that necessary.

But we should resist using a technical signal as a complete judgment about a human being.

A watermark can tell us something valuable about a tool.

It cannot tell us who had the experience, who did the thinking, who owns the message, or who is willing to stand behind the work.

Those questions still belong to people.

And I think they will matter even more as AI becomes part of everyday work.