“AI-Generated” Doesn’t Tell You Who Made the Creative Decisions

“AI-generated” is a useful label — but an incomplete one

If generative AI helped create an image or video, calling it AI-generated can be factually accurate. But that label leaves out a different question:

Who actually made the creative decisions?

In Pawsitive Messages, every finished video starts with something the AI did not invent: a real photo supplied by the pet parent.

From there, the creative process becomes much more collaborative. The pet parent:

• supplies the subject
• defines meaningful context about the pet
• decides what message feels right
• can change the words
• decides whether the idea is worth continuing
• makes visual choices
• decides which result becomes the finished video

AI is deeply involved. So is the human.

We wanted to see how much of that human direction shows up in the audit trail

Not every creative choice is captured equally in historical product data. But across our dataset of finished videos, we looked for documented signs that the human continued shaping the work after the first generated message appeared.

The result was striking: 77% of the videos we analyzed showed at least one documented post-generation human action.

That is a lower-bound observation, not a complete count of every creative decision. Some choices available in the product were not consistently logged across every version of the experience.

Editing is only one kind of creative participation

One of the easiest human actions to see in the data is editing. Roughly 24% of those finished videos carried a recorded user edit. That means the generated wording was not simply accepted as-is.

But editing is only one form of authorship-like creative direction. A person can contribute meaningfully without rewriting a sentence. Other creative decisions include:

• choosing one idea over another
• deciding which scene to continue
• choosing layout, color, or message placement
• deciding something is not right and asking for another version

The absence of a text edit does not mean the human was absent from the process.

The first AI output often was not the final creative choice

In newer Pawsitive experiences, people may be shown multiple message options. Among the finished videos that began with multiple message candidates, at least 62% used something other than the default first option.

That number is intentionally conservative. Some finished videos that kept the first option may still represent an intentional choice — the audit trail simply cannot distinguish every one of those cases from a default state.

What it does show is that a large share of finished work moved beyond the first available AI answer.

Choosing among options is itself a creative act. See what distinguished the birthday messages people chose to keep →

The source material matters too

Every Pawsitive video begins with a user-origin photo of that person’s pet. That is not a small detail.

The AI is not inventing an anonymous dog or cat from scratch and asking someone to adopt it as their own. The pet already exists. The relationship already exists.

The photo, identity, personality, and emotional context come into the process before generation begins. The technology helps transform those inputs into a message, scene, and video. But the subject of the work originates with the human.

Creative provenance is more useful than a binary label

A simple label can tell you: AI participated.

It cannot necessarily tell you:

• who supplied the subject
• who defined the context
• who chose among alternatives
• who changed the words
• who made the design decisions
• who decided what moved forward
• how much human judgment happened between the first generation and the finished piece

Those are provenance questions. And they matter because creation is often a process, not a single button press.

This does not make the AI irrelevant

Quite the opposite. AI is central to how Pawsitive Messages works. It helps generate language, visual scenes, and finished creative output.

The point is not to pretend otherwise. The point is that “AI-generated” describes the technology involved more clearly than it describes the entire creative process.

A finished piece can be AI-assisted and still contain substantial human direction. Both things can be true.

What this means when people call something “AI slop”

“AI slop” has become shorthand for content that feels generic, disposable, or generated without meaningful human intention. That criticism is sometimes fair.

But the presence of AI alone does not tell you whether a human made meaningful creative choices. A better question is:

How much intention sits between generation and the finished work?

That is what the provenance trail starts to reveal.

About the data

We analyzed aggregated data and studied patterns. No individual person or pet is identified.

We examined a cohort of finished Pawsitive Messages videos with documented creation details and looked for post-generation human actions such as editing, choosing among message alternatives, and continuing a creation into later stages. The product has evolved over time, and not every historical experience logged the same actions in the same way.

For that reason, the percentages reported here should be treated as minimum documented participation, not a complete measurement of every human creative decision.

These findings describe behavior inside Pawsitive Messages. They do not establish legal authorship, copyright ownership, or a universal definition of human-created versus AI-created work.

What we learned

A binary “AI-generated” label can tell you that generative technology was involved. It cannot tell you the full creative history of the finished piece.

In Pawsitive Messages, the work begins with human-origin source material and often continues through human selection, editing, design choices, and decisions about what becomes final.

The more useful concept is not:

Was AI involved?

It is:

What was the creative provenance of the finished work?