Your Users Are About to Ask "Who Made This?" Have an Answer Ready

A few stories landed in the same week, and they rhyme. Google changed a setting so your uploads train its AI. Reddit is using AI to fight AI spam. Midjourney is in court arguing the studios suing it do the same thing behind closed doors. Different fights, same root question: who made this, with whose data, and did anyone say yes?

Your users are starting to ask that out loud. Let me catch you up on what changed and what you should decide before your next review.

Consent got moved, not offered

Google updated its Search services privacy settings in June and opted people into wider AI training. Upload a photo to Google Lens, talk to Search Live, practice with Translate, and that media can be saved to train its models. TechCrunch's Sarah Perez walked through the opt-out: two new settings, a "Save Media" box, retention you can set to 3, 18, or 36 months.

The framing was "more control." The default was on. That gap is the story. If you split a privacy toggle into two and leave the new one checked, you got consent on paper, not in spirit.

When your users find out, and they will, the anger is not about the training. It is about being enrolled without a clear ask. Look at your own defaults this week.

The label is becoming the product

People now look at a poster, a video, a UI and wonder if a machine made it. That question barely existed two years ago. Allie Paschal calls the answer "creative provenance": proof of who made it, what tools were used, and how.

"Made by humans" is starting to work like "Made in the USA," a differentiator you print on the box. Campaigns show storyboards and rough cuts as evidence a person spent real time. The process is now part of the pitch, not a slide buried in a case study.

But provenance is not one-size-fits-all. Nobody cares who designed the nav bar as long as checkout works. They care a lot if a film they loved turns out to be AI. Paschal draws a line from expressive work, where the how is the meaning, to functional work, where it barely registers. Know where each of your surfaces sits before you slap on a label.

Disclosure only counts if you can back it

Here is the trap in showing your work: the work can be staged. Paschal flags it plainly. Behind-the-scenes content only shows what the creator wants you to see. Proof can be manufactured too. So a "no AI" claim you cannot defend is worse than no claim at all.

The middle ground is where product teams live. When AI is embedded in the experience, making a recommendation or writing a summary, users want to know its role. IBM's Carbon system puts an AI label right in the UI and lets people click through for detail on how output was generated. That is disclosure you can stand behind, tied to the exact moment AI touched the screen.

Everyone's hands are a little dirty

The cleanest version of the trust problem is Midjourney's court fight. Disney, Universal, and Warner Bros. sued it for copyright infringement, calling it a "bottomless pit of plagiarism." Midjourney's answer, per Variety by way of Mashable, is to demand the studios disclose their own AI training data, model weights, and board decks. Its lawyer's argument: if you are "doing the very thing they seek to punish," that matters.

The studios call it a fishing expedition. Maybe. But the move works because it is plausible. Plenty of companies suing over AI are also building with it. That is the credibility gap your users now assume by default.

So when you make a public claim about how you use AI or data, expect someone to check your internal practice against it. The reputational risk is not the tool. It is the daylight between what you say and what your teams actually do.

The deep cut

Reddit is now using LLMs to catch the spam that LLMs created. It blocks 23 million spam views a day and cut user exposure to spam 20% from January to March. Good numbers. But the quieter lesson is in how it works. Reddit pairs the models with human review, because detection alone gets you false flags and missed context.

That is the practical payoff across all of this. Whatever you automate, disclose it and put a person in the loop for the judgment calls. A toggle without a clear ask, a "no AI" badge you cannot prove, an AI moderator with no human check, those all fail the same way. They break trust the moment a user looks closely. Build the disclosure and the human backstop now, while it is a choice, not after a screenshot goes around.

Three questions for your team

1. Where in our product is a data or AI setting opted-in by default, and could we defend that default out loud to a user who just found it?

2. For each surface where AI touches the output, do we show its role in the UI, and can we prove any "human-made" or "no AI" claim we make?

3. If a reporter compared our public AI stance to what our teams actually do internally, where is the gap, and who owns closing it?