Danielle Okafor has been making a living as a freelance content strategist and essayist for eleven years. She's worked with indie labels out of Pilsen, tech startups in the West Loop, and brands that have no business asking a writer to also be their art director. She's good at all of it. She started using AI tools about two years ago, quietly, the way most honest people do. She told almost no one. Last fall she started telling everyone... and then watched what happened to her engagement numbers. We talked for about an hour at a coffee shop on Milwaukee Avenue. This is the part she let me print.
Let's start with the obvious question. Why did you hide it in the first place?
Because I watched what happened to other people who didn't. There was a photographer I followed, really talented, shoots a lot of editorial stuff. She mentioned in a caption that she used Midjourney for a mood board before a shoot. Not in the final image. A mood board. The comments were brutal. People saying she wasn't a real photographer anymore. She lost brand partnerships. And the thing is... nothing she actually delivered changed. The final photographs were hers. Shot by her. Edited by her. The AI touched nothing the client received. But the perception was poisoned.
So when I started using AI to rough out structure on longer pieces, or to generate a handful of headline options when I was stuck... I just didn't say anything. Not because I was ashamed of the work. Because I'd seen the math. Honesty had a cost I couldn't afford that month.
The platforms pushing disclosure frameworks... do they understand that math?
They understand it exists. I don't think they care about solving it. The frameworks being proposed right now mostly fall into two buckets: AI-assisted or AI-generated. And that binary is so useless it's almost funny. What does AI-assisted even mean? I used Grammarly. I Googled a fact. I ran a paragraph through Claude to see if the logic held. All of that is assistance. None of that is the same as handing a prompt to a model and publishing whatever comes back.
The truth is there's a spectrum and nobody wants to draw a line on a spectrum because wherever you draw it, someone gets hurt. So instead you get these vague labels that don't protect audiences and don't protect creators. They just create a new thing to be accused of.
You mentioned your engagement dropped after you started being more open about your process. How bad was it?
On one platform, about 30 percent reach drop over six weeks. Which, if you're monetizing directly off that, is a real number. It's not catastrophic but it's not nothing. A piece I was genuinely proud of... thoughtful, specific, took me three days... got about half the shares of a piece I'd written two years earlier that I think is actually weaker work. The only real difference was context. People knew more about how the newer piece was made.
And here's what gets me. The older piece... I used a thesaurus obsessively writing that one. I read four articles before starting. I talked through the thesis with a friend over the phone. Nobody asks about any of that. The inputs we've decided are acceptable are invisible. The inputs that involve software with a chat interface are suddenly a moral failing.
So is the problem the label itself, or the culture that makes the label dangerous?
Both. But you can't fix culture with a policy framework. You can't write guidelines that make people less afraid. What you can do is design a system where honesty doesn't cost you algorithmically. Right now the platforms are asking creators to be transparent and then their own recommendation systems punish that transparency. That's not a disclosure problem. That's a design problem. And nobody in the rooms where those decisions get made is a freelance essayist who's behind on rent.
The people designing these frameworks have institutional protection. They work for universities or think tanks or platforms with enough scale that their individual reputation can absorb the hit. A solo creator does not have that buffer. So the framework gets designed for a hypothetical honest person who doesn't have to pay for their honesty.
What would actually work? If you could design the system from scratch.
Honestly... I think it starts with separating process transparency from output labeling. Those are two different things and we keep conflating them. An audience has a legitimate interest in knowing if the thing they're reading or looking at was substantially generated by AI with minimal human shaping. That's fair. That's a real question about what they're consuming.
But process transparency... how I got to the final thing... that's a different conversation. And it should be optional. Not because creators should hide things, but because nobody asks a novelist which sentences they wrote drunk and which ones they wrote sober. Nobody asks if the breakthrough happened in the shower. The process has always been private. We only started demanding it be public because a specific kind of tool got involved.
So the framework I'd want: clear labeling for substantially AI-generated content with minimal human editorial control. And then... let everything else be a conversation the creator chooses to have or not have. Stop designing systems that punish nuance.
What do you say to the creators who are using AI heavily and not disclosing at all?
I say I understand why, and I also think they're making a bet they'll eventually lose. Not because disclosure will become mandatory... maybe it will, maybe it won't. But because audiences are getting better at feeling the difference. Not identifying it, not proving it. Just feeling something slightly hollow. And that's the reputational damage that actually sticks.
Nobody tells you this but... the disclosure conversation is almost a distraction. The real question is whether what you're making has something in it that came from your actual life, your actual thinking, your actual point of view. If it does, no label will kill it. If it doesn't, no absence of a label will save it. The work either has a heartbeat or it doesn't. AI can help you find the structure. It cannot hand you the heartbeat.
Last one. Are you still using AI in your process?
Yes. Less than I was at peak, because I got lazy for a while and noticed it in the work. More thoughtfully now. I use it the way I'd use a smart friend who reads a lot and has no ego about being wrong. I ask it hard questions about my argument. I use it to stress-test logic. Sometimes I'll generate ten variations of an opening line just to see what territory exists, and then I throw all ten away and write an eleventh that's actually mine.
Do I label that? No. And I'm not sure I should have to. That feels like being asked to cite every conversation I've ever had that shaped an idea. At some point the work is either mine or it isn't. I think mine is. I'm living with that answer.