What the OpenAI entrepreneurship report didn't say

OpenAI published a report this month called Empowering Entrepreneurship in ChatGPT. The headline number is good: by their conservative estimate, at least four million people used ChatGPT to start, plan, run, or grow a business in a single month. That's roughly one in fifty active ChatGPT users — across retail, services, trades, health, education, and the long tail.

It's a confidence-building stat for anyone using AI tools to run something small. We're not outliers. Four million of us are doing the same thing.

And then the report stops being useful.

What the report says

The report's most-quoted breakdowns are by industry (retail, professional services, trades, health, beauty) and by activity (marketing, branding, operations, customer service). Both are the top of the funnel. The single biggest activity, at 26%, is marketing and copywriting. The next is branding and identity, at 9 to 15% depending on cohort.

That much is worth knowing. The model is doing what marketers and brand consultants used to do, for free, all day, for four million people.

What the report doesn't say

The report is carefully silent on five things, and the silence is the point.

It doesn't say what worked. The methodology uses privacy-preserving classifiers to count messages. There's no visibility into which businesses earned a pound, which ones stalled, and which ones never made it past a slide deck. Four million people used ChatGPT for business. The report cannot tell you how many of them have a business.

It doesn't break down digital products. The industry buckets — retail, professional services, trades — are offline-shaped. Solo creators selling PDFs, prompt kits, Notion templates, AI music, faceless YouTube channels, and digital downloads don't have a column in the chart. They're absorbed into "professional services" and rounded out of view. The single largest growth lane in the AI economy doesn't appear as a category.

It doesn't quantify revenue. No median earnings, no time-to-first-sale, no conversion rates, no customer acquisition cost. The closest the report comes is observing that "active entrepreneurs are disproportionately represented on higher-tier plans". In other words: the people earning enough from AI tools to justify £20 a month for ChatGPT Plus are the cohort doing it seriously. That's the only revenue signal in the document.

It doesn't name a single person. No case studies. No founders. No real names. The whole document is in the third person plural — "entrepreneurs do X", "founders use Y" — which is exactly how a corporate report avoids the awkward question of which specific X and which specific Y actually paid the rent.

And it doesn't explain why most of those four million will earn nothing.

Why most of the four million won't earn

I don't have OpenAI's data, but I can tell you what I see in operator forums, indie-hacker communities, and the people who actually email me. The pattern is the same every time. The 4 million includes:

• People who asked ChatGPT to "write a business plan for my idea" and have done nothing since.
• People who generated a brand identity, designed a logo in Midjourney, and stopped there.
• People who drafted a Substack post a month for six months and never sent the first one.
• People who built a Notion template, didn't list it anywhere, and called the project finished.
• People who watched a "how I made $10K with AI" YouTube video and ran the same prompt.

None of this is an insult. We've all been in at least one of those categories at least once. The reason most of the four million won't earn isn't laziness. It's that the model produces output, and output isn't the product. The product is a finished, distributed, pricing-decided, checkout-wired, customer-supported, kept-up-to-date thing. ChatGPT can write every word of a business; it cannot ship one.

The operators who do earn — and four million is enough that there are tens of thousands of them — share one feature. They treat the AI as a layer in a system they're running, not as the system itself. They know which layer changed when the output gets worse. They close the loop themselves.

What to do with the four-million number

Use it as motivation, not as evidence. The number is real. The market is real. The opportunity is real. What the OpenAI report can't give you is the operating discipline that turns "I used ChatGPT for my business" into "I run a business that uses ChatGPT".

That's what I write about here, and that's what the AI Operator Starter Pack is built around. Three lessons on what's actually under the hood (model, harness, layers), four exercises that turn vague AI frustration into evidence, a printable reference card for the wall, and ten operator-grade prompts for the work that matters. £9 for the first 100 readers; £12 afterwards.

— Kol Tregaskes, AI Resource Hub. Published 2026-05-05. OpenAI's full report is at openai.com.