Most AI Use Isn’t What the Hype Promised

In partnership with

DEEP DIVE
Most AI Use Isn’t What the Hype Promised

This was all over the AI Twitter corner last week.

“OpenAI LEAK: how people really use ChatGPT”

Now, is it leaked when they released an official 63-page paper that probably took weeks to prepare? So we got:

• OpenAI 63-page paper analyzing ChatGPT conversations.

• Anthropic shared data about Claude’s usage around the world.

Buried inside the charts are insights that actually tell us a lot about what AI adoption looks like in the real world.

And honestly, they line up with what I’ve been feeling.

Let’s break it down.

Claude: Rich vs Poor Countries

Anthropic’s data shows an interesting split:

• In richer countries, Claude is mostly used for augmentation — collaborating with the model.

• In poorer countries, usage leans heavily toward automation — asking the model do the whole job.

From their paper:

“In countries with higher Claude use per capita, Claude’s uses tend towards augmentation, whereas people in lower-use countries are much more likely to prefer automation.”

Why does that matter?

Because it mirrors how work looks in different economies. In more complex, knowledge-heavy industries, people can’t just hand over control.

AI gives the best ROI as a copilot. Something that sharpens their work, not replaces it.

Meanwhile, in economies with more repetitive tasks, the appeal is the opposite: let the AI take the wheel.

That’s not just an academic point. It’s a reminder that how you use AI is shaped by the type of work you do, and why “AI is taking all the jobs” is a way too blunt take.

ChatGPT: The Real Usage Patterns

On the OpenAI side, we now have billions of conversations mapped.

Here’s what stands out to me:

• Search is king. If you add up “Seeking Information,” “Practical Guidance,” and “Technical Help,” you get about 60% of all ChatGPT usage.

I think this speaks a little to the AI adoption and what it truly looks like.

But also the fact that ChatGPT is not your one-stop shop AI and will never be.

Right now it’s more a Google++ and writing assistant.

• Writing is sliding. A year ago, 36% of conversations were about writing. That’s down to 24%. My take: people realized AI-written blogs and LinkedIn posts all look the same. The novelty wore off.

• Images blew up. Remember that Studio Ghibli wave back in April 2025? That wasn’t just your feed. OpenAI’s chart shows a massive spike in image generation right when GPT-4 with image tools came out.

There’s also an interesting split between “asking” vs “doing.” People ask ChatGPT for info, summaries, or ideas way more than they delegate full tasks. Again, it fits the theme: augmentation, not automation.

(Also, apparently about 23 people use ChatGPT to “Do” self-expression. I’d love to meet them, I have questions.)

My Takeaways

Here’s what all this tells me:

1. ChatGPT is mostly used as a “better” Google

   After 20 years of typing things into a search bar, it’s hard to change habits. ChatGPT isn’t reinventing work yet, but it’s slowly replacing the query box.

2. Augmentation is where the ROI is.

   Richer economies, more complex tasks, more augmentation. AI isn’t good at automating end-to-end workflows — not yet.

3. The writing hype has cooled.

   People are using AI more like an editor or assistant now — polish, summarize, rephrase — not as a ghostwriter. That’s actually healthier for adoption and to reduce AI slop.

4. Automation is still overhyped.

   Despite all the “AI agents will run your business” hype, the data shows people still prefer to stay in the loop. Full delegation isn’t here yet.

Why This Matters for You

If you’re running a service business, this is the practical takeaway:

• Stop chasing the dream of “AI runs everything.” That’s not where the value is (yet).

• Focus on augmenting: research, editing, summarizing, brainstorming. Those are the real gains today.

• Don’t buy every shiny SaaS that claims to automate your whole workflow. Most of them will fail the same way those MIT-tracked pilots failed.

• Keep it natural. The best AI adoption happens when it slips into what you’re already doing, not when you force it in because of a YouTube demo.

The data confirms what I’ve been saying for a while: AI isn’t about end-to-end automation, it’s about giving you leverage on the stuff you already do.

And that’s the only sustainable way adoption turns into ROI.