Key Takeaways

  • SparkToro analyzed 11,800 real ChatGPT prompts from more than 12,000 opted-in users and found that the share of conversations actually relevant to marketers is far smaller than vendor pitches suggest, which means your AI visibility budget should be sized to reality, not to the 900 million weekly active user headline.
  • OpenAI's own September 2026 breakdown shows 49% of messages are Asking, 40% are Doing, and 11% are Expressing, and writing dominates the Doing bucket, so most ChatGPT time is personal productivity rather than brand discovery.
  • Information seeking has climbed to 20.2% of consumer messages, near writing at 21.4%, which is the slice where AI visibility actually matters and where a CMO should concentrate spend.
  • OpenAI has started giving brands limited data on how they appear in ChatGPT, which will reset the GEO vendor market faster than most CMOs are planning for.
  • The right move right now is a small, measured AI visibility pilot tied to the information-seeking slice of your category, not a full GEO retainer built on vanity usage numbers.

For about eighteen months now, every pitch deck landing in a CMO's inbox has opened the same way: ChatGPT has 900 million weekly active users, your buyers are there, you need an AI visibility strategy today. The number is real. The leap from that number to a line item in next year's budget is where a lot of marketing leaders have been stuck, because nobody has shown them what people are actually doing inside those sessions.

That's the quiet discomfort in every one of these conversations.

Last week Rand Fishkin's team at SparkToro published the first study I've seen that tries to answer that question with real prompts instead of survey recall. They looked at 11,800 ChatGPT prompts from more than 12,000 opted-in American users, and the headline, for marketers, is that most of what happens inside ChatGPT has nothing to do with us. That's not a reason to ignore it, it's a reason to size the bet correctly.

My read is that this study, paired with OpenAI's own usage report and the new brand-visibility tool they just shipped, finally gives CMOs enough signal to stop guessing. So let's walk the numbers, say what they mean, and talk about what a reasonable marketing leader should actually do on Monday.

What's actually in those 11,800 prompts

The useful thing about the SparkToro sample is that it captures what people type, not what they remember typing. People lie to surveys about how they use AI, usually in the direction of sounding more sophisticated than they are. Prompts don't lie. And when you look at the prompts, the vast majority are personal: homework help, recipe ideas, relationship advice, code snippets, travel plans, writing a sympathy card. The marketing-relevant slice, the queries that look like someone evaluating a product or researching a vendor or comparing brands in a category, is a real slice, but it is a slice.

That lines up with what OpenAI's own team found when they classified consumer intent in September 2025: 49% Asking, 40% Doing, 11% Expressing, with writing dominating the Doing category. Writing is drafting an email, finishing a paragraph, cleaning up a note. Most of that activity is personal productivity rather than anything resembling purchase intent. If you're a B2B tech CMO reading those percentages, the honest interpretation is that maybe one in five ChatGPT sessions touches the kind of research behavior your buyers do, and even inside that twenty percent, only a fraction is about your category.

I want to be careful here, because I don't want to tell you AI visibility doesn't matter. It does, and it's going to matter more. The point is that the calibration everyone has been using, buyers live in ChatGPT now, is wrong at the budget level. A more accurate frame is that a meaningful minority of your ICP occasionally uses ChatGPT for category-adjacent research, and that minority is growing.

The slice that actually matters: information seeking

Here's the number I'd anchor a CMO's thinking on. OpenAI's October 2026 signals report says information seeking has grown to 20.2% of consumer messages, close to writing at 21.4%. A year ago, writing was running away with it. Now information seeking is pulling even, which is exactly the behavior shift that should matter to marketing.

Information seeking is the slice where somebody asks ChatGPT what the best vendors are, how two tools compare, what people are saying about a product, whether a company is credible. That's your window. It's not every session, it's not even most sessions, but it's the fastest-growing one and it's the one where your brand either appears or doesn't. If I'm sizing an AI visibility investment, I'm sizing it against that twenty percent, not against the full 900 million.

And then I'd ask one more question: inside that twenty percent, what fraction is my category? For most B2B tech companies, the answer is small. That's fine. If the category slice is small but trending up, that's a good reason to plant a flag early and cheaply. If it's small and stable, you still want to be there, but the right investment is modest and patient rather than a big front-loaded bet.

OpenAI just moved the goalposts on GEO vendors

While we were all reading the SparkToro study, OpenAI quietly shipped something that changes the market for generative engine optimization. According to coverage from last week, OpenAI is now giving brands some data about how they appear in ChatGPT, directly competing with the sprawling industry of GEO specialists who built their pitch on being the only source of that visibility.

This is the Google Search Console moment for AI visibility. For years, SEO vendors sold ranking reports until Google gave the data away, and then the market shifted from reporting to strategy. The same thing is going to happen here, probably faster, because the AI market is compressing timelines on everything. If I'm a CMO evaluating a GEO retainer right now, my question for the vendor is simple: what do you do that OpenAI's own tool won't do in six months? If the answer is a dashboard, that's a no. If the answer is a repeatable method for showing up in the information-seeking slice, that's worth paying for.

What I'd actually do on Monday

Here's the pivot I keep offering the marketing leaders I talk to. Stop sizing your AI visibility investment against total ChatGPT usage. Size it against the slice that touches your category, and treat the whole thing as a measured pilot for the next two quarters while OpenAI's own tooling matures.

Concretely, that means three things. First, run your category's most likely information-seeking prompts, maybe fifty of them, through ChatGPT today and write down what you see. Who shows up, who doesn't, what sources it cites, where you're mentioned and where you're not. That gives you a baseline, and it costs you an afternoon. Second, make sure the handful of sources that ChatGPT actually pulls from in your category, which is usually a short list of analyst sites, Reddit threads, review platforms, and a few trade publications, have accurate information about you. That work moves the needle more per dollar than any GEO retainer I've seen. Third, revisit this in ninety days with the same prompts and see what moved.

That's it. That's the pilot. If you see real movement in the information-seeking slice, you now have evidence to justify a bigger investment. If you don't, you've spent an afternoon instead of a six-figure retainer. If you want a structured version of this measured, evidence-first approach to AI in your go-to-market, the growth blueprint walks through it.

When this approach is the wrong call

I want to be honest about where this reasoning breaks down. If you sell directly to consumers in a category where ChatGPT is already the default research tool, like travel planning, recipe apps, or study tools, the SparkToro data probably understates your exposure, because those are exactly the personal-use categories that dominate the sample. You should move faster and spend more than the measured pilot I just described.

And if you're a very early-stage B2B company where any mention anywhere is worth something, the cost of a small AI visibility push is low enough that the calibration conversation is almost academic. Do it. The ROI math only matters once you have real budget at stake.

The companies I'd slow down are the mid-market B2B tech companies being pitched $15,000 a month GEO retainers based on the 900 million number. That's the place where the SparkToro data should make you ask harder questions before you sign.

Frequently Asked Questions

How much of ChatGPT usage is actually relevant to marketers?

A minority of it. SparkToro's analysis of 11,800 real prompts shows most ChatGPT usage is personal productivity, writing help, and information seeking on non-commercial topics. OpenAI's own data puts information seeking at about 20% of consumer messages, and only a fraction of that touches any given category, so the marketing-relevant slice is real but much smaller than vendor pitches suggest.

Should CMOs invest in AI visibility optimization right now?

Yes, but as a measured pilot rather than a full retainer. Run fifty category-relevant prompts, baseline where your brand shows up, fix the source sites ChatGPT actually cites in your category, and revisit in ninety days. That's a cheap, evidence-first approach that respects how early the data still is.

What does OpenAI's new brand visibility tool mean for GEO vendors?

It compresses the market. GEO vendors who sell dashboards and reporting are about to compete with free native tooling from OpenAI, similar to what Google Search Console did to SEO reporting tools. Vendors who sell repeatable methods for showing up in the information-seeking slice will keep their value.

How does information seeking compare to writing in ChatGPT usage?

As of September 2026, information seeking is 20.2% of consumer messages and writing is 21.4%. A year earlier writing was dominant, so information seeking is the fastest-growing bucket and the one most relevant to brand discovery and consideration.

How reliable is this data?

It's getting there. The SparkToro study and OpenAI's own usage reports are the first datasets based on real prompts rather than survey recall, which is a big step forward. The honest limitation is that both samples skew consumer, and dedicated B2B research behavior inside ChatGPT is still underdocumented, so treat current numbers as directional while you build your own baseline.

Sources