If I were OpenAI’s new CMO, I’d be enjoying the comeback, since OAI is so very much back.
Then I’d be supremely annoyed at what’s basically a folder of economics reports. 👇
Geeky context: I’m obsessed with AI, brand newsrooms, and The Big Lebowski (which, sadly, has nothing to do with this post. Yet).
Colin Fleming took over as OpenAI’s CMO for business a few months ago. He’s sitting on one of the biggest marketing assets in tech, and it’s currently dressed for a central bank conference.
The folder
OpenAI publishes Signals, which is genuinely great. A dashboard, research briefs, and a downloadable dataset under a Creative Commons license (nice move!).
It shows that Singapore sends the most ChatGPT messages per capita of the 147 countries in the ranking. It shows that at work, people are more than twice as likely to use ChatGPT to actually DO something than they are off the clock (professional managers, amiright?). Multimedia is 7.8% of messages globally and more than one in ten in Brazil and Colombia.
Anthropic’s Economic Index goes even deeper (not like they’re competing or anything).
Recipe requests to Claude are 2.3 times more frequent at 6 p.m. than average. Sleep advice peaks around 5 a.m. (hi, new parents). Personal conversations jump from about 35% of traffic on weekdays to just under 50% on weekends. On April 14, the day before the US tax deadline, tax conversations were eight times as common as a normal day in May.
My favorite: people who’ve used Claude for six months or more have a 10% higher success rate in their conversations. The newbies mostly write haikus, check sports scores, and ask for party food ideas. An ambitious crew.
These are the best behavioral datasets in marketing history. More than a billion people on the ChatGPT side alone, showing exactly how humans learn to use a new kind of tool.
And the data drops quarterly, the briefs read like they were written for the Fed, and the stated audience is “policymakers and researchers.”
That a newsroom does not make.
I’ve been on the wrong side of this
I’m not throwing stones from a glass data center.
At Freightos, our data newsroom started as “a boring weekly email about container prices, sent to a few dozen reporters who mostly ignored it. For years.”
Then “COVID hit and the boring email nobody read was suddenly the number everyone needed.” The index ended up “cited at the White House, traded on an exchange,” and it “became a seven-figure business.”
Overnight success…in eight years.
So I know the feeling of sitting on data and assuming the data is the asset. Turns out the asset is showing up with it every week, in a voice people actually want to read. We had a fraction of what these labs have, and a lot less money, and it still worked once we showed up consistently.
The labs did the expensive part. Privacy engineering, classification at scale, legal sign-off. The cheap part (a couple of humans and a calendar) is what’s missing.
What I’d change on Monday
The gap is tiny. Here’s what I’d do:
-
Hire two journalists. One paid to be right, FT-style. One paid to be read, BuzzFeed-style. Keep the economists, they’re excellent. Their job is to be correct. The journalists’ job is to make sure anyone notices.
-
Go weekly, not quarterly. Quarterly is a report. Weekly is a habit. Hell, go daily. You’re f**king OpenAI.
-
Put a named human on the byline. Institutions don’t get DMs. A person on the “how people actually use AI” beat, posting in public, out-distributes a PDF by a mile (fuzzy methodology, sue me).
-
Keep it ungated and make it embeddable. They already did the brave part with Creative Commons. Now build the chart embeds so every reporter on earth can paste the 6 p.m. recipe spike into their story.
Stories I’d publish in month one
1️⃣ The feature stack. What % of users have ever touched memory, projects, voice, or a custom GPT? My guess is embarrassingly low, and that’s the story.
2️⃣ Fantasy football! What % of users prepped their fantasy draft with ChatGPT, day by day until kickoff. How many asked whether to take a kicker in round 6. Bonus chart: how stupid my questions were (very).
3️⃣ The 6 p.m. query progression. Recipes at 6, sleep questions at 5 a.m., taxes at 8x normal right before the deadline. The daily shift from work tool to life tool, one hour at a time. Publish it weekly and call it the forecast.
4️⃣ The six-month glow-up. Anthropic already found that veterans succeed more often than newbies. Show what people do in month one vs. month six. Every user reading it will immediately want to know which side they’re on (that’s called a retention feature, and it’s free).
5️⃣ How your industry actually uses AI. Built for business. Usage by function and vertical, monthly, ungated. When do pricing and sourcing questions spike? This would beat every whitepaper in existence, and it’s literally Colin’s job to sell to those people.
Every one of these is a Reuters citation. None can be copied without the data. And the hard part is already done.
The kicker
Signals data is Creative Commons.
Which means a wrapper company with one good writer could run OpenAI’s newsroom for them, on OpenAI’s own data, and collect the audience. If I were at OpenAI, that’s the part that would keep me up at night.
If I were that writer, I’d start tomorrow.
It would really tie the room together, man.
Frequently Asked Questions
Does OpenAI have a brand newsroom?
Not really. OpenAI has Signals, a data hub run by its economic research team that publishes quarterly research briefs, an interactive dashboard, and a downloadable dataset under a CC BY 4.0 license. It’s aimed at researchers and policymakers, not at building an audience. A brand newsroom would put named journalists on that data, publish weekly, and write for the people actually using ChatGPT.What is the Anthropic Economic Index?
Anthropic’s ongoing research program tracking how Claude is used across the economy, built on a privacy-preserving analysis system called Clio. In 2026 it has published three reports: ‘Economic primitives’ (January), ‘Learning curves’ (March), and ‘Cadences’ (June), covering task mix, user tenure, hour-by-hour usage rhythms, and a survey of roughly 9,700 people.Why would a foundation AI company need a brand newsroom?
Because models are commoditizing and the durable moat is the audience that trusts you. A newsroom built on proprietary behavioral data makes you the industry’s source of truth, the stat reporters cite and the chart boards screenshot. Freightos, Ramp, and Redfin all did this with far smaller datasets.What's the difference between a data hub and a brand newsroom?
A data hub measures and publishes. A newsroom tells stories on a cadence, with named humans, for a specific audience, and distributes them where that audience already is. The labs have built excellent data hubs. Neither has built the newsroom on top.That’s the end of this one.
Keep reading