ChatGPT went from 300 advertisers to 820 in one quarter
Booking, Intuit, Home Depot and L.L.Bean are in. The ad load roughly doubled.
TGIF!
In April the biggest advertisers on ChatGPT were Speechify and Jotform. Time to stop treating it as a curiosity.
Sensor Tower’s US data says advertisers on ChatGPT went from around 300 in April to more than 820 in July, roughly 160 of them joining in July alone, while mobile users saw about twice as many ads per hour. The April top ten has been replaced by Booking Holdings, Intuit, Home Depot and L.L.Bean. Financial services went from 2% of platform ad spend to 12%.
These are panel estimates, not OpenAI’s books, and the base is tiny: $100 million in annual recurring ad revenue reported in May, against Google’s $81.6 billion quarter. So watch the advertiser count instead. Three hundred to 820 in a single quarter, while the ad load doubled and nobody left. That’s a marketplace forming, and marketplaces are cheapest while everyone is still arguing about whether they’re real.
The honest position for most of you is a small test with a clear kill criterion, run this quarter rather than next. Not because it will work. Because in six months the same answer costs four times as much.
TALK ON THE MARKET
Marisa Jones, analyst at eMarketer, on Pinterest’s quarter:
“Its AI tools have shown early promise, but they haven’t generated the same level of enthusiasm as competitors’ offerings.”
Pinterest guided to 13–15% growth against 18% last quarter and the stock fell 9%. Shipping AI ad features stopped being a differentiator this year, and the platforms are finding that out on earnings calls.
NUMBERS MATTER
55% and 63%. Share of US consumers who expect to spend more this holiday than last year, and the share who may hold out for Black Friday or Cyber Monday before starting, in Optimove Insights’ 2026 report of 648 US consumers. The money is there and it’s arriving late, which is precisely the shape of quarter that punishes a channel test you started in November.
DAILY PROMPT
I want to run a contained test on a new ad channel next quarter. Here is my paid mix with spend, CPA and volume by channel: [paste]. Act as a head of growth defending this to a CFO. Tell me which channel to take the test budget from and why it’s the least painful place to lose it, what the test must prove to earn a second round, the kill criteria and the date I check them, and three ways this test could produce a number that looks good and means nothing.
Use this while you’re building the Q4 plan, before the budget hardens. The kill criteria are the point. A test without one quietly becomes a subscription.
Have a tool or service that you want to get in front of our audience? Reach 20,000+ marketers who read AI-Ready CMO every day.
Or reply to this email and request a media kit.
CMO CORNER
Good leadership makes fewer decisions by making better ones earlier.
There will be a meeting about advertising inside assistants. Have it now, while it costs a small test budget and one person’s attention, or have it in nine months when a competitor’s name keeps coming up and the decision arrives with a deadline attached. Same decision, wildly different price.
We provide consulting services for CMOs on AI strategy for marketing teams through our Teamless offering.
We run an interview series with marketing leaders about how AI actually changes their work. Real use cases, not predictions.
If you’re doing something worth talking about, the opportunity is free and available for anyone. Simply reply to this email with your pitch.
TOOL UPGRADE
A new channel is only worth testing if you can see what it did:
Replo best for shipping a landing page per test without waiting on an engineering sprint. Try Replo here →
Amplitude best for following what the traffic did after the click, rather than stopping at the session. Try Amplitude here →
Anara best for asking the data plain questions when whoever writes your SQL is on holiday. Try Anara here →
We have 300+ curated, filtered, and categorized AI tools in our directory. Browse to find the best ones that help your marketing team.
We negotiated discounts, bonuses, or extended trials on AI marketing tools.
For $10/month, you save $1,724/month on tools. Plus, you can access all our content archive & chat as a bonus.
We even have a 10% OFF on all paid plans bonus for free subscribers.
LEARN THIS
If the channel test isn’t the right first pilot, here are the alternatives with the maths done: the AI Pilot Use Case Library. Forty-plus use cases scored by effort, impact and pilot fit, from bulk meta descriptions to turning competitor G2 reviews into battlecards. For the leader who wants a shortlist rather than a brainstorm.
We publish workshops, courses, guides, manuals, and more about AI marketing for free for everyone. Browse our learning resources for free.
BRAND SLOP
Google shipped an AI image generator inside Google Earth on a Thursday and pulled it on the Friday. One day. It let anyone zoom to real coordinates, type a sentence, and lay a photorealistic aerial scene over genuine satellite imagery.
Dutch researcher Henk van Ess published a post called “How to plant a nuclear plant in Iran” and did exactly that, plus refugees near the Mexican border and a bomb crater beside a hospital in Gaza. BBC Verify made a toppled Eiffel Tower and Russian tanks in Kyiv. AFP produced explosions in Paris.
Google says the images never entered the main Earth platform and that it’s rolling back to build guardrails. They did carry SynthID watermarks, which is the detail worth taking with you. Watermarking is forensics. It tells you what happened after somebody already believed it.
Van Ess made the sharpest point in the story: it takes a government request to remove a real building from Google Earth, and one sentence to add a fake one.
So before you ship anything with a generate button in it, work out what your product implicitly promises and whether generation quietly breaks that promise. Google Earth’s whole promise was that it shows you what is there.
That’s it for today!
See you next time,
— Peter & Torsten
