August 28th-September 3rd // Estimated Reading Time: 9 minutes
In This Edition!
ChatGPT ads are sexy, but is there substance?
Ad Astra…
Top Headlines🔥
Data reveals ChatGPT ads not performing
Source: MediaPost
August 28, 2026
Summary: As OpenAI's ads business hits a $1B annual run rate, anecdotal evidence is emerging that ChatGPT Ads are underperforming. One Canadian marketer told MediaPost about a small campaign where he got nearly double the expected impressions, but a 0.6% clickthrough rate, compared to the 2% he typically sees on search ads. None of his 60 clicks converted. Another agency CEO paused a campaign after OpenAI reported 57 clicks while Google Analytics logged fewer than 20 visits using UTM tracking, a gap other agencies echoed in the comments of his LinkedIn post. Other analyses have also shown that the ads aren't relevant 14% of the time.
Opinion: These are early, small-sample complaints, not a referendum. We believe in the long-term value of AI as an advertising channel.
What makes all of this tricky, however, is a) consumers have to get comfortable with ads in AI, b) the ad formats still have to be figured out, c) targeting based on AI conversations and context is still a work in progress, and d) OpenAI’s ad platform has a ways-to-go in terms of all the capabilities that are needed to make ads perform. All of this will take time. With OpenAI trying to IPO at a $1T valuation, and ads being a key part of a successful IPO, OpenAI doesn’t have the luxury of time. You know who does? Google.
New Products & Features 🚀
What It Does: GPT‑6 Astra is OpenAI's new flagship model, built to handle computer use, coding, and complex professional work like filling out forms, building websites, and running data analysis. Astra crosses a "critical" threshold for cybersecurity risk, capable of finding and exploiting software flaws, so OpenAI is holding back some features and adding extra monitoring before letting it loose. It's rolling out now to a limited set of organizations, with Plus, Pro, Business, and Enterprise users getting access in the coming days.
Quick Take: OpenAI's pitch here is "better and cheaper." As are most new model pitches nowadays. It leans heavily into cost-efficiency as a selling point in addition to performance. The frontier model race continues. Maybe Astra will help ChatGPT ads go to the moon?
What It Does: Google Pics generates and edits images using Google's Nano Banana model. It can isolate objects for targeted edits and change text inside an image, including translations, without redoing the design. It's rolling out inside Google Slides and Docs now, with Drive coming soon. Available to Google AI Pro and Ultra subscribers and most Workspace business customers.
Quick Take: For marketers, the most relevant feature may be in-image text editing. It could help them more easily localize creative across markets without rebuilding files or looping in design teams. And integrating it into Google Workspace makes the Google Suite stickier as well.
What It Does: Search Console's generative AI performance reports show how often you showed up in AI Overviews, AI Mode, and Discover. Google also rolled out a setting that lets site owners remove their content from those AI features entirely. Sites with low AI impressions may not see a report at all.
Quick Take: The reports tell you how often your pages showed up in an AI Overview or AI Mode answer. They don't tell you how many people actually clicked through to your site from there. Showing up in AI is still a catch-22 for content creators and publishers; you want to be visible in AI so people can find you, but you don’t want to lose traffic and time spent to AI either. Not having analytics makes the decision even harder.
What It Does: Comscore AI Intelligence now measures sponsored placements inside chat responses, in addition to organic citations. It links real user AI interactions to their browsing behavior, so brands can see whether someone clicked through to their site after seeing an AI answer, or got what they needed from the chat response and never clicked at all. It's based on opt-in panel data, not platform-verified, so treat findings as directional.
Quick Take: Comscore previously tracked organic citations, not paid placements inside chat responses, so this adds the missing paid layer, giving brands a way to compare paid and organic visibility. AI paid and organic visibility measurement is still a huge challenge (and opportunity), and is theoretically best solved by an independent 3rd party. It’ll be a mad dash between measurement vendors to own this over the next few years.
What It Does: Bazaarvoice's AI Visibility package converts customer reviews, ratings, and photos into structured data AI crawlers can read. Review platforms often render this content through JavaScript, which AI crawlers pass over.
Quick Take: Reviews are a strong source of content marketing for brands. Comparison shopping, based on user reviews, is a big consumer use case for AI. It’s imperative for brands that their reviews be crawled by AI!
What It Does: Optimizely built its own AI models for marketing tasks, like writing campaign copy and analyzing A/B tests, so customers no longer need general-purpose LLMs. The models hit 10x greater cost efficiency than frontier models in early testing. Optimizely also released Mark-Bench, an open source benchmark that tests models on marketing tasks, where Optimizely's AI beat Claude Code on accuracy at half the cost.
Quick Take: One of the biggest objections to any AI marketing platform nowadays is “well can’t my LLM just do that?” If the rebuttal from said AI marketing platform is “yes, but we can do it more efficiently, and we can prove it”, that’s compelling. Especially in the current AI cost conscious environment.
AI Use Case of The Week💡
Wellness brand Rouge Care let AI agents plan and buy its TV ads🔒
The Setup: Wellness brand Rouge Care wanted to run a connected TV campaign to build awareness among affluent households in Toronto and Calgary who hadn't yet tried red-light therapy. The brand set a goal of 250% return on ad spend for the quarter.
The AI Solution: Agency Klever Programmatic ran the campaign through PubMatic's AgenticOS. A planning agent turned Rouge Care's brief into audience and inventory options, ranging from tight matches for precision to wider options for added reach. A buying bot then executed on those choices. Klever's traders still approved pricing and final inventory, and the system handled most of the setup on its own. Mid-campaign, an agent flagged stalled reach and proposed new segments, with a human signing off before the change went live.
The Results:
• The $25K campaign led to $125K in sales
• 500% return on ad spend, double the campaign goal
Why This Matters: Agent-to-agent media planning and buying is starting to show up in real campaigns. Of course, the human-in-the-loop guardrails are critical. But this is also a story a brand chose to share. Fully agentic buying is still nascent, so maybe this is about the PR angle, but there’s obviously potential long-term.
Your Action: If you break down each part of the campaign process — planning, audience discovery, deal negotiation, campaign setup and launch, campaign optimization, and measurement — it's easy to see how AI already plays a role at each turn. So what value does "fully agentic" buying create? Is it the connectivity and feedback loop between each step?
Test it on your next campaign: map which of those steps you already run through some AI tool, then time the handoffs between them — how long does it take for an insight from your planning tool to reach the person making the audience targeting decision, or for a measurement report to influence optimization? If a lot of these handoffs are manual today, and take too much time, you have a case for going fully agentic.
Other Notable Headlines📌
Nextdoor combines human recommendations, AI to promote local businesses - A new hub uses AI to summarize neighbor recommendations and rank local businesses, with verified accounts and location checks meant to keep out bots and paid reviews.
We taught Claude how U of Digital teaches. Now it helps us scale. - U of Digital built a shared Claude workspace using the "Ask Your Org" feature, organizing company knowledge into five layers so team members can skip inputting all the context and utilize consistent skills when using AI.
Nvidia to buy Hugging Face for $12.9B - Hugging Face is an open-source platform where developers share and deploy AI models and datasets. Nvidia CEO Jensen Huang pledged the platform will stay open to all models, clouds, and hardware, not just Nvidia's.
Google's new Gemini model may narrow the coding gap with Anthropic and OpenAI🔒 - Google's flagship Pro series is months behind schedule, but Gemini 3.8 Flash, launching this week, is said to be performing well in internal tests.
CartographAI now helps ad tech vendors market themselves to advertisers - The ad tech and mar tech vendor discovery startup launched a paid accelerator program helping vendors like The Trade Desk and MNTN reach the right buyers.
'Oreogate' on Instagram went viral. But was it all a fake ad for an AI company?🔒 - Yes. An IG video series about a school lunch dispute over Oreos, told through screenshots, with a Mom rapping in the background, turned out to be a paid partnership with Suno, an AI music generator.
That’s It For This Week 👋
The U of Digital AI Edge Newsletter is intended for AI Edge Pro subscribers, but occasional forwarding is okay!
To subscribe visit Uof.Digital/Newsletters/ai-edge or AI Edge Pro, or contact us directly for group subscriptions.
And remember, U of Digital helps teams drive better outcomes through structured education on critical topics in marketing / advertising like AI, mar tech, programmatic, CTV, commerce media, and more. Interested in learning more about how we can supercharge your team? Let’s talk!
Thanks for reading!
