ChatGPT ads move into image generation, and trust is the test

0:05Host:Leaders Insights. Today: ChatGPT ads move into image generation, and trust is the test. Let me start with the thing I keep seeing people get wrong. Did OpenAI launch ads inside image generation on Monday, or did it announce that it might?

0:20Expert:It announced a test. BleepingComputer's report from October 5 says OpenAI will begin testing the format later this month, in the US, with an initial group of advertisers. Nothing is live in your account today. That distinction matters because a lot of the commentary is reacting to screenshots that do not exist yet.

0:39Host:Fine. But why should anyone running LLM systems at work care about an ad unit? That sounds like a marketing story.

0:46Expert:Because the ad unit is the small part. The larger part is measurement. OpenAI shipped attribution partners, conversion data integrations and incrementality testing in the same post. Attribution means tying a purchase back to the ad that preceded it. Incrementality means asking whether that purchase would have happened anyway without the ad. Those are the two things that turn a placement into a real advertising business.

1:11Host:Give me the names, because the names tell you how serious this is.

1:15Expert:Securities.io lists them. Attribution partners include AppsFlyer, Triple Whale, Adjust, DV Rockerbox, Northbeam, Branch, Singular, Kochava, Airbridge and Tenjin, with Fospha, Measured and INCRMNTAL on full-funnel measurement. Geo-based incrementality work is running with Haus, Measured and WorkMagic, and there is early-stage testing with Kantar and Cint on brand metrics. That is not a pilot roster. That is most of the performance marketing supply chain.

1:48Host:OpenAI also put out numbers. How much of that is vendor marketing dressed as evidence?

1:53Expert:A fair amount, and I want to be precise about who said what. OpenAI and several partners disclosed early findings, including WorkMagic reporting statistically significant lift for the wellness brand Dose, with 67% of incremental purchases coming from net-new customers and a geo-lift study estimating 2.3 times more incremental orders than last-click attribution captured. Triple Whale said 93% of Portland Leather's visitors from ChatGPT Ads were new. Those are partner-reported, single-brand results, published by parties with an interest in the channel working. The fact is that the numbers were disclosed. The opinion is mine: I would not plan a budget on two case studies.

2:39Host:There was one more figure in there that I thought was the most revealing.

2:42Expert:OpenAI's own early analysis of global campaigns optimised for deeper-funnel outcomes found 52.7% of eligible one-day view-through conversions happened within an hour of the matched ad impression. A view-through conversion is when someone sees an ad, does not click it, and buys later. Over half landing inside an hour suggests intent that was already in motion. Whether the ad caused it or caught it is exactly the question incrementality tests exist to answer, and OpenAI describes that work as early stage.

3:17Host:Now the uncomfortable one. OpenAI says ads do not influence the answers. Can anyone outside the company check that?

3:24Expert:Not today, no. The company says ads will be clearly labeled, remain separate from the image being created, and that ads do not influence ChatGPT's answers. That is a claim about internal ranking systems that no external party can audit from the outside. There are pilots with DoubleVerify and Integral Ad Science, two firms that verify where ads appear, to evaluate brand suitability in controlled testing environments, and Digiday notes those pilots will not give either company access to real user conversations.

3:57Host:So the verification deliberately avoids the data that would make verification meaningful.

4:01Expert:That is the trade-off, and OpenAI is explicit that privacy is why. My view: it is a reasonable privacy position and a weak assurance position at the same time. Both can be true.

4:13Host:This is not OpenAI's first collision with its own users over this.

4:17Expert:Right. TechCrunch reported the February rollout of ads to Free and Go tier users in the US, and before that, OpenAI faced a backlash late last year when it tested app suggestions that looked like unwanted ads. Anthropic ran Super Bowl commercials mocking the idea, with actors playing chatbots delivering advice next to badly targeted ads, and Sam Altman called those ads dishonest. The temperature on this topic has been high for nearly a year.

4:45Host:What is the strongest argument on OpenAI's side? Steelman it.

4:49Expert:Two things. First, the economics. Subscribers on Plus, Pro, Business, Enterprise and Education do not see ads, per TechCrunch, so the people being monetised are the ones who chose not to pay. TechCrunch's framing was that OpenAI needs revenue from the chatbot to cover the cost of developing the technology and growing the business. Ads fund free access. Second, OpenAI built the guardrails before the scale. Placement guardrails assess whether a conversation is appropriate for advertising, with the stated aim of keeping ads out of emotionally vulnerable, sensitive or otherwise unsuitable contexts. Compare that with how display advertising grew up on the open web, where nobody thought about context until after the damage.

5:35Host:And the advertiser side is not exactly thrilled either, from what I read.

5:39Expert:No. Digiday is the most useful reporting here because it is written for buyers rather than users. Since OpenAI started selling ads in February, advertisers have repeated three complaints: they cannot measure properly, there are not enough formats, and there is not enough inventory to absorb what they want to spend. Digiday's read is that this round of updates addresses two of them. The unaddressed one is inventory. Shamsul Chowdhury at Zeno Group pointed out that ChatGPT has roughly one ad placement against Google's many, and called it "a disconnect between supply and demand."

6:16Host:Which means prices go up and the pressure to add placements goes up with them.

6:21Expert:That is the mechanism to watch. Scarce supply with eager demand pushes toward more surface area, and more surface area is where the separation between ad and answer gets tested. BleepingComputer frames the test as limited to the US for now, while OpenAI tries to monetise 1.2 billion weekly users. One billion two hundred million. Even a small share of that attention is a large business.

6:46Host:For someone running LLMs inside a company, is any of this actually their problem?

6:51Expert:Three ways it becomes your problem. If you run marketing analytics, ChatGPT Ads is now a channel your stack will be asked to support, and the partner list tells you whether your vendor is ready. If you build consumer-facing products on top of ChatGPT, the free-tier experience your users see is changing. And if you write AI policy for your organisation, you now need a line on commercial influence inside assistant outputs, even if the current answer is that the paid tier you buy is exempt.

7:21Host:Give me one concrete thing to do this week.

7:23Expert:Open the partner list in OpenAI's October 5 post and check it against your own analytics and customer data tools. If you use AppsFlyer, Adjust, Branch, Triple Whale or Northbeam, the integration decision is already on your desk. Then write down, in one paragraph, what evidence you would accept that advertising does not shape model output. Keep it. When the US test goes live later this month, you will have a standard you set before the screenshots started flying, rather than one you invented in reaction to them.

7:57Host:Sources for today's episode: OpenAI will show visual ads in ChatGPT while you generate images, OpenAI Adds Visual Ad Format and Measurement Tools to ChatGPT, Image Credits:OpenAI, OpenAI moves to make ChatGPT Ads more measurable — and more visual, OpenAI starts putting visual ads inside ChatGPT, beginning with image generation - SimpMusic Blog, OpenAI to test visual ads during ChatGPT image generation. That's it. The AI decision tools are live at mba-training.com.