Subscribe
Share
Share
Embed
10 minutes of expert insights every weekday. Your morning ritual for staying ahead in retail media.
I Really, Really Hoped That Ads in AI Wouldn't Suck. Now We Find Out.
===
[00:00:00] Kiri Masters: I got married the smart way. My husband and I eloped at City Hall in New York. No planning, no audience, no budget, no fuss. For us, it was perfect. [00:00:15] Then the doubt crept in. Shouldn't we at least have a party for our family's sake? Aren't they missing out? Are we missing out? It was meant to be casual, but I got swept away.
[00:00:27] Fancy white dress hall [00:00:30] caterers, a hundred people. It was fun and I'm glad we did it, but damn it turns out we just had a wedding Anyway, that's how ads in a trusted [00:00:45] environment usually go to. It starts with principles and restraint. Then comes the pressure to scale, the urge to add just one more thing and suddenly you've rebuilt the very machine [00:01:00] you were trying to avoid.
[00:01:01] A couple of weeks ago, one of the leaders of Open AI's advertising division, Asad Awan, went on the record about exactly how ads will work inside Chat GBT. The full [00:01:15] interview is on the Open AI podcast and it's definitely worth listening to that regardless of where you sit in the retail media ecosystem.
[00:01:23] But I'm gonna break down a few of the things that I found interesting.
[00:01:27] What I was really comforted to hear [00:01:30] what else kind of stood out to me and a few more observations.
[00:01:34]
[00:01:36] Kiri Masters: Back in November, I wrote a piece for the drum arguing that ads inside AI assistance don't necessarily have to be [00:01:45] terrible. It was by my own admission. Hopelessly optimistic. Now predicting that OpenAI would launch ads wasn't exactly bold.
[00:01:55] It seemed kind of obvious from some of the hiring moves that they've made, [00:02:00] including hiring Fiji, CMO from Instacart to run the applications business. The harder questions were about. How they would actually do it. So now that this thing actually exists, people have started to see these ads show up in chat, GPT.[00:02:15]
[00:02:15] It's worth checking whether that optimism on my part was warranted. And so here's where my earlier thinking lines up with what. Assad a one explained in the podcast, so first I argued that the [00:02:30] ad layer should be architecturally separate from the model. The answers shouldn't be corrupted by who is paying to be there, and I was comforted to hear a one [00:02:45] confirm this explicitly
[00:02:46] that the model doesn't even know that an ad exists. In the response, if you ask chatt PT about an ad on the screen, it will say, I don't know. So you have to actually press a button to bring the ad into the [00:03:00] conversation. Now that is a harder wall between the model and the ad layer than even what I was expecting or hopeful for.
[00:03:08] So that is good to know. Second thing I argued for is contextual [00:03:15] targeting over surveillance based tracking a one described matching ads to conversation themes with sensitive topics like health, politics, violence automatically excluded from [00:03:30] ad inventory.
[00:03:31] Users can see exactly what data is being used. They can clear their history, they can turn off personalization entirely. Again, more robust controls than I'd assumed. And thirdly, I suggested that [00:03:45] the new model, which, ad industry veteran Brian O'Kelly calls Agentic Advertising, which is all about targeting content rather than people that could avoid the privacy nightmares that broke [00:04:00] trust in traditional digital ads.
[00:04:03] Again, AANS rubric fits this mold. That user trust comes above user value, which comes above advertiser value, which comes above revenue. Now that sure [00:04:15] sounds like AI ads may have been paying attention to some of the challenges happening over here in this strange little corner of the media universe called retail media.
[00:04:24] You and I have talked about how retailers like Costco are putting things like [00:04:30] membership renewals and membership value ahead of ad revenue growth. They know that they could make a heck of a lot more money through media sales, but they are holding back on that because. They [00:04:45] understand that they could in fact kill the golden goose of their core retail business if they dial in the ads too much.
[00:04:52] So those were some of the easy calls that I landed. Let's talk about some of the more interesting things that I didn't [00:05:00] anticipate. First, this philosophy called One good ad is enough. If there is no relevant match. OpenAI would rather show no ads than any ads. Here's what AAN said in the [00:05:15] interview.
[00:05:15] You don't want advertisers to pay randomly for impressions. You don't want users to see too many ads. You want to show the one right ad. Now, this is the polar opposite of. Onsite [00:05:30] retail media advertising. Think about someone like Amazon, where search results now have, a couple dozen sponsored product ads per page with close to near saturation.
[00:05:44] [00:05:45] OpenAI has a very different bet that is. Restraint that builds long-term value rather than filling every available slot. Second is the algorithmic play. Asada one described a future where a small [00:06:00] shoe brand doesn't need to hire three performance marketers.
[00:06:03] Instead, the founder tells chat, GPT sell more shoes in the Midwest, and the system handles campaign creation, bidding, and optimization. Now, to be honest, this is where [00:06:15] my Spidey sense goes up. Algorithmic performance models like this. Can work well for SMBs, but sophisticated advertisers, brands and agencies alike want more [00:06:30] granularity and precision.
[00:06:31] What I hear anecdotally again and again, is that these algorithmic black box models are not. Very popular amongst the sophisticated large advertisers out there. So I really hope that [00:06:45] it's not black box or nothing here. Did you know that leading retail media networks drive 85% of their [00:07:00] ads through mid and long tail advertisers?
[00:07:03] Miracle Ads provides full funnel ad formats tailored to both one P and three P advertisers leveraging unique AI capabilities that provide unprecedented [00:07:15] levels of relevance and engagement. Retailers who want to capture ad spend from the long tail of three P Marketplace sellers use miracle ads in their tech stack.
[00:07:26] Learn more@miracle.com. That's [00:07:30] M-I-R-A-K l.com.
[00:07:33] Number three, the audience split ads appear only on the free and go tiers and paying subscribers.
[00:07:41] See no ads. Now, that's a classic freemium play. Not [00:07:45] super surprising, but think about what that means. Some of the power users don't see any ads. Advertisers only get the people who are using the free tier. And is that audience as [00:08:00] valuable as the power users? So that's interesting. Maybe like traditional media.
[00:08:05] We, We have examples where I pay a subscription for the Wall Street Journal, let's say, and I still see ads. I pay [00:08:15] for. Disney, uh, still see ads. Um, So that might be a direction that they go in as well. Right. Let's talk now about what this means for retail media. So this wasn't part of the podcast interview, [00:08:30] of course did not come up, but here's, you know, the links that I'm gonna draw back there.
[00:08:35] The most immediate implication is that there is now a high intent. Ad surface that sits upstream of the [00:08:45] retailer. So this is the most obvious threat to retail media. New ad surface that could be quite appealing to a brand advertiser. As of right now, this is extremely limited. It's, advertisers paying on a, [00:09:00] uh, a CPM basis.
[00:09:02] Actually Debbie Ajo Williamson, who writes a a a great Substack, a link to in the comments said that she had anecdotally heard that the insertion orders for the chat GPT ads were all being done [00:09:15] on spreadsheets. So it is definitely early, early days here, but, um, this is conceptually a threat to retail media just because it's a new ad surface that sits upstream of a retailer, but that is [00:09:30] not the only challenge and it's ~easy to~
[00:09:32] ~I, it is ~easy to disregard. ai, media because if you consider that to be the only threat, the bigger threat in my mind is all of these signals that [00:09:45] retailers have built their ad businesses on, which is the strength of their first party data. They know what you actually bought. They know what you're considering buying.
[00:09:54] They know what you clicked. They know what you clicked and didn't purchase. That is the strength of [00:10:00] retail media, whether we're talking about onsite or offsite, all of this context planning, researching. And then finally, of course, the closed loop, eventual sale. So this migration of research in the [00:10:15] shopper journey takes away those signals.
[00:10:18] so that is a topic that I've covered. Extensively in past episodes. I'll link up to those in the show notes.
[00:10:24] But I bring this up because Assad Awan in that interview teases something that [00:10:30] directly connects to that thesis, which is where he described a future where ads work behind the scenes. An AI agent that knows that you like ramen, and they discover a new vegan [00:10:45] ramen brand.
[00:10:46] And the AI surfaces that and may eventually facilitate that purchase for you. So if that becomes real and the entire discovery to purchase journey happens inside that AI layer [00:11:00] again, , I think that is, uh, you know, the technology isn't creating a, a, a, a perfect experience around that yet. But that is the.
[00:11:11] Potential eventual threat as well . Now [00:11:15] from this conversation, it seems that Asan said all the right things, but every platform says the right things. At launch, Google didn't set out to make search results indistinguishable from ads. [00:11:30] Amazon didn't plan for sponsored products to overwhelm organic results.
[00:11:33] The real test isn't how this looks today. It's what happens in year three when there's an entire division within OpenAI with revenue targets [00:11:45] and. If it eventually becomes a public company and people are asking whether that wall between the model and the ads really needs to be quite so rigid,
[00:11:55] Will that AI ads ecosystem face a version of the retail media [00:12:00] doom loop When the easy money from advertisers is gone?
[00:12:04] The answer that Asad Juan gave in the podcast to that question is this, we are in the business of trust. You can't drift when the [00:12:15] incentive is set up to be the best at this . Maybe Open AI is diversified revenue streams with enterprise customers, with subscriptions, things like that. Maybe that gives them less pressure to squeeze [00:12:30] ads than for traditional publishers, for example, or retailers where the media revenue stream has become more and more important.
[00:12:39] I learned the hard way that a small party doesn't stay small [00:12:45] unless you keep coming back to your principles. Open eyes. Real tests won't be launch week. It'll be year three when the growth targets show up.
[00:12:55]