TL;DR
- OpenAI and TikTok are testing the same idea: the ad click no longer sends people away, it opens a conversation with the brand.
- For shoppers, the question that used to die on a landing page now gets answered while the interest is still warm. That matters most on Black Friday, when people make a lot of decisions in a few days and every unanswered question costs a sale.
- For brands, the real prize is the conversation itself. Whether they'll ever get to hear it is still open.
The click used to be a goodbye
For twenty years, the job of an ad was to make someone leave. Leave the feed, leave the search results, land somewhere else, and hope the landing page answered whatever question made them curious in the first place. Most of the time it didn't, and we called what happened next a bounce rate.
Now picture two people on the same afternoon. One asks ChatGPT which sofa would survive a dog and a toddler. The other scrolls past a sofa on TikTok and stops, wondering whether it would even fit through her door. Different moments, different moods, one asked a question, the other was simply caught by something. Yet, if the tests running right now go where they seem to be going, both of them end up in the same place: talking to the brand, in a brand controlled chat, without ever leaving the app they were in.
That is the shift I want to pinpoint before this Black Friday. The ad is no longer the door out. It is becoming the room you stay in and ask.
Let’s sit on it for a while. This resembles the zero-click marketing strategy that we love to talk about in B2B, but it’s topic for another article.
So, I have been following how agents are reshaping commerce for a while (you can find my article about agents taking over the errand, not the relationship from last month), and this felt like the advertising side of the same story finally showing its hand.
Full disclosure, I work on brand-owned experiences for a living, so read me with that bias in mind.❤️
Two doors into the same room
In mid-September, OpenAI began testing what it calls Sponsored Agents with a small group of US advertisers [1][2]. The mechanics, as described so far: someone clicks an ad in ChatGPT and can choose to open a clearly labelled conversation with an agent sponsored by that business, separate from ChatGPT's own answer [3]. The brand builds the agent from what it supplies (website content, service details, product feeds), and the agent can keep going across several turns, including questions nobody scripted. Brands pay for the first click, not per message [2].
Then, yesterday, TikTok announced its own set of agentic tools [4]. The Lead Agent runs inside TikTok DMs or on lead forms on the advertiser's own site, and works from a knowledge base of "brand-approved information, such as offering details, FAQs, eligibility requirements." The Shopping Assistant answers product questions in a chat window with, in their words, "AI-powered answers based on information provided by the merchant."
There is also Buy Direct, a one-click checkout, which I'll leave aside here because the checkout is not the interesting part, well at least for me. The conversation is.
I know, the two are not identical, and the difference is worth holding onto:
ChatGPT is question-led: the person has already told the platform what they need, so the brand arrives at the exact moment of need.
TikTok is curiosity-led: the interest is sparked mid-scroll, often before there is a need at all, so the brand arrives at the moment of interest.
Two very different doors.
Yet both open into the same room, a conversation that runs on the brand's own information, inside someone else's platform, while the person is still warm.
Not a one-off
Once you notice the pattern, it is everywhere this year.
Google's Business Agent for Leads, announced at Marketing Live in May and in open beta in the US, replaces the static lead form inside an ad with a Gemini-powered chat trained on the advertiser's website [5].
Meta's Business Agent, launched globally in June, answers questions, recommends products, books appointments and qualifies leads across WhatsApp, Messenger and Instagram; Meta says more than a million businesses already use it, on top of "more than one billion active threads with businesses" every day [6].
The retailers are doing something adjacent, and the contrast is useful. Inside Amazon's Rufus, brands can pay for sponsored prompts, a suggested question like "Why choose [brand] shirt?", billed per click [7].
Walmart has opened ads inside its Sparky assistant too [8]. There, the brand buys a line in the retailer's script; the retailer's agent stays the narrator. With OpenAI, TikTok, Google and Meta, the brand gets its own voice, on someone else's stage.
None of this is entirely new, to be fair. Click-to-WhatsApp ads have been sending people into brand conversations for years.
What is new is that a machine can now hold millions of those conversations at the same time, in the brand's tone, at three in the morning, in the middle of the biggest shopping week of the year.
Why peak season is the real test
Think about what Black Friday and Cyber Monday actually look like from the shopper's side. A whole year of "maybe later" gets decided in a handful of days, with a timer running on every deal. Is this discount real, or was it the same price last month? Will my size still be there tomorrow? Does it work with what I already have? Can I return it if it doesn't? Every one of these questions used to be a small cliff on a landing page, and with a countdown on top of it, a lot of interest fell off.
The numbers suggest the conversations are coming either way. Adobe expects US online holiday spend of $275.1 billion between November 1 and December 31, and traffic referred by AI services to grow 130% year on year, from a still-small base, as they are careful to note [9]. Salesforce, which of course sells AI agents itself, reports that 41% of shoppers say a brand-owned AI assistant answering their questions makes them "much more confident" in a purchase [10]. Take the vendor numbers with the usual pinch of salt; the direction is hard to argue with.
To be clear about where we are: most of what I am describing is still in testing, eligibility-based, US-first, with no published results. So, very few brands will run a sponsored agent this November. Hence, I don't see this BFCM as the main event. I see it as the preview, the last peak season where an ad mostly reports a click, and maybe the first where a handful of them report a conversation.
The part I am most curious about: from signals to sentences
This is where I have to admit my real excitement, and it is not the format.
Until now, ads gave us signals: an impression, a click, a view, a bounce. All of them tell us that something happened. None of them tell us why.
A conversation is a completely different kind of material. Which words people choose to describe what they want (rarely the words on our product pages). The question asked right before someone leaves. The objection that comes up a thousand times in the same week. The competitor someone names in their third message. How a person phrases hesitation when the deal ends at midnight.
Multiply that by millions of people talking at the same time, and you have something no focus group, survey or search report has ever given a brand: the language of demand, in the customer's own words, at the exact moment they were deciding.
And yet, I don't know how much of it brands will ever get to hear. OpenAI's ads policy is very clear that "advertisers do not have access to your chats, chat history, memories, or personal details," and that they "only receive aggregate information about how their ads perform such as number of views or clicks" [11]. For Sponsored Agents specifically, OpenAI has said it has yet to decide what insight advertisers will receive from these conversations [2]. Brands, for now, can guide an agent's goals and tone, yet can't approve every answer or plug in a chatbot they already own [2]. So the brand gets a voice in the room. Whether it also gets ears is still open.
History gives a hint of which way platforms tend to lean. In 2020, Google narrowed its search terms report, the closest thing marketers ever had to customers' own words, to "only include terms that were searched by a significant number of users," citing privacy [12]. And to be fair, that instinct protects something real. People talk openly to these assistants partly because they trust that advertisers can't read over their shoulder, and that trust is the whole reason the conversation is valuable.
So here is where my curiosity lands. I don't need to know who asked. I would love to know how they asked. Patterns, not people. The ten ways customers describe the same need, the three questions that always come before the sale, the one worry nobody answers. That is a trade I think most brands would gladly take, and one that keeps the customer as the one who benefits, because the point of hearing it is to answer the next person better.
Your FAQ just became your creative
Which brings me to the practical part, and the opportunity I think is sitting right there. In every one of these formats, what the agent says comes from what the brand has written down about itself. The eligibility rules, the sizing logic, the delivery cut-offs, the return policy in plain words, the tone. In other words, the knowledge base is the ad now. The brief we used to write for a 15-second spot is becoming the document an agent reads before it talks to a customer.
That is good news for the brands who have been listening all along. Every question your customer service team answered last November, every chat transcript, every "does this come in…" in the comments, is suddenly creative material. The brands that already know what their customers ask will have a head start, whether or not they run a single sponsored agent this year.
And if I had one suggestion for this BFCM, it would be simple: write down the questions. Whatever channel they arrive through, collect the real questions people ask during peak week, in their own words, and treat them as next year's brief. If the platforms end up sharing the language of these conversations, you will know exactly what to listen for. If they don't, you will already have your own.
Where I've landed, for now
I am an enthusiast when it comes to new things with FOMO running my veins on a daily basis. And for this one, I have to say I am optimistic yet cautious. For the person on the other side of the screen, an ad that can answer is simply more useful than an ad that sends them somewhere to search for the answer themselves. For brands, the trade is clearer than it first looks: we are being invited to speak inside platforms we don't own, and the price of admission, for now, is not hearing everything that was said.
The question I keep coming back to is not whether ads will become conversations. They clearly are. It is this: if our ads could finally hear our customers, what would we want to learn first, and are we ready to change something because of it?
I would love to hear what you are seeing, especially if you are part of any of these tests going into peak season. Any signals I’m missing or misinterpreting? Tell me I’d love to know!
References
- AdTech Radar, "OpenAI Launches Sponsored Agents for ChatGPT Ads," 17 Sept 2026. adtechradar.com
- Citybiz, "OpenAI's Sponsored Agents Let Brands Sell Inside ChatGPT, With Limits on Control and Data," 23 Sept 2026. citybiz.co
- Quasa, "ChatGPT Sponsored Agents Test Keeps Answers Separate." quasa.io
- TikTok for Business, "TikTok Agentic Solutions," 5 Oct 2026. ads.tiktok.com
- Search Engine Land, "Google tests new conversational ad formats in AI Mode and Search," May 2026. searchengineland.com
- Meta, "Be There for Every Customer With Meta Business Agent," 3 June 2026. about.fb.com
- Shopifreaks, "Amazon reveals pricing and measurement details for sponsored ads inside Rufus AI shopping assistant," 1 Apr 2026. shopifreaks.com
- eMarketer, "Walmart bets on agentic AI with advertising assistant, Sparky ads." emarketer.com
- Search Engine Watch, "Adobe expects AI shopping referrals to more than double this holiday season," 30 Sept 2026. searchenginewatch.com
- Salesforce, "2026 Holiday Predictions: A Unified Yet Fractured Journey." salesforce.com
- OpenAI, "Testing ads in ChatGPT," Feb 2026 (updated Aug 2026). openai.com
- Search Engine Land, "Google Ads to limit Search Terms reporting, citing privacy," Sept 2020. searchengineland.com
