Suppose you are about to buy a chair.
The ad tells you: limited-time offer, design award winner, save 300 yuan today.
But what you actually want to ask is: “My home is only 30 square meters—will it be too large?” “I have a cat—is the fabric easy to clean?” “My back hurts—will it become uncomfortable after a long time?” “The doorway is only 75 centimeters wide—will it fit through?”
An image cannot answer those questions. A landing page may run for dozens of screens, but you may not want to dig through it.
Now imagine someone beside the ad who knows the product, never pressures you, and is even willing to say, “This model is not right for you.” That changes the experience.
This is what interested me most about Sponsored Agents. They may change less about what an ad looks like than about the fact that an ad can finally listen to a person’s whole question.
For the First Time, an Ad Can Become a Conversation
On September 16, OpenAI announced new ChatGPT Ads capabilities, including Sponsored Agents in testing.
After seeing a clearly labeled ad, a user can choose to enter a conversation with the brand’s agent. They can ask about their space, budget, preferences, and concerns, understand whether the product actually fits, and then decide whether to visit the brand’s site.
OpenAI used a dining table as its example. Seeing the table is only the beginning. People will also ask how many it seats, whether it fits their dining room, and how it should be maintained.
Ads used to send people to a page, and the page tried to present everything at once. Now another layer appears between them: first understand this person’s real problem.
On September 23, ChatGPT Ads also began expanding to Indonesia, Malaysia, the Philippines, Singapore, Thailand, Vietnam, and Taiwan. OpenAI said it had reached more than 60 countries and regions.
That does not mean Sponsored Agents are open in every market. They were still being tested with selected advertisers in the United States. But the direction is clear:
Advertising is moving from “look at me” to “what would you like to know?”
A Good Salesperson Is Better at Judging, Not Merely Talking
Traditional advertising is built to answer one question: “Why should you buy me?”
Conversational advertising must answer a harder one: “In your specific situation, am I actually right for you?”
The sentences are close, but their logic is very different. The first amplifies advantages; the second must understand limitations. The first pursues a click; the second needs trust.
A useful brand agent cannot merely recite marketing copy. It needs to know dimensions, price, materials, delivery times, and support boundaries. It must also know which needs the product cannot solve.
Sometimes its most valuable sentence may be: “If this is what matters most to you, I would not recommend this model.”
That may lose an order in the short term. In the long term, it may be the first time a user believes the thing on the other side of the screen is not trying only to close a sale.
Landing Pages Will Remain, but Their Job Changes
Conversational ads naturally raise a question: will brands no longer need websites, product pages, or design?
My judgment is the opposite.
When AI summarizes a product for a user, the brand’s own content becomes evidence for its answer. Vague specifications, invented cases, and contradictory pages will only be exposed faster.
The layers will have clearer roles:
- Ads create the meeting: they appear when a need emerges.
- Conversation supports judgment: it discovers the context, concerns, and limits.
- The website supplies proof: it shows real cases, details, prices, and the brand’s attitude.
- The app delivers experience: it makes the promise happen and builds a lasting relationship.
- The product fulfills the claim: the product itself must still create the final value.
AI does not make brands unimportant. It makes vague brands harder to hide and serious products easier to understand.
What This Could Mean for My Own Products
Imagine someone sees Yixiu described as “helping you relax,” then asks: “I often wake at three in the morning. I do not want a voice—only ten minutes of nature sounds. Is there something suitable?”
A good answer should not exaggerate medical effects. It can explain what Yixiu does and does not provide, then recommend Morning Forest Birds or another track based on time and sound preference, with a preview first.
For Wendao, someone may ask: “I feel stuck at work. What perspective can the Tao Te Ching offer?” The agent does not need to pretend it can make the decision. It can introduce a reading path from the original text and explanation to personal insight and everyday practice, then let the person decide whether this is how they want to read each day.
For Xiazi, a person may say: “I work in cross-border commerce. I only want yesterday’s important changes in AI and consumer brands, with links to the English sources.” The conversation should not repeat all 18 posters. It should help them find the part that matters to their work.
These are product-design exercises, not claims that these ads or agents are already live. But they clarify something for me:
Product communication should not only say what I have. It should help a person decide what it has to do with them.
An Undervalued Opportunity for One-Person Companies
In the past, only large companies could afford presales teams that covered multiple time zones, knew the entire product line, and spoke several languages. A one-person company usually wrote an FAQ and hoped people would read it.
Conversational sales creates another possibility:
- questions asked at three in the morning can receive answers grounded in real material;
- one product can be explained in different languages to different markets;
- questions people would never send in a private message may finally be heard;
- every question teaches the creator what users are actually hesitating over.
There is no magic in which “connecting AI” automatically creates sales. An AI salesperson’s ceiling depends on the product facts you give it; its floor depends on whether you are willing to admit the product’s boundaries.
What a one-person company needs is not smoother sales talk but a maintainable product knowledge base: real specifications, use cases, who it suits, who it does not, common questions, comparison criteria, delivery processes, and support rules.
Five Things You Can Do Now
1. Collect 30 real questions. Use messages, comments, support conversations, and trials with friends. Do not invent them at a desk.
2. Write a page called “Who this is not for.” A product willing to reject the wrong order is more likely to earn the right user’s trust.
3. Put proof on your own website. Cases, prices, specifications, and update history should be findable and citable.
4. Become clear in your first language before translating. AI can translate words; it cannot repair a vague position.
5. Record questions, not only clicks. What people repeatedly ask often reveals the product’s next improvement.
Advertising Is Still a Mirror
I have never liked ads that chase people around. They know what I clicked, but not why I hesitated. They keep urging me to buy without asking whether the thing suits me.
Conversational advertising offers another possibility: when a user chooses, a brand can answer one specific question seriously instead of occupying a few seconds of attention.
It will also magnify bad products faster. If the product has no value, the agent is merely a more diligent promoter. If the facts are false, it will say the wrong thing more fluently.
The next generation of advertising may compete less on whose model is smarter and more on who understands their product, who respects the user’s question, and who will trade one honest answer for a longer relationship.
The best ad may truly stop being an image. It may feel like a knowledgeable, measured person who does not follow you around asking for an order—yet happens to be there when you need them.
---
Fact note: OpenAI announced Sponsored Agents on September 16, 2026, as a test for selected U.S. advertisers. Its September 23 announcement concerned expansion of ChatGPT Ads to six Southeast Asian countries and Taiwan; it did not mean Sponsored Agents were fully available in those markets. The Yixiu, Wendao, and Xiazi examples are design explorations.