ChatGPT Ads 如何運作:廣告形式、競價方式與測試建議
ChatGPT Ads Are Real Now. Here's How They Actually Work
ChatGPT Ads 已在 Free 和 Go 方案的對話答案中投放,亦可在圖片生成流程中以視覺廣告形式出現。廣告主透過 OpenAI Ads Manager 建立活動,新活動可選 Target CPA 或 Target CPC;美國電商品牌亦可使用品牌智能體,讓點擊進入產品對話。目前價格、最低投放額及公開成效基準均未公佈,作者建議以小額且可追蹤轉換的測試,對照現有渠道評估。
作者提出以單一產品、可追蹤轉換和現有渠道作對照的小額測試框架,讓廣告主在缺乏公開價格與成效基準時衡量投放。
ChatGPT Ads Are Real Now. Here's How They Actually Work
I haven't run a ChatGPT Ads campaign myself yet. Nobody outside a small group of early advertisers has, really. The product is barely a month old in its current form.
So this isn't a war story. It's the honest map: what the ads are, where they show up, how you buy them, what they cost (spoiler: OpenAI won't say yet), and the one number that should decide whether you test them now or wait.
The short version
ChatGPT now runs ads. They appear inside answers on the Free and Go plans, to a product that has about 1.2 billion weekly users. OpenAI says more than 100,000 advertisers are already buying, with ads live in 89 countries as of October 2026. In November the pieces landed fast: Target CPA and Target CPC bidding for new campaigns on November 9, visual ads during image generation from November 10, verified business profiles and brand agents under "Businesses in ChatGPT" on November 10-11, and Aleph named exclusive reseller in 44 more countries on November 12.
If you sell something people research before buying, this is the first new ad channel in years that is actually worth understanding. Not because it's proven. Because the place people ask questions is becoming the place they decide.
Where the ads actually show up
Forget banner ads. There is no sidebar here. ChatGPT Ads live inside the conversation, and they take a few different shapes:
Suggestions inside answers. Someone asks for the best noise-cancelling headphones under $200. The answer names a few products. One of them can be a paid placement, labelled as sponsored. The ad is part of the answer, not next to it.
Visual ads during image generation. From November 10, image generation can surface visual ads. Someone generating images for a project sees a relevant visual placement in that flow.
Business profiles. "Businesses in ChatGPT" launched in beta November 10-11. Verified businesses get a profile, a dashboard showing mentions and clicks, and for US ecommerce brands, a brand agent: a version of ChatGPT that answers questions about that brand's products specifically. Paid ads can point at a brand agent, so the click lands in a conversation about your product, not on a landing page.
That's the shape to picture. Not an impression on a feed. A suggestion inside a decision.
How to start
Buying runs through OpenAI's Ads Manager. The mechanics will look familiar if you've bought search ads before:
- Set up a verified business profile through Businesses in ChatGPT, if your market has it. This is the foundation: your brand agent and your mention/click data live here.
- Create a campaign in Ads Manager and pick a goal. New campaigns now offer Target CPA (pay per conversion, tell the system what a conversion is worth) or Target CPC (pay per click, with a target cost) bidding, added November 9.
- Point the ad somewhere useful. A product page, or a brand agent if you have one. The brand agent is the interesting option: the user keeps chatting instead of bouncing off a landing page.
- Watch mentions and clicks in the dashboard, not just clicks. The Businesses dashboard reports both, which tells you whether people are even seeing your brand in answers.
Who can buy right now: advertisers in the 89 countries where ads run, buying directly from OpenAI, plus the 44 markets Aleph covers as exclusive reseller (announced November 12), including Bangladesh, Cambodia, Hong Kong and Sri Lanka. In those 44, you contract and pay locally through Aleph, in local currency, with credit terms. I wrote up the Aleph deal separately if you're in one of those markets.
One limit to know before you plan anything: ads only appear for users on the Free and Go plans. Paying subscribers don't see them. Your audience is the free tier, which is most of the 1.2 billion.
What it costs
Honestly? Nobody outside OpenAI and its early advertisers knows. No public rate card exists. No minimum spend has been published. Target CPA and Target CPC are bidding strategies, not prices: you set a target and the system bids toward it, and what you actually pay depends on competition for that audience.
That is normal for a launch this young and frustrating if you're budgeting. If you need a number for a plan, you don't have one yet. Talk to OpenAI or, in the 44 Aleph markets, to Aleph's local team, and treat the first campaign as price discovery, not a scaled buy.
The number that should decide it for you
Here's the stat I keep coming back to. According to Similarweb data reported by Ahrefs, Google's AI Mode sends clicks to websites on only 1.6 to 2.5 percent of queries, versus 17 to 19 percent for classic search. People ask an AI a question, get the answer, and leave. No click.
ChatGPT Ads live inside that same behaviour. The ad's job is not to win a click from a search results page. It's to get chosen inside an answer someone already trusts. That cuts both ways:
- If your product is the kind of thing people ask about before buying (headphones, software, a course, a local service), an ad inside the answer reaches them at the exact moment of decision. That is worth more than an impression on a feed.
- If people don't ask about your category, or the answer is usually a single factual reply, there is nowhere for your ad to live. A banner would have been seen. An in-answer ad that never triggers is invisible.
And the demand signal is real: Profound counted US users asking about ChatGPT Ads themselves 729,000 times in October 2026. People are already asking the product questions. Advertisers are just late to the answer.
Ads vs. being mentioned for free
This is the question every marketer is actually asking: do I need to pay, or will ChatGPT just mention my product?
Both exist. Organic mentions come from the model's training and the web it can reach: good reviews, comparison articles, a product page that answers real questions. Paid ads put you in the answer when the organic mention doesn't.
The honest answer is that nobody knows yet how much paid placement outweighs organic mention, or whether users trust a sponsored suggestion inside an answer the same way they trust an unpaid one. Early tests will tell. My guess, and it is only a guess: the brands that win here will be the ones that are mentionable organically and pay to be present. Paying without being a good answer is a fast way to spend money on clicks from people who bounce. Being a good answer without paying means hoping the model picks you.
Who should test now, and who should wait
Test now if: people already ask ChatGPT about your category, you have one product with a clear price and a clear conversion (a purchase, a signup, a booking), and you can afford a small experiment you might lose. The learning is the point of the first campaign.
Wait if: your product is impulse-driven, your conversion happens offline with no tracking, or you need a published rate card before your finance team approves anything. Check back in a quarter: pricing, formats and real performance data will exist by then.
A first test I'd actually run
If I were spending my own money, the first campaign would be small, boring, and measurable:
- One product only. The one with the clearest use case and the best reviews. Not your whole catalogue.
- One conversion. A purchase or a signup, tracked properly. Not "traffic."
- Brand agent on, if you can get one. Send the click to a conversation about your product, not a generic homepage. That's the one genuinely new mechanic here; test it.
- Target CPA bidding, with a target you'd accept if it worked. If you don't know your acceptable cost per acquisition, you don't have a test, you have a donation.
- Two weeks, then judge it on cost per conversion against your best existing channel. Not on clicks, not on mentions. Mentions are nice. Conversions pay rent.
Run it next to the channel you already know, same product, same two weeks. If ChatGPT can't beat or match it, you learned something cheap. If it can, you found a channel before your competitors did.
What we don't know yet
- Pricing and minimums. No rates or minimum spends published.
- Real performance data. No public case studies, no benchmark CPC or conversion rates.
- How users react to sponsored answers. Do they click? Do they trust them less? The first public data will matter more than any launch announcement.
- Targeting depth. What you can target beyond the conversation's context hasn't been detailed publicly.
- Whether the free-tier audience converts. 1.2 billion weekly users is a huge number. How many of them buy things through a chat window is the open question.
The bottom line
ChatGPT Ads are not a better version of Google Ads. They're a different bet: that the answer, not the results page, is where buying decisions will happen. The product is young, the pricing is unpublished, and the performance data doesn't exist yet. But 100,000 advertisers have already placed the bet, and the audience is 1.2 billion people asking questions every week.
If your customers ask before they buy, a small test now is cheap education. If they don't, wait for the data. Either way, don't sleep on it for a year. The advertisers who learn what a good ChatGPT ad is while the format is still new will have an edge over the ones who arrive once it's obvious.
Have you run a campaign yet? If so, I'd genuinely like to know: what did a click cost you, and did the brand agent change anything? That's the data point this whole conversation is missing.
Sources: OpenAI announcements on ChatGPT Ads (October-November 2026), including Businesses in ChatGPT beta (November 10-11), Target CPA/Target CPC bidding (November 9), and visual ads in image generation (November 10); Aleph/OpenAI reseller announcement (November 12, 2026); Profound data on US ChatGPT Ads queries (729,000, October 2026); Similarweb data via Ahrefs on AI Mode vs. classic search click-through (1.6-2.5% vs. 17-19%). No pricing, minimum spends, or performance benchmarks have been published. This article is researched analysis, not hands-on testing. I have not run a ChatGPT Ads campaign.
來源:DEV Community · dev.to