WayToGlobal Insights · ChatGPT Ads

WayToGlobal Insights · ChatGPT 广告

ChatGPT Ads for DTC Brands: A Category-Level Q4 Playbook

ChatGPT 广告投放实操:品类级 Q4 手册

2026-10-28 · 9 min read · checked against primary sources

2026-10-28 · 约 9 分钟读完 · 逐条核对原始来源

All insights全部文章

30-second read

  • ChatGPT Ads hit a $1B annual run rate in under 200 days — the fastest new ad surface open to brands in years.
  • A category playbook: which DTC verticals test first, with what creative and budgets.
  • Measurement is the open question — set the attribution story before the spend.

title: "ChatGPT Ads for DTC Brands: A Category-Level Q4 Playbook" slug: chatgpt-ads-for-dtc-brands date: 2026-10-27 meta-description: "How ChatGPT Ads actually works, what two verified ecommerce campaigns returned, and how to set attribution so you can judge the channel yourself."


# ChatGPT Ads for DTC Brands: A Category-Level Q4 Playbook

You're weighing a Q4 budget for ChatGPT Ads. So you search, and you find a dozen confident answers pointing in opposite directions: some say AI-referred traffic converts better than Google, some say AI crawlers take your site and send nobody back, and some describe ChatGPT advertising as "keyword ads, but smarter."

Each of those is partly right. What they get wrong is scope. This article assembles the verifiable pieces: the ad system mechanism OpenAI has published, two ecommerce campaigns with public results, and how attribution actually works — because most of the contradiction comes from nobody stating which methodology produced which number.

Three things are settled before we start.

This is a real channel you can buy. On August 31, 2026, OpenAI announced that ChatGPT Ads had reached $1 billion in annualized revenue run rate less than 200 days after launch, with more than 100,000 advertisers running campaigns. On September 23 it added seven Asian markets — Indonesia, Malaysia, the Philippines, Singapore, Thailand, Vietnam, and Taiwan — bringing availability to more than 60 countries. One detail is easy to miss: ads are shown only to Free and Go subscribers. Pro and Enterprise plans stay ad-free.

The mechanism is not keyword advertising. Plenty of write-ups — including several in Chinese — compare ChatGPT Ads to "smarter keyword ads." The direction is right. The mechanics will mislead how you set up the campaign.

The channel has no published benchmark, so you build your own. OpenAI publishes no CTR, conversion rate, or customer-acquisition-cost figures. The only numbers available are brands' own, and the comparison has to be yours. So the useful question isn't "is this channel good" — it's "what am I comparing it against?"

1. The Unit of Decision Is a Conversation, Not a Query

The ChatGPT Ads delivery pipeline: from conversation state to impression
The ChatGPT Ads delivery pipeline: from conversation state to impression

On October 21, 2026, OpenAI's engineering team published a complete technical description of the ad system — currently the only first-party account of how it works. The most important line: the ChatGPT ads system does not run on its own models and its own code. It runs alongside answer generation, independently, in its own service. That architectural boundary produces two consequences. If the ads system fails or slows down, ChatGPT's answers are unaffected. And showing no ad is an ordinary, successful outcome — the system delivers only when an ad is clearly beneficial to users and advertisers alike.

What decides how you operate the campaign is the unit of decision. Conventional ad systems start from discrete events: a search submitted, a page viewed. People arrive at ChatGPT with a goal, and the goal gets more specific across a conversation. OpenAI's own example: someone asks how to train for a race, and several turns later asks about shoes. The unit of decision is therefore the session — ad state is remembered and updated across turns and across conversations.

The operational consequence is concrete: your creative has to match the thing the user is discussing right now, not a category keyword. Tecdo's October 24 explainer, "The Trigger Logic of ChatGPT Ads: Intent Resonance," puts the rule plainly: an ad is displayed when the user-side and advertiser-side intents match. Their example is small-apartment bedroom storage — a user asks how to save space, ChatGPT infers "space-saving bedroom cabinets suitable for small apartments," and matches ads against that.

The full pipeline runs in six stages (see Figure 1): preparing retrieval queries → broad retrieval → eligibility filtering → relevance and quality scoring → the auction → final checks and display.

Two details are worth keeping. The auction enforces a minimum relevance bar — the orchestrator only nominates contextually appropriate candidates, and pricing belongs to a separate auction tool. And server-side delivery isn't the same as a viewable impression: the ad has to arrive before ChatGPT finishes generating its answer, or it doesn't appear that turn. The system logs why — the turn wasn't eligible, retrieval returned no candidates, candidates failed hydration or eligibility, or an ad was delivered and received an interaction. So zero impressions are diagnosable, not a black box.

2. Two Verified Ecommerce Campaigns, and Their Caveats

Two published ecommerce campaigns: Loop Earplugs and Newegg
Two published ecommerce campaigns: Loop Earplugs and Newegg

OpenAI's customer stories page carried seven case studies on October 27. Only two are consumer ecommerce purchases: Loop Earplugs and Newegg. The rest are tools, lead-gen, and developer products. A thin case list isn't evidence of weak performance — OpenAI only publishes brands that agree to be published — but it does mean the verifiable ecommerce sample is small.

Loop Earplugs is the more transferable of the two, because the structure of the approach transfers: for each new market, start with a product-feed campaign, then layer on non-feed creative split by use case once the feed campaign has results. A product feed is a catalog — name, description, price, images, availability — and ChatGPT Ads uses it to generate ads featuring specific items. Loop prepares a feed per market with the right language, currency, and pricing.

The numbers (per the case page footnote: Loop's Ads Manager data, August 18–September 8, 2026; post-click attribution on the order-created event, excluding view-through conversions): five live product-feed campaigns returned 5.1× ROAS and a 3.7% post-click conversion rate combined; a non-feed EU "concerts and music" campaign returned 1.8× ROAS. After the test window, Loop roughly 2.5×'d daily spend on the EU feed campaign and opened new campaigns in Sweden, Norway, and Switzerland.

Loop's own read on why is more useful than the ROAS: non-feed creative performed better once it moved from generic "earplugs" to a specific situation like "festival earplugs." In their words, "someone describes a situation, not a product category, so a campaign built around a situation has something to connect to."

Newegg took a different shape: evergreen product-feed campaigns as the base, seasonal campaigns layered on, and extra budget concentrated on promotion windows. It recorded 3× ROAS across campaigns over 28 days and 7× ROAS on a two-week Fantastech Sale campaign. Performance marketing manager Alec Shao also noted that, compared with other channels, visitors referred from ChatGPT were more likely to convert and purchase. Two caveats matter: Newegg is a multi-category retailer rather than a DTC brand, and the case page does not disclose the attribution window — so 3× and 7× cannot be set directly beside Loop's 5.1×.

That is precisely why you build your own control. In OpenAI's October 5 measurement update, DV Rockerbox measured WeightWatchers' attributed CPA on ChatGPT Ads at 15.3% below its blended paid-search benchmark, and WorkMagic's geo-lift study estimated 2.3× more incremental orders for Dose than last-click attribution captured. The first is a brand measured against itself; the second shows last-click understates the channel. But 2.3× is one brand's estimate, not the channel's typical multiplier.

3. How to Account for It: You Set the Attribution Window

ChatGPT Ads attribution: the two settings you control and the known risk
ChatGPT Ads attribution: the two settings you control and the known risk

This is the most practical section of the article. ChatGPT Ads gives advertisers two adjustable attribution settings. You choose the click-through attribution window, sized to your buying cycle. You can additionally enable a one-day view-through window to capture conversions that follow an impression without a click. And one qualifier has to be quoted exactly: these settings control attribution in reports; they do not change how campaigns optimize.

That is not a minor point. If you read the channel on last-click alone, you will also understate it — in OpenAI's early analysis of global campaigns optimized for deeper-funnel outcomes, 52.7% of eligible one-day view-through conversions occurred within an hour of the matched ad impression. But you also can't put ChatGPT Ads on the same ruler as your existing paid search while running last-click only.

The measurement stack is in place. Pixel and Conversions API send purchases, leads, installs, and sign-ups back and feed optimization. Ten attribution and reporting partners are listed (AppsFlyer, Triple Whale, Adjust, DV Rockerbox, Northbeam, and others), with Hightouch, Tealium, and LiveRamp on the data-ingestion side, Haus, Measured, and WorkMagic on geo incrementality experiments, and DoubleVerify and IAS piloting brand-suitability evaluation. An Event Quality Score rates the quality of your conversion signals, and qualifying advertisers can use Negative Phrases to block contexts they don't want.

And the honest headline: OpenAI publishes no industry benchmarks — no CTR, no average CPC, no acquisition-cost reference. "Is this channel good" has no answer until you have your own comparison data. The method is straightforward: same attribution settings, same time window, ChatGPT Ads against the paid search or social channels you already run.

Two boundaries belong in your compliance checklist. Advertisers don't get access to private conversations. And placement guardrails keep ads out of emotionally vulnerable, sensitive, or otherwise unsuitable contexts. Which means you cannot determine from public material whether a sensitive category is eligible in your market — only OpenAI's sales support can answer that.

4. The Q4 Calendar, and What to Finish Before the Rush

OpenAI's holiday page lays out a specific calendar. These figures are all OpenAI internal data and cannot be independently verified, so that qualifier travels with every number:

  • - More than 1.7 billion retail and ecommerce conversations on ChatGPT each week (internal data, September 1–30, 2026);
  • - Approximately 43% of shopping-related ChatGPT conversations involved comparing and evaluating options (internal data, September 21–October 18, 2026);
  • - U.S. product-search activity on ChatGPT rose 52% during the 2025 holiday season (internal data, Nov 1–Dec 31 versus matched weekdays Oct 4–31);
  • - During Cyber Five 2025, U.S. product-search activity increased 47% on average and 94% on Black Friday versus matched weekdays across the prior four weeks.

OpenAI's recommended order of operations: complete account, conversion measurement, and product catalog setup before holiday demand builds; align campaigns and budgets with key shopping dates; build distinct creative around seasonal products, holiday offers, and shipping deadlines; optimize continuously on early signals. There's also a new-advertiser offer: spend $500, get $500 in ad credit, ending December 31.

On timing, CommerceV3's September 30 Q4 Playbook offers a comparison worth borrowing: ads can go live quickly, while earned AI visibility takes weeks to accumulate, so the work belongs in September and October rather than November. That's a vendor's view rather than a platform rule, but it lines up with Loop's timeline — first campaign in June, budget increased late August, results in early September.

5. What This Evidence Doesn't Cover

Beyond the missing benchmarks, two gaps are worth naming. There are no published rules on which categories placement guardrails restrict, so a DTC brand in a sensitive category has to ask the platform directly. And this article covers only the paid side: no verifiable public experiment establishes a causal relationship between AI organic visibility (GEO) and paid ad performance. You can do both; you cannot assume the first makes the second work better.

A few widely repeated claims did not survive this review. Botify's "5.4x" means AI bot traffic growth in 2025 in one of its posts, and a 5.4x jump in genuine human visits from ChatGPT after OpenAI's agentic commerce launch in another — different metrics, and mixing them produces a completely wrong conclusion. A PerkFuel article claiming "ChatGPT converts better than Google" attributes that to an academic study whose public abstract describes "a descriptive study of 973 ecommerce websites," with no mention of "50,000 transactions" — and PerkFuel itself sells AI-referred traffic. Category-level ChatGPT Ads content is genuinely thin in Chinese too: Tecdo's GEO blog has seven posts, only two of them on ChatGPT Ads and both general explainers; YinoLink's Q4 content is a generic calendar; MeetSocial's focus is gaming.

Where We Fit

WayToGlobal is an agency for DTC e-commerce brands and AI-application companies heading overseas, as described on our site: brand positioning and narrative, content and discovery (including visibility on ChatGPT, Perplexity, and Gemini), market entry and growth, and brand assets. Our case results are qualitative; we don't publish unverified metrics. If you're deciding whether to put budget into ChatGPT Ads, build the comparison this article describes before you commit.

Sources and verification. All OpenAI data and cases come from OpenAI's own site (engineering post, October 21, 2026; milestone announcement, August 31, 2026; Asia-Pacific expansion, September 23, 2026; measurement update, October 5, 2026; holiday page and customer stories, checked October 27, 2026). Academic paper metadata comes from Crossref, Semantic Scholar, and OpenAlex publication records. Tecdo, Botify, and CommerceV3 pages were opened and read directly, with their stated methodologies carried into the text.

Want a category plan for ChatGPT Ads — not a generic one?

Leave your product category, price point and current channels — we'll come back with a concrete take. And if you want new channels opened and run for you, that is exactly what we do.

30 秒看完

  • ChatGPT Ads 上线不到 200 天冲到 10 亿美元年化——近年对品牌开放最快的广告新位。
  • 品类级打法:哪些 DTC 垂类先测、用什么素材、预算怎么摆。
  • 归因是最大的未知数——花钱前先定好口径。

如果你在考虑给 ChatGPT Ads 配一笔 Q4 预算,大概会先搜一圈,然后撞上一堆方向相反的结论:有的说 AI 引流转化率比 Google 还高,有的说 AI 爬虫抓走了你的站却一个访客也没送来,还有的直接把 ChatGPT 广告讲成"升级版关键词广告"。这些说法各自只说对了一件事。

这篇文章把可核的部分拼起来:OpenAI 公开的广告机制、两个跑出结果的电商案例、归因到底怎么算——那些数字之所以互相打架,是因为没人说清口径。

先说三件确定的事。

这是一个能下单的渠道。 OpenAI 在 2026 年 8 月 31 日宣布,ChatGPT Ads 上线不到 200 天就达到 10 亿美元年化收入运行率,已有超过 10 万家广告主投放过;9 月 23 日又新增印尼、马来西亚、菲律宾、新加坡、泰国、越南、台湾七个市场,累计覆盖超过 60 个国家。有个容易忽略的前提:广告只展示给 Free 和 Go 套餐用户,Pro 和 Enterprise 订阅仍是无广告版本。

它的机制和关键词广告不一样。 很多人(包括中文内容里)把它类比成"更聪明的关键词广告",方向没错,但会误导你怎么操作。

它的账目前只能自己算。 OpenAI 没有公布任何行业基准的点击率、转化率或获客成本。所以正确的问法不是"这个渠道好不好",而是"我该拿什么跟它比"。

一、机制:决策单元是会话,不是一条搜索词

ChatGPT Ads 投放链路:从会话状态到展示的六个环节
ChatGPT Ads 投放链路:从会话状态到展示的六个环节

OpenAI 工程团队 2026 年 10 月 21 日公开了广告系统的完整技术说明,这是目前唯一一份一手机制描述。最关键的一条是:ChatGPT 的广告系统不用独立模型和独立代码去跑,它与回答生成并行、互不依赖。这条架构边界有两个后果——广告系统挂了或慢了,ChatGPT 的回答不受影响;不展示广告是"正常且成功"的结局,系统只在广告明显对双方都有利时才投放。

真正决定你怎么操作的是决策单元。传统广告系统从离散事件起手:一次搜索提交、一次页面浏览。而 ChatGPT 里的用户带着目标来,目标在多轮对话里不断变具体。OpenAI 的例子是:用户先问怎么训练跑步,几轮之后才问跑鞋。系统的决策单元因此是会话——广告状态跨轮次、跨会话记忆并更新。

后果很具体:广告素材要对应"用户当下在聊的那件事",而不是一个品类词。钛动科技 10 月 24 日《ChatGPT Ads 的触发逻辑:意图共鸣》把这条机制表述为"当用户侧和广告侧的意图相互匹配的时候,就会展示对应的广告",例子是小户型卧室收纳——用户问怎么省空间,ChatGPT 归纳出"适合小户型的省空间卧室柜",据此匹配广告。

完整链路六段(见图 1):准备检索 query → 宽口径召回 → 资格过滤 → 相关性与质量打分 → 竞价 → 终检与展示。

两个细节值得记住。竞价有明确的最低相关性门槛,编排器只提名上下文合适的候选,定价交给独立的拍卖工具。服务器端投放也不等于用户看得到:广告必须赶在回答生成完成之前抵达,否则这一轮不展示。系统会把失败原因写进日志,所以零曝光是可归因的,不是黑箱。

二、两个已核电商案例,以及它们各自的口径

两个已公开的电商投放案例:Loop Earplugs 与 Newegg
两个已公开的电商投放案例:Loop Earplugs 与 Newegg

OpenAI 官网客户案例页在 10 月 27 日挂着七篇,其中面向消费者电商购物的只有 Loop 和 Newegg 两家,其余是工具类、线索类和开发者产品。案例少不代表效果差——OpenAI 只公开愿意公开的品牌——但可核的电商样本就这么些。

Loop Earplugs(耳塞,DTC 消费品牌)的打法可以直接迁移:每进入一个新市场,先起一个 product-feed 计划,等它跑出结果,再叠加按具体使用场景拆分的非 feed 创意计划。product feed 是一份商品目录(名称、描述、价格、图片、库存),ChatGPT Ads 用它生成展示具体商品的广告;Loop 为每个市场准备对应语言、币种和价格的目录。

跑出来的数字(案例页脚注口径:Loop 的 Ads Manager 数据,2026 年 8 月 18 日至 9 月 8 日;post-click 归因、order-created 事件、排除 view-through 转化):五个在投的 product-feed 计划合计 5.1× ROAS、3.7% post-click 转化率;欧盟"concerts and music"非 feed 计划 1.8× ROAS。测试期结束后它把欧盟 feed 计划日预算提到约 2.5 倍,随后在瑞典、挪威、瑞士开新计划。

Loop 自己的总结更值钱:非 feed 创意从泛泛的"earplugs"改成"festival earplugs"这种具体场景词之后表现更好。用他们的话说,"有人描述的是一个场景,不是一个产品品类",所以围绕场景建的广告才有东西可以对上。

Newegg 走另一条路:常青 product-feed 计划打底,叠加季节性计划,促销期集中加预算。结果是 28 天跨计划 3× ROAS,两年一度的 Fantastech Sale 专项计划两周 7× ROAS。市场经理 Alec Shao 还提到,相比其他渠道,来自 ChatGPT 的访客"更可能转化并下单"。但要说清楚:Newegg 是多品类零售商不是 DTC 品牌,这个案例页也没披露归因窗口——3×、7× 与 Loop 的 5.1× 不能直接横向比。

这正好说明为什么这个渠道得自己建对照。OpenAI 10 月 5 日的测量更新里提到,DV Rockerbox 测得 WeightWatchers 在 ChatGPT Ads 上的归因 CPA 比其混合付费搜索基准低 15.3%;WorkMagic 的 geo-lift 研究估计 Dose 的增量订单是 last-click 归因捕获的 2.3 倍。前者是"跟自己比",后者说明只按 last-click 看会低估——但 2.3 倍是单一案例的估计。

三、怎么算账:归因窗口是你自己调的

ChatGPT Ads 的归因三层结构与已知风险
ChatGPT Ads 的归因三层结构与已知风险

这是全篇最实际的一节。ChatGPT Ads 的归因设置有两个可调项:点击归因窗口由广告主自选,按你的购买周期定长度;可以额外勾选 1 天的 view-through 窗口,捕捉没有点击、仅曝光之后发生的转化。一句限定必须照抄:这些设置只影响报告里的归因,不改变投放优化。

这不是小事。只按 last-click 看这个渠道,你会同时低估它——OpenAI 在一篇以深层漏斗结果为目标的全球计划早期分析里发现,52.7% 的合格 1 天 view-through 转化发生在匹配广告曝光后 1 小时内——但你也无法在只做 last-click 的情况下把它和现有付费搜索渠道放在同一把尺子上比。

测量工具链已经成型:Pixel 和 Conversions API 把购买、留资、安装、注册等动作回传并用于优化;归因与报告侧有十家合作方(AppsFlyer、Triple Whale、Adjust、DV Rockerbox、Northbeam 等),数据接入侧有 Hightouch、Tealium、LiveRamp,增量实验侧有 Haus、Measured、WorkMagic,品牌安全评估在和 DoubleVerify、IAS 做试点。另有一个 Event Quality Score 给转化信号打分,符合条件的广告主可用 Negative Phrases 屏蔽不希望出现的语境。

所以"这个渠道好不好",在你没有自己的对照数据之前没有答案。做法就是用同一套归因口径、同一时间窗口,拿 ChatGPT Ads 对上你现有的付费搜索或社媒渠道。

两件官方划了边界的事该写进合规清单:广告主拿不到用户的私人对话;放置护栏会把广告挡在情绪脆弱、敏感或不适宜的语境外。这意味着某个敏感品类能不能投,公开材料判断不了,只能问平台。

四、Q4 的时间表,以及节日前该做完的事

OpenAI 的假日专题页给了一份相当具体的时间表。这些数字全部是 OpenAI 内部数据,无法第三方复核,引用时必须带这个限定:

  • - 每周超过 17 亿条零售与电商类对话(内部数据,2026-09-01 至 09-30);
  • - 约 43% 的购物相关对话涉及比较与评估选项(内部数据,2026-09-21 至 10-18);
  • - 2025 假日季美国站内商品搜索活动同比 +52%(内部数据,对比匹配工作日);
  • - Cyber Five 2025 期间,美国站内商品搜索活动平均 +47%、黑五当天 +94%(内部数据,对比前四周匹配工作日)。

OpenAI 给的准备顺序是:假日需求起来前把账户、转化衡量和商品目录设置做完;对齐关键购物日期的预算与优化;按季节商品、节日优惠和发货截止日做不同创意概念;用早期信号持续优化。开户还有个促销:满 500 美元投放送 500 美元额度,12 月 31 日截止。

关于节奏,CommerceV3 在 9 月 30 日《ChatGPT Ads + AI Search:Q4 Playbook》里给过一个可参考的对比:广告可以很快上线,earned(自然获得)的 AI 可见性需要数周积累,所以该做的是 9–10 月就开始。这是服务商观点、不是平台规则,但它和 Loop 的时间线对得上——第一个计划 6 月上线,8 月底加投,9 月初才拿到结果。

五、这份材料覆盖不了什么

除行业基准外还有两个缺口。一是哪些品类受放置护栏限制没有公开细则,落在敏感语境的 DTC 品类只能直接问平台。二是本文只覆盖付费投放这一侧;AI 自然可见性(GEO)和付费广告之间有没有因果关系,目前没有可核的公开实验能证明——两者可以同时做,但不能假设做了前者后者就一定变好。

还有几条流传很广的说法,本轮核对后降了级。Botify 那个"5.4 倍",同一篇文章里指的是 AI 爬虫流量在 2025 年的增长,另一篇页面指的是 OpenAI 上线 agentic commerce 后来自 ChatGPT 的真实人类访问量增长 5.4 倍——口径不同,混用会得出完全错误的结论。PerkFuel 那篇"ChatGPT 转化率超过 Google"归因到一篇学术研究,但该研究的公开摘要是"对 973 个电商网站的描述性研究",没有"5 万笔交易"的说法,而且 PerkFuel 本身就是卖 AI 流量的服务商。中文侧的品类实操内容也确实稀薄:钛动 GEO 博客共七篇,ChatGPT Ads 相关只有两篇通用向科普;易诺的"Q4"只有一份通用日历;飞书深诺的重心在游戏垂类。这是"我们检索到的公开内容里没有",不是"没人做"。

我们是谁

WayToGlobal 是帮 DTC 电商品牌与 AI 应用公司出海的代理,公开定位写在官网:品牌定位与叙事、内容与发现(含 ChatGPT / Perplexity / Gemini 可见性)、市场进入与增长、品牌资产。我们的案例结果只做定性描述,不发布未经验证的数字。要不要在 ChatGPT Ads 上押一笔预算,先做本文那套"用同一口径自建对照"。

引用与核对:本文 OpenAI 数据与案例均来自 OpenAI 官网(工程文 2026-10-21、里程碑公告 2026-08-31、亚太扩展 2026-09-23、测量更新 2026-10-05、假日专题与客户案例页,核对日 2026-10-27)。完整事实表见编辑内部留档。

想要一份品类级的 ChatGPT Ads 投放方案,而不是通用模板?

留下品类、客单价和当前投放渠道,我们回来给你具体判断;新渠道需要代开户、托管投放,正是我们的主业。