OpenAI Tests Sponsored Agents in ChatGPT, Adds HubSpot and Shopify Ad Integrations

OpenAI today introduced Sponsored Agents — a fundamentally different ad format that replaces the browser redirect with a branded AI conversation inside ChatGPT — alongside a natural-language Ads Manager plugin and the first commerce and CRM integrations for its seven-month-old advertising platform. The announcement, published September 16 on OpenAI's website, escalates ChatGPT Ads from an ad-slot experiment into something closer to a full-stack commerce and marketing channel, one where the click no longer sends users somewhere else but keeps them inside the session with the advertiser's own AI agent.

What a Sponsored Agent Actually Does — and What It Does Not Do

Under the existing ChatGPT Ads model that launched in February 2026, sponsored placements appear as labeled cards below the AI's responses, visible to Free and Go tier users in the US and a growing number of international markets. Clicking one takes the user to an advertiser's website, the same mechanism that has defined digital advertising since the mid-1990s.

Sponsored Agents change that mechanic. When a user sees a relevant ad and chooses to engage — the interaction is opt-in — clicking opens a separate, clearly labeled conversation thread with a business-sponsored AI agent rather than a new browser tab. That agent stays inside ChatGPT but is architecturally isolated from both the original conversation the user was having and from ChatGPT's ordinary model responses. OpenAI's announcement uses the example of a shopper who spots a dining table in an ad but wants to know whether it fits a specific room, how many people it seats, or what finish care requires before committing to a click. Those questions can now be answered by the brand's own agent without the user ever leaving ChatGPT.

The isolation design is not incidental. OpenAI has maintained since the February 2026 ad launch that its chat model's responses are completely separate from the advertising system — advertisers cannot influence what ChatGPT says organically, and the ad infrastructure does not have access to raw conversation content. The company's help center states explicitly that ads "do not influence ChatGPT's answers" and that "advertisers do not have access to your chats, chat history, memories, or personal details." The Sponsored Agent architecture extends this principle: the branded conversation runs in its own labeled thread, and the user chooses to enter it rather than being routed into it automatically.

Wayfair and Angi, the home services marketplace (NASDAQ: ANGI), have publicly confirmed participation in the pilot. Angi said in a press release that an Angi-branded sponsored agent may appear in relevant ChatGPT conversations about home services, and that homeowners can share project details and, if needed, enter Angi's existing service-request flow to connect with local professionals. Angi co-founder Angie Hicks said the pilot lets homeowners "go directly from discussing a home project in ChatGPT to connecting with a skilled local pro who can help." Wayfair, according to reporting by ContentGrip, is using the pilot to explore whether conversational commerce helps shoppers make more confident decisions, while applying guardrails around product accuracy, transparency, and handoffs to human customer service when needed.

The pilot is currently limited to select advertisers in the United States.

How the Agent Infrastructure Is Configured

An advertiser setting up a Sponsored Agent configures it across several inputs: an automated crawl of their website that generates a business profile and common customer questions; a product feed containing inventory, pricing, and item details; optional Model Context Protocol (MCP) tool connections for live data access; and lead-generation forms or service-request flows that represent the conversion endpoint.

Read more: OpenAI Agents API launches in public beta with managed Codex harness for all developers

That configuration layer is meaningful because it determines what the agent actually knows. An agent drawing on a live product feed can confirm real-time pricing and availability; one operating only on a static site crawl may answer with stale information. Wayfair's explicit emphasis on "product accuracy" guardrails reflects a real operational risk: an AI agent that hallucinates dimensions, finishes, or availability creates the same category of harm as a misleading ad, but faster and at conversational scale.

A New Ads Manager: Campaigns Built in Plain Language

Alongside Sponsored Agents, OpenAI is rolling out an Ads Manager plugin — available in ChatGPT and in Codex, OpenAI's coding environment. The plugin takes a plain-language description, a website URL, or a campaign brief and converts it into a structured campaign ready to launch in Ads Manager. It can also update live campaigns, troubleshoot delivery problems, analyze performance metrics, and surface prioritized recommendations.

The plugin requires advertiser confirmation before any change is applied, a design choice that explicitly positions it as an assistant rather than an autonomous operator. Every recommended update is previewed and approved. This matters practically: an advertiser using natural language to manage campaigns needs a clear separation between "the AI suggested this" and "the AI did this."

A companion feature announced today adds AI-assisted creative directly inside Ads Manager. When drafting an ad, advertisers can now receive copy and imagery suggestions generated from their landing page and campaign objective. OpenAI also introduced opt-in AI-powered text customization that adjusts existing headlines and descriptions to better fit the context of a specific conversation, and automatically translates ad copy into the user's preferred language. Neither feature publishes anything automatically; human review precedes any live change.

HubSpot and Shopify: Embedding ChatGPT Ads Where Advertisers Already Work

Two integrations announced today address a structural friction point in reaching smaller advertisers: the requirement to manage a new ad platform from a separate interface.

HubSpot becomes OpenAI's first CRM partner for ChatGPT Ads. Businesses already managing customer data in HubSpot can now connect a ChatGPT Ads account and create campaigns, track performance, and follow up on leads without leaving HubSpot's workspace. The integration uses HubSpot's existing customer context — contact history, pipeline stage, company data — to inform ad targeting and lead qualification.

Shopify becomes the first ecommerce partner. A free ChatGPT Ads app, developed by OpenAI, is available today in the Shopify App Store for US-based merchants. The app automatically pulls the merchant's product catalog through Shopify Catalog, meaning merchants can begin building and tracking ChatGPT Ads campaigns without a separate product-feed upload process. The app's international launch, covering all markets where ChatGPT Ads are available, is scheduled for September 23, 2026.

The significance is partly organizational. Shopify had already integrated its catalog with ChatGPT's broader commerce layer through Agentic Storefronts — a feature that activated for all eligible merchants in March 2026 and made Shopify product inventory discoverable through ChatGPT's organic product-recommendation responses. The September 16 app converts that organic presence into a paid advertising surface, giving merchants a direct campaign-creation pathway from within their existing Shopify admin.

For HubSpot, the integration extends the platform's role from CRM and marketing automation into a new paid-media channel without requiring users to learn a separate campaign interface.

Read more: OpenAI ships WebMCP site tools: ChatGPT now calls your app instead of scraping it

How ChatGPT Ads Differ From Google, Meta, and Bing

Sponsored Agents are the sharpest differentiator ChatGPT Ads has from every competing platform, but the context matters for understanding how new the category actually is.

Google's AI Mode — the conversational search surface powered by Gemini — has carried sponsored placements since launching shopping ads with Direct Offers in February 2026. At Google Marketing Live in May 2026, Google announced Conversational Discovery Ads and Highlighted Answers, both designed to embed paid placements inside AI-generated conversational search results. Google also previewed a custom in-ad AI agent capability for AI Mode, though that format's availability for general advertisers remains limited. These formats run within Google's existing Performance Max and AI Max infrastructure; they live inside the search results page rather than opening a separate branded conversation thread.

Meta does not have native in-chatbot advertising inside Meta AI. The company uses conversations within Meta AI as a targeting signal to improve ad placement in the regular Facebook, Instagram, and WhatsApp feeds — a fundamentally different architecture that does not expose the chatbot session to sponsored content.

Microsoft's Bing runs "Sponsored Answers" within Copilot conversations, placing sponsored content units within chatbot exchanges. The format is more interactive than a traditional display ad but does not open a separate branded agent thread; it surfaces sponsored information within Copilot's own response.

The result is that Sponsored Agents, as currently designed, have no direct structural equivalent at a major platform. The format is at an early, US-only pilot stage and its commercial effectiveness is unproven, but the architectural distinction is real: a click opens a branded AI service interaction rather than a URL.

What the Platform Looks Like Seven Months In

The September 16 announcements come as ChatGPT Ads establishes early-stage but meaningful scale. The platform launched February 9, 2026, for Free and Go tier users in the United States with a limited set of major launch advertisers including Target, Albertsons, and Williams-Sonoma. It reached $100 million in annualized revenue in under 60 days, according to a company-reported milestone confirmed by multiple outlets in late March 2026. On August 31, OpenAI announced the platform had crossed $1 billion in annualized revenue run rate, roughly 200 days after launch, with tens of thousands of advertisers now active in more than 40 countries.

For context: Google's AdWords took approximately four years to reach $1 billion in annual revenue. The speed of ChatGPT Ads' ramp reflects both the scale of ChatGPT's user base — approximately 900 million weekly active users as of February 2026, a figure OpenAI has reported but not updated since — and the relative novelty of a platform where intent is explicit and contextual without requiring keyword bids.

OpenAI has stated a $2.5 billion advertising revenue target for 2026 and a longer-range ambition of $100 billion by 2030, according to investor-presentation reporting. Reaching $2.5 billion in recognized 2026 revenue from a $1 billion annualized run rate in late August would require significant acceleration through the fourth quarter. The $2.5 billion figure is company-claimed and investor-directed; it should not be treated as a confirmed forecast.

Privacy Architecture and the Data-Ownership Question

OpenAI's stated privacy structure for ChatGPT Ads maintains several hard separations: advertisers cannot access individual conversation content, ads do not influence ChatGPT's organic responses, and reporting is aggregated rather than user-level. Free users can opt out of ad personalization and clear their ad interaction history. Plus, Pro, Business, Enterprise, and Edu subscribers do not see ads at all, and ChatGPT does not show ads to accounts identified as belonging to users under 18.

An April 2026 privacy policy update formalized advertising data practices that had not been fully disclosed at launch. Marketing cookies are now enabled by default for Free-tier US users, with opt-out available in settings. OpenAI shares limited user identifiers — cookie IDs and device IDs — with marketing partners. Conversations and private content are not shared. Miranda Bogen of the Center for Democracy and Technology warned at the January 2026 announcement that introducing personalized ads starts OpenAI "down a risky path previously taken by social media companies."

The Sponsored Agent format introduces a new variant of this question. When a user shares budget constraints, room dimensions, or service preferences with a Wayfair or Angi-branded agent inside ChatGPT, who owns that interaction data? The current advertiser-reporting architecture answers "not the advertiser, in raw form" — advertisers receive aggregated performance metrics. But OpenAI's own platform receives the substance of those conversations, which may inform its contextual matching and future targeting. This data-ownership dynamic becomes more consequential as Sponsored Agents mature: the richer the branded interaction, the more intent data flows through OpenAI's infrastructure rather than through the advertiser's own analytics.

What the Sponsored Agent Pilot Has Left Unanswered

Sponsored Agents are in an early US test with select advertisers. Several questions about the format's durability and commercial viability remain unanswered.

Agent accuracy and brand liability are unresolved. If a Sponsored Agent returns incorrect information — wrong dimensions, unavailable inventory, inaccurate pricing — the harm is similar to a misleading advertisement but harder to audit because the error occurs in a live conversational exchange rather than in static ad copy. Wayfair's guardrails signal that early advertisers are treating agent accuracy as a brand-risk issue, not just a customer-experience footnote.

Measurement remains structurally limited. ChatGPT Ads does not provide individual-level conversion attribution comparable to Google or Meta. Conversions API and pixel-based tracking, launched in May 2026, cover post-click events but depend on server-to-server data and are subject to the same signal-loss constraints that affect all advertising measurement in an era of cookie deprecation and privacy regulation. For Sponsored Agents specifically, where the "conversion" may be a lead form submitted or a service request initiated inside ChatGPT rather than a purchase on an external site, the attribution model is still being defined.

User acceptance is unproven at scale. The pilot is limited precisely because OpenAI needs to verify that Sponsored Agents do not erode the trust users place in ChatGPT's answers. The company has consistently cited no negative impact on consumer trust metrics and low dismissal rates since the February launch, but those measurements cover the existing card-below-response format, not the more immersive experience of a branded agent conversation. Whether users distinguish the opt-in nature of Sponsored Agents from a sense that ChatGPT is routing them to advertisers remains an open empirical question.

The Click Has Been Redefined — But Only in a Pilot

The fundamental architecture of digital advertising — you click an ad, you leave the platform, you arrive at the advertiser's destination — has held for thirty years because every major publisher found it simpler to redirect than to host the transaction. Sponsored Agents represent OpenAI's bet that ChatGPT is different: that its users arrive with sufficiently specific intent, and trust the platform sufficiently, that a branded conversation inside the session produces better outcomes for users and advertisers than a redirect.

Whether that bet proves commercially durable depends on whether agent accuracy is solvable at ecommerce scale, whether advertisers find the data-ownership tradeoff worthwhile, and whether OpenAI's broader ambition to become a commerce infrastructure layer — not merely a publisher selling placements — runs into regulatory or platform-access headwinds as its role in consumer transactions deepens. The HubSpot and Shopify integrations are the clearest signal of that ambition: they position ChatGPT Ads not as a channel companies manage alongside their existing workflows, but as infrastructure embedded inside the tools where commercial decisions already happen.