OpenAI announced a new set of ChatGPT Ads features on September 16, 2026. The update combines Sponsored Agents, natural-language campaign management, AI-assisted creative work, and integrations with HubSpot and Shopify. For marketers, the bigger change is not another ad format. It is a workflow in which an ad can lead into a clearly labeled business conversation and campaign operations can begin from a prompt.
This guide separates the announced features from the practical questions marketers still need to answer: who can use them, where human review belongs, what should be measured, and how to compare them with existing AI advertising tools.
ChatGPT Ads and Sponsored Agents at a glance
In the official ChatGPT Ads announcement, OpenAI describes four connected changes:
- Sponsored Agents let a user choose to start a clearly labeled conversation with a business after seeing an ad.
- An Ads Manager plugin in ChatGPT Work can create, update, and analyze campaigns from natural-language instructions.
- Ads Manager can suggest copy and imagery based on a landing page and campaign objective.
- HubSpot and Shopify integrations connect campaign work with customer and catalog data.
Sponsored Agents were being tested with selected advertisers in the United States at launch. The Shopify app was available to U.S. merchants, with international availability planned for markets where ChatGPT Ads operates. Availability can change, so businesses should verify the current product page before planning a launch.
Why conversational advertising is different
A traditional ad usually sends the user to a landing page, product page, or form. A Sponsored Agent adds an optional conversational step. A shopper could ask whether a product fits a room, compare materials, clarify delivery, or understand which plan suits a specific need before visiting the business website.
That can reduce the distance between interest and a useful answer. It also creates a new quality bar. The agent needs approved product information, a clear identity, correct boundaries, and a safe handoff to the company's website or team. A persuasive answer built on outdated catalog data can damage trust faster than an ordinary static ad.
OpenAI says the sponsored conversation is distinct from ChatGPT's independent answers and separate from the original user conversation. That separation should remain visible in the user experience, measurement, and data policy.
The new AI campaign workflow
The campaign workflow begins with a prompt, but it should not end there. A practical process looks like this:
- Prepare the source material. Confirm the landing page, offer, audience, approved claims, exclusions, and measurement goal.
- Generate a campaign draft. Use natural-language instructions to propose structure, copy, imagery, targeting, or translations.
- Review claims and creative. Check pricing, availability, legal language, product imagery, brand voice, and regional requirements.
- Launch a controlled test. Start with a narrow audience and a clear comparison against an existing campaign.
- Analyze useful outcomes. Measure qualified visits, completed actions, helpful conversations, and downstream value rather than only clicks.
- Feed approved learning back. Keep a record of which changes were accepted and why, instead of letting the system silently rewrite the campaign.
Tools such as AdCreative.ai and Jasper remain relevant for teams that want dedicated creative generation or broader marketing content workflows. The content marketing category provides more alternatives, while ChatGPT is the directory entry for OpenAI's general assistant.
What marketers should keep under human control
OpenAI's announcement says advertisers can review and edit suggested copy and imagery before adding them to a campaign. That review step is not a formality. It is where teams protect accuracy, brand safety, and legal compliance.
Keep these decisions explicit:
- Which product claims are approved and which need evidence.
- Whether an image accurately represents the product, package, or result.
- Which audience exclusions and regional restrictions apply.
- How translated copy is reviewed for meaning, not only grammar.
- What the Sponsored Agent may answer and when it must decline or escalate.
- Which customer and catalog fields can be used for campaign work.
- Who approves budget, launch, and major optimization changes.
AI can accelerate the first draft and surface patterns in performance data. It should not be the only authority for claims, eligibility, or spend.
Sponsored Agents need a knowledge and safety plan
A business-sponsored conversation is useful only when the agent has dependable information. Build the knowledge layer from approved product pages, policies, current inventory signals, and support material. Define which source wins when two documents conflict, and attach an update process to every fast-changing field.
The agent also needs a refusal boundary. It should not invent compatibility, make unsupported performance guarantees, or turn a product conversation into advice outside the business's competence. High-impact areas such as health, finance, or legal services need additional review and may require a human before any recommendation.
Conversation logs can include personal preferences and purchase intent. Businesses should decide what is retained, what enters the CRM, how consent works, and how a user can request deletion. Integration is valuable when it reduces duplicate work, but automatic data movement should never be invisible.
How to measure an AI advertising pilot
Start with a test that can answer a specific question. For example: does a sponsored conversation help users choose the correct plan, or does prompt-assisted campaign creation reduce production time without increasing correction work?
Useful measures include:
- Conversation start rate after an ad impression or click.
- Percentage of conversations that reach a useful answer.
- Escalation, refusal, and unsupported-answer rates.
- Qualified site visits and completed downstream actions.
- Human review time per campaign draft.
- Number and severity of corrections before launch.
- Cost per qualified outcome, including tool and review costs.
Avoid optimizing solely for longer conversations. A concise answer that sends a qualified user to the right product may be better than a long exchange that never resolves the question.
Workflow automation still has a role. Zapier, n8n, or Activepieces can route approved leads, create review tasks, or synchronize campaign results without giving a generative system unrestricted access to every marketing system.
Frequently asked questions
Are Sponsored Agents available to every advertiser?
No. OpenAI said the feature was being tested with selected U.S. advertisers at the September 16, 2026 launch. Check current eligibility before building a campaign around it.
Can ChatGPT create and manage campaigns from prompts?
OpenAI announced natural-language campaign creation, updating, and analysis through an Ads Manager plugin in ChatGPT Work. Advertisers remain responsible for review, approval, budget, and policy compliance.
Do AI ad tools replace creative teams?
They can accelerate drafts, variants, translation, and analysis. Teams still need human judgment for positioning, evidence, brand voice, legal review, and the final decision to publish.
What should a first Sponsored Agent answer?
Start with a narrow set of high-frequency product questions backed by current approved data. Add a visible website handoff and a human escalation path before expanding the knowledge scope.
Conclusion
ChatGPT Ads combines conversational discovery with prompt-assisted campaign operations. The opportunity is a shorter path from interest to a useful answer; the risk is scaling inaccurate claims or opaque data movement just as quickly.
Treat Sponsored Agents as a governed customer touchpoint, not an unrestricted chatbot. Compare dedicated tools in the content marketing directory, define one measurable pilot, and keep approval with the people accountable for the campaign.