Yes on the organic answer, no on the surface around it. OpenAI began testing Sponsored Agents on September 16, 2026: a labelled, separate chat with an advertiser's agent on the Free and $8-a-month Go tiers, where Similarweb found 26% of answers already carry a sponsored placement. Heather Wardle's Games Without Frontiers supplies the test: ask what conditions the design creates, not whether it is an ad.

Last week a founder I work with sent me a screenshot of a ChatGPT answer about payroll providers with a sponsored card under it and asked whether the answer above the card was still honest. I think it is. But that is the wrong question, and the reason it is the wrong question is the useful part.

The facts first. On September 16, 2026, OpenAI announced it is testing Sponsored Agents: after a ChatGPT ad appears, a user can click into a clearly labelled conversation with an agent sponsored by the advertiser, ask follow-up questions and follow a link to the business's site. OpenAI says that conversation is "distinct from ChatGPT's independent answers and separate from the original conversation," and the test is limited to select advertisers in the United States. The same announcement wired ChatGPT Ads into HubSpot and Shopify as the first CRM and ecommerce partners, with the Shopify app going international from September 23. The context matters more than the feature. Ads exist only on the Free tier and the $8-a-month Go tier; OpenAI's own ads principles state that Plus, Pro, Business and Enterprise do not include them, and that "ads do not influence the answers ChatGPT gives you." The pilot opened on February 9, 2026, self-serve buying on May 5, sponsored placements reached 31 European markets from August 24, and per ppc.land's timeline Digiday reported a $1 billion annualised run rate by August 31, under 200 days in. Similarweb's August 17 data found 26% of ChatGPT responses shown to Free and Go users carried a sponsored placement. What is still undisclosed, as ppc.land noted: what the advertiser receives from a sponsored conversation, which model runs the agent, who writes its instructions, and how the unit is priced.

The book I keep reaching for here is not a business book. Heather Wardle's Games Without Frontiers is a short academic history of how gambling mechanics seeped into video games, and its central move is the one founders need. Wardle argues that the loud debate over whether a loot box legally "is gambling" misses the point; the questions that matter are about the conditions a design creates, who it is optimised for, and what is not disclosed, because loot-box odds are set by the producer and usually hidden, and machine learning is already used to encourage repeat purchases. Her teenagers understood the business model perfectly, "the game was free so they need a way to earn money," and still felt compelled to spend: savvy, and subservient. Swap "is it gambling?" for "is it an ad?" and you have the ChatGPT debate. The label is real and I believe the answer-independence principle as stated. The conditions are what changed: a free product funded by advertisers, a conversational surface that a rival executive called the fastest-growing engagement format in advertising, and an agent whose instructions you cannot read, built by the party that wants the sale. Wardle's history also says every new communication technology, telegraph to mobile, expanded the reach of the older commercial mechanism. A chat that answers your follow-up questions is that step for advertising.

So the protocol I now run, which I call the sponsored-surface rule, is about conditions, not labels:

  1. Research on an ad-free tier, always. Vendor, hiring, legal and financial questions go through a Plus, Pro or Business seat, or through Claude or Perplexity, never a Free or Go account. Twenty dollars a month is a rounding error against a bad tooling decision.
  2. Never take a decision inside a sponsored chat. If I open one at all, it is to extract a spec sheet or a price, which I paste back into my own conversation and ask the model to argue against.
  3. Ask for the disqualifier, not the recommendation. "What would make this vendor wrong for a 12-person company?" is a question no sponsored agent is instructed to answer well.
  4. Verify the number at the source. Any price, limit or integration claim from a chat gets checked on the vendor's own pricing page before it enters a memo, sponsored or not.

Where I would push back on my own alarm. Nothing published this week suggests the organic answer is being steered, and a separate, opt-in, labelled conversation is a cleaner design than a search page with four ads above the first result. The 26% figure is Similarweb's estimate of ad presence, not a measure of influence. And a sponsored agent that answers "does it fit my room" honestly is genuinely useful; the risk is not that it lies but that a founder in a hurry treats a salesperson with perfect recall as a researcher. The failure mode is on our side, which is exactly what Wardle's teenagers could have told us.

Sources: OpenAI, "Reimagining advertising with AI," September 16, 2026 (openai.com/index/reimagining-advertising-with-ai); OpenAI, "Our approach to advertising and expanding access to ChatGPT," ads principles and tier details (openai.com/index/our-approach-to-advertising-and-expanding-access); ppc.land, "OpenAI lets advertisers run ChatGPT ads from HubSpot and Shopify," September 17, 2026, including the Similarweb, Digiday and rollout timeline (ppc.land); TechWyse, "OpenAI Adds Sponsored Agents to ChatGPT Ads Platform," September 17, 2026 (techwyse.com); Heather Wardle, Games Without Frontiers? Socio-historical Perspectives at the Gaming/Gambling Intersection (Palgrave, 2021).

Related: How to Find Your Passion · Best Self-Improvement Books · How to Make Better Decisions · What University Will Not Teach You