Until this week, an ad in ChatGPT behaved like an ad anywhere else. It sat next to the answer, you clicked it or you did not, and if you clicked you left for the advertiser's website. On 16 September, OpenAI started testing something different. It is called Sponsored Agents: you see an ad, you tap it, and instead of a landing page you get a separate, labelled conversation with an AI agent that the business has set up. You can ask it about sizing, compatibility, delivery, whether it works with the thing you already own. When you are ready, it hands you a link to buy.
In other words, the ad no longer just points you at a salesperson. The ad is the salesperson.
This matters for anyone who spends money on marketing, because ChatGPT advertising is no longer an experiment. It crossed a $1 billion annualised run rate in under 200 days, it runs in more than 40 countries, and it is now wired directly into Shopify and HubSpot. Before you move budget into it, though, it is worth knowing exactly who sees these ads, what a click costs, and what your conversion rate has to be for the numbers to work. That is what this piece is for.
What OpenAI launched on 16 September
Four things arrived together, and only one of them got the headlines.
| Launch | What it does | Where |
|---|---|---|
| Sponsored Agents | Tap an ad, get a separate labelled chat with the brand's agent, then a link to the advertiser's site | Test with select advertisers, US only |
| AI campaign creation | An Ads Manager plugin in ChatGPT Work turns a website or brief into a campaign, suggesting copy and images; a human must approve before anything runs | Advertisers |
| Shopify app | US merchants install a ChatGPT Ads app; products already in the Shopify Catalog feed straight into campaigns | US now, international from 23 September |
| HubSpot integration | Create campaigns, track performance and follow up on leads from inside HubSpot | HubSpot customers |
The Shopify and HubSpot pieces are the quiet heavyweights. Together they mean a small online store or a B2B firm can go from "never advertised in ChatGPT" to "campaign live" inside tools they already use, without an agency. That is how an ad platform stops being something big brands test and becomes something ordinary businesses spend on every month.
How big this already is
The growth numbers explain why OpenAI is pushing so hard.
| Metric | Figure |
| Ads pilot began | February 2026, US, Free and Go tiers |
| Self-serve Ads Manager, all US advertisers | 5 May 2026, CPC bidding added alongside CPM |
| Annualised run rate | $1 billion, reached by 31 August, in under 200 days |
| Monthly revenue at that point | roughly $83 million |
| Revenue actually booked in the first eight months | an estimated ~$330 million |
| Markets | 40+ countries; Europe opened in 31 markets in August, self-serve from 31 August |
| Advertisers | tens of thousands |
| 2026 target | reported at roughly $2.3-2.5 billion |
| 2030 target | $100 billion |
Two honest notes on that table. First, a run rate is last month multiplied by twelve, not a year of revenue; the money actually booked so far is closer to a third of a billion dollars. Second, the 2030 target is enormous. Digiday calculated it needs something like 216% compound annual growth to get there. For context on why OpenAI needs it anyway, remember the company that lost about $38.5 billion last year while chasing a $1 trillion IPO. Advertising is the revenue line that can scale to the size of the losses. Subscriptions alone probably cannot.
Who actually sees your ad
This is the part most "should you advertise on ChatGPT" guides skip, and it changes the whole calculation.
| ChatGPT user | Sees ads? |
| Free, logged-in adult | Yes |
| Go (the cheap paid tier) | Yes |
| Plus, Pro | No |
| Business, Enterprise, Education | No |
| Predicted to be under 18 | No |
| Anyone asking about health, mental health or politics | No ads near those topics |
Read the middle rows again. The people paying $20 or $200 a month, the heaviest users and on average the ones with the most money to spend, never see your ad. Neither does anyone using ChatGPT through their employer's Business or Enterprise account, which rules out most of the B2B decision-makers you might most want to reach while they are at work. Free and Go users can also opt out in exchange for fewer daily messages.
That is not a reason to avoid the platform. The free tier is vast, and ChatGPT's weekly audience is approaching a billion people. But it means you are buying a specific audience: consumers, on personal accounts, often on free plans. For a grocery chain or a mid-priced consumer brand that is exactly right, which is probably why retail and grocery took 44% of all ad impressions in the first two weeks of the pilot. For a company selling $50,000 software contracts, it is a much weaker fit than the headline audience size suggests. (If you want to understand the tier split from the user side, I broke down what the $20 and $200 plans actually get you.)
What a click costs, and the break-even math
Pricing is auction-based, so there is no rate card. The figures that come up repeatedly in advertiser guides for mid-2026 are roughly $25 to $60 per thousand impressions on CPM buying and $3 to $5 per click on CPC, and OpenAI's own help centre suggests starting CPC bids in that $3 to $5 range. Treat those as starting points, not guarantees. Your category and your competition will move them.
The useful question is not "is $4 a click expensive?" It is "what conversion rate do I need for $4 a click to pay?" That depends on how much profit one sale brings you.
| Your gross profit per sale | CPC $3: conversion needed to break even | CPC $5: conversion needed to break even |
| $20 | 15% | 25% |
| $50 | 6% | 10% |
| $100 | 3% | 5% |
| $300 | 1% | 1.7% |
The rule behind that table is simple: break-even conversion rate = cost per click รท profit per sale. Anything above that line makes money; anything below it loses money on every click.
What it shows is uncomfortable for small-ticket sellers. If you make $20 of profit on an order, you need one in every four to seven clickers to buy, which is a conversion rate most online stores never see from any channel. If you make $100 or more per sale, the numbers start to look normal. ChatGPT ads, at today's prices, are a channel for higher-margin products and considered purchases, not for impulse items with thin margins. Run your own margin through that formula before you run a single campaign.
Why Sponsored Agents could change the conversion math
Here is where the new format matters. A normal ad's job ends at the click. Everything after that, whether the landing page answers the buyer's question, whether they find the right size, whether they give up because shipping costs are hidden, is lost conversion you already paid for.
A Sponsored Agent moves that whole middle part of the sale into the conversation. The person who clicked can ask the exact question that would otherwise have sent them back to search results. If the agent is good, it answers, reassures, and hands over a link at the moment the buyer is ready. That is how an ad channel can lift conversion rather than just traffic, and on the table above, a few points of conversion is the difference between losing and making money.
The catch sits in the phrase "if the agent is good".
Your sponsored agent's promises are your promises
A Sponsored Agent is your business talking, automatically, to strangers, at scale. It will quote prices, describe return policies, claim compatibility, and sometimes improvise when it does not know. Everything it says is said in your name.
This is not hypothetical. In a well-known 2024 case, a Canadian tribunal ordered Air Canada to honour a bereavement-fare refund that its website chatbot had simply made up, rejecting the airline's argument that the bot was responsible for its own words. The lesson carries straight over: when a customer relies on what your agent told them, you own the consequences, and paying OpenAI to put that agent in front of more people only raises the stakes.
So before a Sponsored Agent goes live, test it the way a customer would. Ask it the awkward questions: the return you do not actually accept, the size you do not make, the delivery date you cannot hit, the discount that does not exist. See what it promises. I have written about how rarely anyone walks up to their own bot as a customer and what mystery-shopping an airline's chatbot turned up. Those lessons apply double to an agent you are paying to amplify.
You can buy the ad, but not the answer
One more rule OpenAI has committed to publicly: ads do not influence ChatGPT's answers, and conversations stay private from advertisers. Sponsored Agents are labelled, run in a separate conversation, and are not supposed to change, rank or become part of ChatGPT's own response.
Take that at face value and it creates two separate games. The paid game is the ad slot and the Sponsored Agent. The organic game is whether ChatGPT names your business when someone asks for a recommendation in the first place. The second cannot be bought, and for many businesses it is worth more, because it is free, it recurs on every similar question, and it carries the credibility of an answer rather than a label. I looked at that second game in why your name is usually missing from the AI's answer, and if you want a quick read on where your own company stands, that is what the AI Company Scan checks.
The smart budget is not "ads or organic". It is using ads to test which products and messages convert inside ChatGPT, then investing in being the organic answer for the questions that matter most. And it fits into the broader shift I described in agentic commerce, where the customer's own AI agent increasingly does the shopping and the question becomes who gets paid along the way.
What this means for you
If you run an online store on Shopify, this is the easiest AI ad channel you will ever be offered, and that is exactly why you should be disciplined. Wait for the app to reach your country (international rollout from 23 September), start with your highest-margin products, set a small fixed test budget, and judge it against the break-even table rather than against clicks. If your average profit per order is under about $30, expect this channel to struggle at current click prices.
If you sell B2B through HubSpot, the integration makes testing cheap, but be clear-eyed about the audience. Your buyers are mostly on Business or Enterprise accounts at work, where there are no ads. You will reach them on personal free accounts, if at all. Test it for top-of-funnel leads, and measure cost per qualified lead, not cost per click.
If you are considering a Sponsored Agent, treat it like hiring a salesperson who works for every customer at once. Write down what it is allowed to promise, test it hard before launch, re-test it after every product or policy change, and keep a human route for anything it cannot answer. The format can lift conversion meaningfully. It can also create liabilities at the same scale.
If you sell marketing services, this is a new line of work. Setting up feeds, writing conversational ad copy, designing and testing sponsored agents, and measuring real ROI across a new channel is exactly the kind of implementation work businesses will pay for, and one of the practical paths in the nine ways people make money with AI.
What to watch
Sponsored Agents leaving the test. If they roll out beyond select US advertisers, the ad slot becomes a conversation slot, and the quality of your agent becomes a competitive advantage.
Real click and conversion data. The $25-60 CPM and $3-5 CPC ranges are early. As tens of thousands more advertisers arrive, auction prices tend to rise. Watch whether performance keeps up.
Whether the "answers stay independent" line holds. The whole trust model depends on it. Any sign that paid placement bleeds into organic recommendations would change the value of both games overnight.
The tier split. If OpenAI ever extends ads to Plus users or work accounts, the audience, and the B2B case, changes completely.
The honest take
What OpenAI is building is not just another ad network. It is an attempt to own the moment between a question and a purchase, the moment that search engines owned for twenty-five years. Sponsored Agents take that one step further, turning the ad into the start of a sales conversation, and the Shopify and HubSpot integrations make it available to the smallest merchants almost instantly.
For businesses, that is a genuine opportunity with an ordinary, unglamorous condition attached: it has to pay. At $3 to $5 a click, shown mainly to free users, it pays for some products and not for others, and your own margins decide which group you are in. And if you let a Sponsored Agent speak for you, you are responsible for every word it says, no matter how clearly it is labelled as sponsored.
So here is the question to answer before you spend anything: at your real profit per sale, what conversion rate does ChatGPT need to deliver, and have you asked your own agent the questions a doubtful customer would?
Sources: Search Engine Land: OpenAI is turning ChatGPT ads into conversations; Digiday: ChatGPT ads business hits $1 billion run rate; OpenAI: Our approach to advertising; OpenAI Help Center: Ads in ChatGPT.
