The biggest shift in digital advertising since social media… and it’s barely getting started.
For a while, ChatGPT felt like the last clean corner of the internet. No ads, no sponsored results, no brand trying to wedge itself between you and an answer. Just a conversation. That officially ended in January 2026.
On January 16, 2026, OpenAI confirmed it was testing ads inside ChatGPT. By February 9, the first live placements were spotted in the wild. And right now, as of late March 2026, the beta is active (slow, limited, and expensive to get into), but it’s real.
Here’s a full breakdown of what we know so far: how the beta actually works, what the numbers look like, what users think about it, whether it’s worth running when it opens up, and where every other major LLM stands on the ads question.
What the Beta Actually Looks Like
The rollout is deliberately narrow. Ads only appear for U.S. users on ChatGPT’s Free tier and the new Go tier. If you’re on Plus, Pro, Team, Enterprise, or any educational plan, you won’t see any ads.
The format is straightforward: contextual text ads that appear at the bottom of ChatGPT’s responses, clearly labeled as “Sponsored” and visually separated from the actual answer.
So if you’re asking ChatGPT about meal planning, you might see an ad for a meal kit service after the response. OpenAI has been firm that ads do not influence what ChatGPT actually says. The answer comes first, entirely independent of whatever advertiser is in the mix.
Targeting is conversational rather than keyword-based. OpenAI uses the subject of the current chat, past conversation history, and previous ad interactions to determine relevance. Advertisers never see your actual conversations, they only get aggregated performance data like views and clicks. Users can dismiss ads, see why a particular ad appeared, and adjust or turn off personalization entirely.
One detail that caused controversy: if you have ChatGPT’s memory feature enabled alongside personalized ads, the platform may reference stored memories when selecting which ads to show. That felt like a step too far for a lot of users.
Who’s In the Beta & What It Costs to Play
Getting into the beta isn’t cheap. The minimum spend commitment to participate is $200,000. That’s unusually high for what’s essentially an alpha test of a new ad format.
Early beta participants were committing up to $1 million each. Three of the world’s largest holding companies are in the mix: WPP, Omnicom, and Dentsu.
OpenAI is working with Criteo as its first ad tech partner, connecting roughly 17,000 advertisers to the platform since early March. There are also reported conversations with The Trade Desk about a broader distribution setup.
A self-serve Ads Manager dashboard is being tested with a small group of partners, but in its current state, it’s basic: advertisers receive weekly CSV reports with click and impression data, not the real-time dashboards they’re used to on Google or Meta.
The CPM is $60 (that’s roughly three times Meta’s average rates and well above Google Display Network pricing). That’s either a signal of OpenAI’s confidence in the quality of its audience, or a way to test how much the market will actually bear before adjusting. Probably some of both.
The Numbers: CTR, CPA & What We Actually Know
Here’s the honest truth about performance data: it’s still very thin. OpenAI’s reporting infrastructure is early-stage at best. Advertisers are getting weekly CSV dumps instead of real-time attribution. So take any numbers floating around right now with a healthy dose of skepticism.
CTR
The estimated click-through rate from early testing is around 1.3%. That sounds low, but context matters. Google Search has a 29.2% CTR built on 25 years of intent-matching precision. ChatGPT’s ads are two months old. Some industry projections suggest well-optimized campaigns could reach 3-8% CTR as the platform matures, though those are forward-looking estimates, not confirmed beta results.
Conversion Rate
This is where the data gets more interesting. Criteo has reported that LLM-referred users convert at roughly 1.5x the rate of users from other referral channels. Some early data points suggest AI-referred traffic may convert at 5x Google rates. Take that with a grain of salt, the measurement is so immature right now that those figures are hard to validate.
CPA
Directional projections suggest CPAs could come in 20-40% below other digital channels once the platform matures. To put it in practical terms: if you’re paying $30 CPM and getting a 2% CTR, your effective CPC is $1.50. If 10% of those clicks convert, your CPA is $15; competitive against mid-funnel Google Search in a lot of categories.
Standard click-through rate doesn’t fully capture what’s happening in a conversational context. A user could see your ad, not click it, keep chatting, and then visit your site 20 minutes later having been meaningfully influenced. That kind of impact doesn’t show up in last-click attribution. Brands that evaluate this channel purely on CTR will likely undervalue it.
One more thing worth watching: frequency. If the same user sees your brand’s ad in 15 consecutive ChatGPT conversations, they’re not going to associate you with intelligence and helpfulness, they’re going to associate you with the thing that made their AI feel less trustworthy. Frequency capping matters here more than on any other channel.
|
CTR (current) |
~1.3% |
Early beta reporting |
Medium |
Early stage (2 months live) |
|
CTR (projected) |
3-8% |
Industry estimates |
Low |
Forward-looking (not confirmed) |
|
Conversion Rate |
1.5-5x other channels |
Criteo beta data |
Medium |
Early stage (measurement immature) |
|
CPA |
20-40% below other channels |
Directional projection |
Low |
Unconfirmed (awaiting scale data) |
|
Attribution |
Weekly CSV dumps |
OpenAI reporting infrastructure |
Low |
Early stage (no real-time data) |
Does It Actually Make Sense to Run ChatGPT Ads?
The honest answer: it depends heavily on what you’re selling and how big your budget is. This isn’t a channel for everyone right now.
Where It Makes Sense
ChatGPT ads work best for categories where customers naturally ask questions during the research phase rather than searching for a specific product by name. Think complex B2B services, SaaS, financial products, education, healthcare, high-consideration consumer purchases like cars or home services.
The intent signal here is fundamentally different from anything in paid media right now. When someone searches Google, they type “best project management software.” When someone asks ChatGPT, they say: “I run a 12-person agency and we’re drowning in project handoffs between client services and design; what tool would actually help?”
That’s not a keyword. That’s a paragraph of context. The advertiser who shows up in that moment, with the right message, has a very different kind of opportunity.
Where It Doesn’t (Yet)
Commodity products with minimal differentiation, B2B brands targeting senior decision-makers (who skew heavily toward ad-free paid tiers), and for anyone who needs tight attribution and real-time optimization… none of those are a good fit right now. The measurement infrastructure isn’t there yet.
There’s also a demographic reality to keep in mind. About 58% of adults under 30 use ChatGPT, and nearly half of all messages come from users under 26. Free and Go tier users skew toward students and early-career professionals. If your target customer is a 45-year-old CFO, they’re probably on a paid tier that doesn’t show ads.
The Practical Recommendation
A sensible starting point: allocate 70-80% of any experimental AI ad budget to Microsoft Advertising, which has mature reporting and proven ROI models, and reserve 20-30% for ChatGPT pilots to build early expertise. As measurement capabilities improve and the platform opens to self-serve, that split can shift based on actual results.
And don’t overlook the organic layer. If your brand isn’t organically surfaced in ChatGPT’s responses, your paid ad is going to feel like an intrusion. The AEO and GEO work, optimizing your content to show up in AI answers, is the foundation. The paid layer amplifies it.
What Users Actually Think (Spoiler: Not Great)
The reception has been rough. On Reddit, 68% of commenters expressed negative sentiment toward the announcement. The announcement post on X racked up over 10 million views, dominated by skepticism.
The core complaint isn’t really about the ads themselves, it’s about what they represent. ChatGPT felt like one of the last places on the internet actually trying to be helpful, with no commercial agenda baked in. The ads feel like a betrayal of that.
There was also a telling preview moment before the official launch: in December 2025, ChatGPT briefly started showing random app suggestions mid-conversation. A $200/month Pro subscriber posted a screenshot of ChatGPT suggesting the Peloton app during a completely unrelated discussion. The backlash was immediate and fierce.
That incident foreshadowed exactly how sensitive users are going to be about any hint of commercial influence bleeding into responses.
Broader consumer research adds another layer of concern.
In early 2026, 36% of U.S. adults said they’re less likely to buy from a brand that uses AI in ads, up from 32% just months earlier. About two in five have a negative opinion of brands that use AI in advertising, with nearly a quarter feeling very negative. Gen Z, which makes up a huge portion of ChatGPT’s free-tier user base, is particularly skeptical.
Early anecdotal signals suggest some users are actively upgrading to Plus specifically to escape ads. That’s a useful data point for OpenAI, ads may be driving subscription revenue as a side effect, but it also means the ad-eligible audience could self-select toward less engaged or less affluent users over time.
Where Every Other LLM Stands on Ads
ChatGPT is the only major LLM actually running ads in the conversational interface right now. Here’s where everyone else is:
Google Gemini
No ads in the chat interface yet, though they’re not ruling out the possibility. The way I see it, Google can afford to wait. The company already generates an estimated $254 billion in search ad revenue annually and has plenty of AI-adjacent ad inventory through AI Overviews in traditional search.
Perplexity
The most interesting case study. Perplexity actually experimented with ads and then pulled them, publicly stating that ads were “misaligned with what users want.” They may revisit it but for now they’ve made the deliberate choice to prioritize answer integrity.
That’s a meaningful differentiator while it lasts.
Grok
No ads yet, though Elon Musk has expressed an interest in changing that. Grok is tightly integrated with X and positioned around real-time social data and conversation.
For now, xAI’s monetization strategy leans on X Premium subscriptions and enterprise API access rather than advertising within the chat interface.
Claude (Anthropic)
Claude continues to be vehemently anti-ad across all tiers (and, in classic Anthropic fashion, actively using that as a selling point right now).
That said, Anthropic is burning through significant capital on infrastructure, and the CEO’s careful wording about potentially “revisiting” the approach suggests this isn’t an unconditional permanent commitment.
The Bigger Picture
The financial pressure driving all of this is worth understanding. OpenAI has 900 million weekly active users, but only about 5.5% (roughly 50 million people) pay for a subscription. The company generated $13.1 billion in 2025, but spent over $22 billion (they are also projecting losses of $14 billion for 2026 already, and it’s only April).
They have also committed to $1.4 trillion in data-center infrastructure over eight years. Advertising isn’t a choice they’re making from a position of comfort. It’s a structural necessity.
Truist analysts have called 2026 an “inflection year” for LLM ads, projecting OpenAI will generate under $1 billion in ad revenue this year, growing to over $30 billion by 2030. They also predict that within a few years, LLM-powered ad channels will become one of the most important pillars of digital advertising alongside Search, Social, and Retail Media. Those are big numbers, but they’re projections, not guarantees, and they’re contingent on OpenAI successfully walking the line between monetization and maintaining the trust that makes the platform valuable in the first place.
The advertising landscape is also about to get a lot more complex. Microsoft has a deep stake in how this plays out given its OpenAI partnership. Google faces real pressure if conversational ad intent proves as powerful as early signals suggest. And every publisher is watching to see whether sponsored answers deepen the zero-click trend that’s already eating into referral traffic.
What to Do Right Now
If you’re running paid media with a $1M+ media budget and you’re in a considered-purchase or high-intent vertical, you should be requesting early access now. The minimum commitment is still $200K, but the platform is going to open to self-serve at lower budget thresholds as the year progresses.
If you’re not there yet on budget, focus on the organic side. Audit your bottom-of-funnel content and make sure it’s written the way users actually talk to AI: specific problem framing, direct answers, dense information over marketing language.
The brands that show up organically in ChatGPT responses today are the ones that will get the most value from paid when it opens up widely.
And set up your measurement infrastructure now. When you eventually do launch on ChatGPT Ads, you’ll want UTM parameters, a dedicated GA4 segment, and a clean CRM source attribution setup ready to go. The last thing you want is to be building your reporting system while simultaneously trying to evaluate whether the channel is working.