ChatGPT ads are live. The ROI isn't proven yet.

Let the data tell the story, because right now, the story around ChatGPT ads is being told mostly by people with a financial stake in your believing it.

Every client conversation we're having this quarter includes some version of the same question: should we be putting budget into ChatGPT ads? The honest answer, once you set aside the first-mover hype, is more nuanced than the agencies selling ChatGPT ad management want it to be. We pulled the actual 2026 performance data, compared it against the four channels that already have a proven track record and found a real conflict between what OpenAI is reporting and what advertisers are experiencing on the ground.

Here's what the numbers actually show, and where that leaves your budget.

Quick take:

• ChatGPT ads show real early traction, but independent reporting also documents underdelivery and unprovable ROI for a meaningful share of advertisers
• Google and LinkedIn still hold the deepest, most auditable performance data of any channel
• Treat ChatGPT ads as a small, capped test for now, not a reallocation of proven budget

Where ChatGPT ads stand right now

OpenAI began testing ads inside ChatGPT in February 2026, for logged-in Free and Go tier users in the United States, then expanded to Canada, Australia and New Zealand in the weeks that followed. OpenAI's own announcement calls the early results "encouraging," reporting no measurable impact on trust metrics and low ad dismissal rates.

The growth curve backs that up on the surface. Independent monitoring firm Cloro.dev tracked ad penetration jumping from just 0.42% of responses in April and May to 26.5% overall and 49.1% within the US, in three weeks by late May. That's roughly a 60-times increase. A self-serve Ads Manager followed in May, removing the original minimum spend requirement and opening the platform to smaller advertisers.

But growth in ad volume is not proof that the ads work. That's where the story splits.

What the case studies reveal

Some early results are genuinely good. Digital agency Jellyfish reported that premium CPMs (cost per thousand impressions) quoted at around $60 actually landed closer to $30, alongside early conversions and high order values, though the agency was careful to frame this as a brand-awareness result, not a direct-response one. Separately, one advertiser tracking conversions independently found efficiency close to Google's non-brand search, an outcome nobody had modelled for on a CPM-only product, according to Digiday's reporting.

Now the other side. AI search intelligence platform Adthena found one client's ads landing a click-through rate of just 0.91%, nearly seven times below the 6.4% benchmark for the same sector on Google search, while a separate enterprise advertiser had used only 3% of a $250,000 commitment after several weeks. Search Engine Land went further, reporting that two agency executives told The Information they could not prove the ads drove measurable business results for clients at all, with deals still negotiated by phone and email rather than through automated buying tools.

Both of these things are true at the same time. That's not a contradiction you resolve by picking a side. It's a signal that the channel is still assembling the infrastructure that makes ROI provable at all.

How does it compare to the rest

Here's where the comparison gets useful. The four established platforms have years of market data behind them. ChatGPT ads don't, yet.

PlatformWhat the market showsWhat the data showsGoogle AdsSearch still captures 39.8% of all US digital ad revenue, the largest share of any categoryMature, proven attribution down to closed-won revenueLinkedIn AdsB2B ad revenue reached $4.59 billion in 2025, up 9% year over yearHighest lead quality for B2B, priciest per clickMeta AdsInstagram's average CPM alone reached $9.46 in Q2 2025Needs your own intent layer (pixel, lookalikes) to work wellX AdsDeclining priority among B2B marketers amid brand-safety and moderation concerns, per ForresterBest for awareness, weak for direct lead generationChatGPT Ads0.91% CTR reported in one sector trial, versus 6.4% for Google searchAttribution tools (pixel, Conversions API) only shipped in May 2026

The gap that matters most for a real-ROI conversation isn't any single number; it's the depth of evidence behind it. Search still commands 39.8% of all US digital ad revenue, according to IAB and PwC data, and LinkedIn's B2B ad business alone reached $4.59 billion in 2025. Those figures are audited, aggregated across an entire market and tracked year over year.

The ChatGPT figures in the table above come from single-client trials. That's not a small distinction. It means any ChatGPT ad number you see right now, ours included, should be read as a data point, not a benchmark.

The real conflict behind the headlines

Here's what the case studies don't tell you outright: OpenAI has an enormous financial reason to want this channel to look good, fast. Internal investor presentations reported by Axios project that OpenAI's ad revenue will climb from $2.5 billion in 2026 to $100 billion by 2030, roughly half of Meta's current ad business.

That same Axios reporting notes something worth sitting with: Anthropic used a Super Bowl commercial to publicly declare its own Claude platform will remain ad-free, a direct competitive jab that only makes sense if ad-free positioning is seen as a genuine differentiator right now, not a relic.

There's a second, quieter tension worth naming. WPP, Omnicom and Dentsu, three of the world's largest ad agencies, were named in OpenAI's official pilot testing program.

Sources tied to those same three agencies are also the ones telling CNBC, anonymously, that they're frustrated with slow delivery and thin insights. Official partner in public, anonymous critic in private. That gap tells you more about where this channel actually stands than either the glowing case studies or the skeptical exposés do on their own.

What Facebook Ads taught us

While everyone talks about ChatGPT ads like they're an unprecedented new frontier, we've actually watched this exact movie before. Facebook Ads launched in November 2007 with its own hyped debut, and its own early trust misstep: the Beacon program, which broadcast users' off-platform purchases to their friends without real consent. Facebook shut Beacon down in September 2009 as part of a class-action lawsuit settlement.

It took years, not months, for Facebook to become a channel advertisers could actually measure and trust. Worldwide ad revenue sat at $267 million when Sheryl Sandberg joined as COO in 2008, grew to nearly $1.9 billion just two years later, and reached nearly $114 billion by the time she left her full-time COO role in 2022. The advertisers who built real advantage weren't the ones who panicked into the platform in year one. They were the ones who tested small, built measurement infrastructure early and scaled only once the data justified it.

That's the pattern worth remembering here. A rocky, hard-to-measure first year isn't a red flag unique to OpenAI. It's what every major ad platform launch has looked like.

Our recommendation for real ROI

If your goal is provable return, not brand exposure, the data points to one clear approach: keep Google Ads and LinkedIn Ads as your foundation, since both have the attribution maturity to trace spend to closed revenue, and treat ChatGPT ads as a small, capped experiment rather than a reallocation.

Before you spend a dollar:

• Install the pixel (a tracking snippet for your site) and the Conversions API (reports offline sales back to OpenAI) that it shipped in May
• Track everything through your CRM to actual customers, not just clicks
• Cap initial spend at roughly 5% to 10% of your current Google Ads budget, enough to gather a real signal without real risk
• Compare results against your own historical cost-per-customer on Google and LinkedIn, not the market-scale figures in the table above, which measure a different thing
• Set a hard 30- to 60-day review window before committing further spend

If the numbers hold up against your own baseline, scale slowly. If they don't, you've lost a small test budget, not a quarter's pipeline.

The people with a financial stake in this channel will keep telling you it's ready. Right now, the data says something more balanced: real promise and real growing pains, at the same time. Ready to build a measurement plan before you spend anything? Get in touch, and we'll map out a test that actually tells you something.