ChatGPT Ads Are Seven Months Old and Still Stuck at Test Budgets. Here's the Measurement Checklist Before You Spend.
Seven months after OpenAI started selling ads in ChatGPT, the story from agencies is not "it works" or "it doesn't." It's "we can't tell." Digiday reported on October 1, 2026 that measurement remains the main sticking point for marketers, and the examples are specific enough to be useful. (Digiday)
What agencies are actually seeing
One UK lead-gen campaign logged form fills while the platform showed zero conversions two weeks later. The same agency saw ChatGPT report about 100 clicks where its own tracking recorded 20, and OpenAI support couldn't explain the gap. Another partner sees conversions arrive 24 to 36 hours late, versus a few hours on Google. (Digiday)
Delivery is erratic too: one campaign served thousands of impressions one week and fewer than 100 the next, with "low bid" warnings that don't say what to bid. (Digiday)
Why budgets are stuck
One executive said clients spend $10 million a month on Google but under $100,000 a month on ChatGPT. Another agency's unanswered measurement questions keep clients at £10,000 to £20,000 test budgets rather than £100,000. And conversion tracking requires signing OpenAI's tracking terms, which many legal teams refuse; one agency said that applies to almost all its clients. (Digiday)
The demand side is not the problem. Kantar's Media Reactions 2026 found a net 75% of marketers plan to increase AI assistant investment in 2027, but only 32% of consumers use assistants to research brands and 23% trust their recommendations. (MarketingProfs) Intent to spend is ahead of both trust and tooling.
It is improving, with caveats
Advertisers can now optimize to cost per acquisition rather than only clicks or impressions, and OpenAI has signed mobile measurement partners AppsFlyer and Adjust. (Digiday) A visual ad format is set to test with Free and Go users in the US later in October, with Hightouch, LiveRamp and Tealium as partners. One advertiser, WeightWatchers, reported a 15.3% lower attributed acquisition cost than its paid-search benchmark; note that this is self-reported and attributed, not independently verified. (MarketingProfs)
Third-party verification can confirm an ad was seen, not whether it worked. Treat platform-reported conversions as a claim to reconcile, not a number to optimize against.
The pre-spend checklist
Get legal to review OpenAI's conversion-tracking terms before the test starts. If they won't sign, you are buying unmeasurable reach.
Tag every ad with UTMs and log landing-page sessions server-side. Reconcile against platform clicks weekly; a 5:1 gap is a red flag, not noise.
Set a fixed budget and a pre-agreed threshold (cost per qualified lead, not clicks) for scaling or stopping.
Expect conversion reporting to arrive a day or more late. Don't judge a flight on day-two numbers.
Run an equal-budget paid-search or LinkedIn test on the same offer so you have a real baseline.
The point for lean teams
If you have one marketer and a finite budget, ChatGPT ads are a learning experiment, not a channel. The cheaper move for the same buyer attention is earning the answer organically: the AI answer a buyer reads before any ad is shaped by your content, not your bid. Spend the test budget only if you can measure it, and keep the rest on being the citation.
Should a B2B company advertise in ChatGPT now?
Only as a capped test with your own click logging and a control channel. Platform reporting is not yet reliable enough to scale on.
Why do ChatGPT click counts differ from my analytics?
Agencies report platform clicks well above their own tracking, and OpenAI has not explained the gap publicly. Always reconcile with your own logs.
Can I optimize ChatGPT ads to conversions?
Advertisers can now optimize to cost per acquisition, but conversion tracking requires accepting OpenAI's tracking terms.
Related
51% of B2B Software Buyers Now Start Research in a Chatbot. Your Analytics Sees None of It.
G2's 2026 buyer survey: 51% of B2B software buyers now start research in an AI chatbot, 69% switched their vendor pick based on what it said, and none of it shows up in your analytics. Here's why the AI dark funnel is structurally different from the old one, and what to do about it.
Salesforce Named Its Outbound AI Agent "Hunter." Here's the Governance Checklist Lean Teams Need Before Trying the Same Thing.
Salesforce's September 11 Agentforce 360 launch gave the world seven named AI agents - including Hunter, an autonomous outbound sales agent still running in pilot. Even Salesforce isn't fully trusting it yet. Here's the governance checklist a lean team needs before it tries anything similar.
Every Data Vendor Just Shipped a "Context Layer" for AI Agents. Your GTM Stack Doesn't Need All of Them.
Egnyte launched a "context layer" on September 15. So did half the data industry this year - Databricks, ZoomInfo, Neo4j, a dozen more. Here's what's real infrastructure, what's rebranding, and the minimum viable context layer a lean GTM team actually needs.