ChatGPT Ads has grown up fast. Since OpenAI started testing ads in February, the platform has added more buying, optimization and measurement options, and self-service access now reaches 52 countries as of August 31.
What it hasn’t added: any way to tell whether your numbers are good.
As Search Engine Journal reports, OpenAI says it does not yet have performance benchmarks across advertisers, industries or campaign types. There’s no Auction Insights equivalent, no impression share, no competitive reporting. Advertisers are spending real money into a black box and grading their own homework.
The CPC spread is wild
Published tests show just how little a single CPC number means right now.
- Hostinger spent nearly $70,000 testing. Head of PPC H\u00fcseyin Ograk initially said CPCs weren’t higher than Google Search and that purchases were coming in. As spend grew, his read got more cautious: CPMs above $65, CTR as the bigger worry, narrow use cases outperforming broad messaging, and inconsistent traffic quality.
- Common Thread Collective scaled a high-AOV ecommerce advertiser from $7/day to over $1,000/day in roughly a month. After $9,620 in spend: $4.41 average CPC, 0.94% CTR, 136,000 weekly impressions, and $19,000\u2013$38,000 in attributed revenue depending on model, for an estimated 3.3x to 6.8x ROAS. They leaned on Triple Whale, not just Ads Manager.
- Floyd Blaikie’s B2B test spent about $7,000 CAD at a $9.29 CPC, $64.34 CPM and 0.7% CTR. Deanonymization identified 146 organizations behind 336 clicks \u2014 and only five matched the ideal customer profile. Ads Manager gave no hint of that.
- Synter spent $4,428.84 for a $9.89 blended CPC, but by market it ran $5.10 in the UK, $10.62 in the US, $17.59 in Australia and $22.89 in New Zealand (the latter two on low volume).
Same channel, radically different economics. And nothing in the reporting explains why.
Why you can’t diagnose the auction
ChatGPT Ads runs a relevance-weighted, second-price auction. Selection leans on the context and intent of the conversation, plus signals from your landing page, creative and advertiser-provided \”context hints\” set at the ad group level.
Those hints are not keywords, and OpenAI is explicit that they don’t guarantee delivery against specific words, audiences or conversations.
So when CPC climbs, you can’t tell whether it’s competition, relevance scoring, inventory, bidding or simply a different mix of conversations. You know your max bid and what you paid. That’s it.
The $3\u2013$5 number is not a benchmark
The figure everyone repeats \u2014 $3 to $5 per click \u2014 is OpenAI’s recommended starting maximum bid. Not a platform average. Not a benchmark. Ads Manager may also flag whether a bid is competitive enough to deliver.
The risk is obvious: repeat it enough and \”recommended starting bid\” quietly becomes \”industry benchmark,\” and people start optimizing toward a number that means nothing.
It matters even less now. In August, OpenAI made Maximize results the default bid strategy for eligible new ad groups, setting and adjusting bids automatically. OpenAI’s docs say it doesn’t guarantee delivery against a specific CPA, CPC or ROAS target. Want a hard ceiling? You have to opt back into manual bidding.
Your reach is smaller than ChatGPT’s headline numbers
Don’t build a media plan off ChatGPT’s total user count. Ads only show to Free and Go plans. Plus, Pro, Business, Enterprise and Edu are ad-free, and OpenAI doesn’t serve ads to accounts identified as under 18.
Note that Go is itself paid, so this isn’t simply \”non-payers\” \u2014 but every higher-priced tier is excluded. For luxury and high-ticket brands, that’s a real question mark, and there isn’t enough public data on the ad-eligible audience’s demographics or purchasing power to answer it. CTC’s high-AOV win shows it can work; it just means broad ChatGPT stats aren’t a proxy for your addressable market.
Geography narrows it further. The pilot launched in February in the US, Canada, Australia and New Zealand; UK, Japan, South Korea, Brazil and Mexico were announced in May, with Brazil and Mexico live in August. Ads are buyable in over 40 markets via sales teams and partners.
How to test it without fooling yourself
Test it \u2014 especially if you’re hunting for incremental channels. Just design the test properly.
Write down your success criteria before you spend. Someone else’s CPC tells you nothing about your auction, your audience or the conversations you’re being matched into.
Plan measurement outside Ads Manager from day one. The most useful findings in these early tests came from elsewhere: revenue attribution tools, engagement depth, and company-level visitor identification. Blaikie’s five-out-of-146 ICP match is the whole argument in one stat.
And hold early results loosely. With this few diagnostic signals, a \”bad\” week could be delivery shifts or conversation mix rather than a genuine performance problem. Gather more evidence before you kill or scale.
OpenAI has said more metrics, reporting views and insights are on the roadmap. Until then, \”good\” is whatever moves your P&L.
Source: Search Engine Journal
