OpenAI Adds CPC Bidding to ChatGPT Ads, and There Are No Keywords to Bid On

OpenAI has expanded its ChatGPT advertising pilot beyond CPM-only buying, adding CPC bidding for campaigns with a Clicks objective plus a beta self-serve Ads Manager ^[1][2]. The operational consequence is larger than a new bid type: this is an auction with no keywords in it, so planning, measurement and exclusion policy have to be built rather than copied from search ^[3][4].

Ads appear below ChatGPT responses, and relevance is determined from signals in the active conversation combined with context hints the advertiser supplies ^[5][6]. Advertisers do not select a list of exact terms ^[6]. In search, an account is a lattice of queries, match types and negative keywords ^[7]; here the starting point is definitions of audience, category, use case and message that can be expressed through provided context ^[8]. According to the source material, the absence of keyword targeting is a fundamental design constraint rather than a missing feature ^[4]. That material also describes negative keywords only as a property of traditional search accounts and does not describe any conversation-level exclusion control in ChatGPT Ads ^[9]. Until one is documented, brand-safety governance in this channel is a matter of how narrowly you write your context, not what you block.

The money. OpenAI recommends starting with maximum bids of $3 to $5 per click, set at the ad-group level ^[10]. Third-party reporting has placed early observed CPCs in the same $3 to $5 band, and the same reporting indicates CPMs fell from roughly $60 at launch, with some figures cited around $25 ^[11][12]. That is a decline of about 58 percent from the launch level ^[1]. Because both CPC and CPM remain available ^[13], the choice is a bet on your own click-through rate: a $25 CPM matches a $3 CPC at a CTR of about 0.83 percent, and a $5 CPC at about 0.50 percent ^[2]. Below those rates, CPM is the cheaper way to buy the same impressions; above them, CPC is. None of these numbers is a rate card. They are pilot observations, and outcomes can move with campaign context, advertiser demand, available inventory, creative relevance and changes OpenAI makes to delivery ^[14][15].

Measurement now includes a Conversions API and pixel-based tracking, returning aggregated signals that can include conversions, click-through rate, CPC and CPM ^[16]. Availability is not maturity. Teams still have to settle conversion definitions, attribution windows and reporting consistency against their existing standards ^[17], and the source material does not establish parity with the reporting depth or workflow maturity of established ad ecosystems ^[18]. Treat the first flight as a controlled learning exercise: limited budget, one narrow business question, measurement plan agreed before launch ^[19].

Access is still gated, with OpenAI opening Ads Manager to more advertisers gradually as part of a controlled pilot ^[2][20]. Three things to watch. Whether the $3 to $5 recommendation survives contact with real demand, or whether it drifts the way the CPM did ^[10][12]. Whether OpenAI ships a conversation-level exclusion mechanism, which is the gating requirement for regulated and reputation-sensitive categories ^[9]. And whether Conversions API reporting holds steady long enough for anyone to compute a defensible cost per acquisition ^[16][17].