ChatGPT Ads: 6 Months In, Advertisers Still Seek Best Practices

Six months into the rollout of ChatGPT Ads, the platform has evolved from a limited testing ground into a self-service advertising network available in 52 countries. While OpenAI has introduced new tools for buying, optimizing, and measuring campaigns, advertisers are still navigating a landscape with minimal context. The core challenge is not the lack of data, but the absence of benchmarks to interpret it.

Early results show a wide variance in performance. Some advertisers report costs per click (CPCs) under $3, while others pay upwards of $13. Campaigns have generated qualified leads for some, while yielding little meaningful activity for others. Without competitive or auction-level reporting, it remains difficult to understand why these disparities exist or what constitutes “good” performance on the platform.

Measuring Performance Within the Platform

ChatGPT Ads Manager provides standard metrics necessary for basic campaign evaluation. Advertisers can track impressions, clicks, spend, click-through rate (CTR), average CPC, and cost per mille (CPM) at the campaign, ad group, and ad levels. Data can be exported to CSV for external analysis.

For conversion tracking, the platform supports the OpenAI Pixel and Conversions API, covering actions such as purchases, leads, and sign-ups. Modeled conversions may also be included when available. Ecommerce advertisers can view attributed sales, cost per attributed sale, and return on ad spend (ROAS) if those metrics are captured.

However, the reporting ecosystem lacks depth compared to established search networks. There is no equivalent to Auction Insights, Impression Share, or detailed competitive reporting. Consequently, even straightforward metrics like CPC lack context, making it hard to discern whether price fluctuations are driven by competition, relevance, or inventory constraints.

The Black Box of the Auction

The opacity of the auction mechanism complicates optimization efforts. OpenAI utilizes a relevance-weighted, second-price auction where ad selection depends heavily on conversation context, intent, landing page signals, and creative quality.

Advertisers provide “context hints” at the ad-group level to describe relevant products or needs, but these are not keywords. OpenAI explicitly states that these hints do not guarantee delivery against specific words, audiences, or conversations. This structure limits visibility into how individual ads match with user queries before entering the auction.

If CPCs rise, the available reporting offers little insight into the cause. It is difficult to separate changes in competition from shifts in relevance or the types of conversations being matched. Furthermore, the second-price model means advertisers know their maximum bid and final payment but cannot see the competitive pressure or relevance score that determined the outcome. This makes interpreting individual CPCs nearly impossible without broader market data.

Varied Early Results Across Industries

As spend increases beyond initial test budgets, the diversity of outcomes becomes clearer. Publicly shared data reveals significant differences in efficiency and effectiveness across various sectors.

Hostinger conducted one of the largest public tests, spending nearly $70,000. Hüseyin Ograk, head of PPC at Hostinger, initially noted CPCs comparable to Google Search and strong engagement. However, as spend grew, he observed CPMs exceeding $65 and concerns over CTR. Specific use cases performed better than broad messaging, and traffic quality proved inconsistent, complicating ROI calculations based solely on direct conversions.

Other advertisers found different patterns. Common Thread Collective reported results for a high-average order value (AOV) ecommerce client that scaled from $7 to over $1,000 daily. After spending $9,620, the campaign achieved an average CPC of $4.41 and a CTR of 0.94%. Depending on the attribution model, the campaign generated between $19,000 and $38,000 in attributed revenue, with estimated ROAS ranging from 3.3x to 6.8x.

In contrast, B2B advertiser Floyd Blaikie’s team spent roughly $7,000 CAD, averaging a CPC of $9.29 and a CTR of 0.7%. Using visitor deanonymization tools, they identified 146 organizations behind 336 paid clicks. Only five of those organizations matched their ideal customer profile. Blaikie’s experience highlights a critical limitation: Ads Manager provides no way to verify if traffic aligns with target profiles, raising concerns about lead quality despite reasonable CPCs.

Geography also plays a major role in pricing. Synter’s test showed an overall average CPC of $9.89, but this varied drastically by region. Costs ranged from $5.10 in the U. K. to $10.62 in the U. S., reaching $17.59 in Australia and $22.89 in New Zealand. These variations underscore the difficulty of establishing universal cost expectations.

Misinterpreting the Bid Recommendation

A figure frequently cited in discussions about ChatGPT Ads is the $3 to $5 starting maximum CPC recommendation. This number originates from OpenAI’s guidance for setting initial bids, often accompanied by bid-strength indicators in Ads Manager.

It is crucial to distinguish this recommendation from a performance benchmark. OpenAI has explicitly stated that it does not yet have performance benchmarks across advertisers, industries, or campaign types. The $3 to $5 range is a starting point for bidding, not an average cost across the platform.

Relying on this figure as a benchmark risks misalignment with actual market conditions. With Maximize results now the default bid strategy for eligible new ad groups, advertisers who desire strict cost controls must opt back into manual bidding. Even then, OpenAI does not guarantee delivery against specific CPA, CPC, or ROAS targets. The recommendation serves only as a baseline for those still setting manual bids.

Audience Limitations and Geographic Reach

Evaluating the opportunity requires understanding the composition of the ad-eligible audience. ChatGPT Ads do not appear to users on Pro, Business, Enterprise, or Edu plans, nor to accounts identified as belonging to individuals under 18. Additionally, ads are shown to Free and Go users, meaning the ad-supported audience includes paying subscribers.

This exclusion of higher-priced tiers raises questions for luxury brands and high-ticket product advertisers. There is insufficient public data regarding the demographics or purchasing power of the ad-eligible audience. Broader statistics about total ChatGPT users are not representative, as they include ad-free tiers. While some high-AOV advertisers like Common Thread Collective have reported success, others may find the audience composition misaligned with their target profiles.

Geographic availability further restricts reach. The pilot launched in the U. S., Canada, Australia, and New Zealand in February. Expansion to the UK, Japan, South Korea, Brazil, and Mexico occurred in May and August. Self-service access reached 52 countries in late August, with ads buyable in over 40 markets through partners.

Strategic Approach for Advertisers

The lack of established benchmarks should not deter advertisers from testing ChatGPT Ads, particularly those seeking to diversify their channel mix. Success requires realistic expectations and robust measurement strategies.

Advertisers should define success criteria before launching, focusing on what the channel produces for their specific business rather than comparing metrics to others. Since CPC and CTR offer limited guidance due to unknown competition and audience dynamics, external measurement is essential. Tools for revenue attribution, engagement analysis, and company identification can provide insights that Ads Manager alone cannot.

Early results demand careful scrutiny. With fewer diagnostic signals, distinguishing between genuine performance issues and variations in delivery or matching is challenging. Advertisers may need more evidence to understand why performance shifts occur.

OpenAI has indicated that additional metrics, reporting views, and insights are planned as Ads Manager develops. For now, however, there is no universal metric to determine campaign health.

“Good” has to start with what the channel produces for your own business.