ChatGPT Ads has changed quite a bit since OpenAI began testing ads in February.
The platform now gives advertisers more ways to buy, optimize, and measure campaigns, with self-service access available across 52 countries to date.
We also have something we didn’t have at launch: real results from advertisers who have spent money on the platform.
Some are reporting costs per click (CPCs) below $3. Others are closer to $10, or even $13. There are examples of campaigns generating qualified leads at competitive costs, along with others that produced little meaningful activity after the click.
The problem is knowing what to make of those differences.
OpenAI says it does not yet have performance benchmarks across advertisers, industries or campaign types. There is also limited competitive or auction-level reporting to help explain why one advertiser might pay substantially more than another.
Six months in, advertisers have more ChatGPT Ads data to work with. But, they still have very little context for deciding what “good” performance should look like.
What Advertisers Can Actually Measure In ChatGPT Ads
ChatGPT Ads Manager gives advertisers most of the basic performance metrics needed to evaluate their own campaigns.
Reporting is available at the campaign, ad group, and ad levels, including:
- Impressions.
- Clicks.
- Spend.
- Click-through rate.
- Average CPC.
- Average cost per mille (CPM).
- Conversions.
Advertisers can also export performance data to CSV for analysis outside the platform.
Conversion tracking is available through the OpenAI Pixel and Conversions API, with support for actions including purchases, leads, and sign-ups. OpenAI may also include modeled conversions when available.
For ecommerce advertisers, reporting can include attributed sales, cost per attributed sale and ROAS when those metrics are available.
That’s about as far as the current reporting goes.
There isn’t an Auction Insights equivalent, Impression Share or competitive reporting. OpenAI also hasn’t published performance benchmarks across advertisers, industries or campaign types.
Those gaps make it difficult to put even a straightforward metric like CPC into context, especially when the auction itself gives advertisers relatively little visibility into what influenced the price they paid.
See also: How To Measure PPC Performance When AI Controls The Auction
What Advertisers Can’t See Behind The Auction
The lack of competitive reporting matters more when you consider how ChatGPT Ads decides which ads to show.
OpenAI uses a relevance-weighted, second-price auction. Ad selection is based primarily on the context and intent of the conversation, along with signals from the landing page, ad creative, and advertiser-provided context hints.
Those context hints give advertisers a way to describe relevant products, needs or situations at the ad-group level, but they aren’t keywords. OpenAI also makes clear that they don’t guarantee delivery against specific words, audiences or conversations.
That gives advertisers less visibility into the individual matching decisions happening before an ad ever enters or wins an auction.
If CPC rises, for example, the available reporting provides little help separating changes in competition from relevance, inventory, bidding or the types of conversations being matched.
The second-price auction adds another layer of complexity. Advertisers know their maximum bid and what they ultimately paid, but they can’t see the competitive pressure behind that price or how relevance affected the outcome.
That makes individual CPCs difficult to interpret outside the context of the advertiser’s own campaign.
And the results advertisers have published so far show just how wide those differences can be.
What Early Advertiser Results Show So Far
We have more first-hand ChatGPT Ads data than we did a few months ago. Some advertisers have also spent enough that we’re getting beyond the $500 test campaigns that were common early on.
The results still vary quite a bit, as seen below from a few advertisers who have shared their data.
Hostinger Spent Nearly $70,000 Testing ChatGPT Ads
Hostinger has one of the larger publicly shared tests I’ve found so far.
Hüseyin Ograk, head of PPC at Hostinger, initially reported that CPCs weren’t higher than Google Search and that the campaigns had already generated purchases. He also described strong engagement from the traffic.
His assessment became more nuanced as spend increased.
After nearly $70,000 in testing, Ograk reported CPMs above $65 but said CTR was a bigger concern. More specific use cases performed better, while broader messaging consistently struggled.
He also noted inconsistent traffic quality and challenges evaluating return on ad spend based only on direct conversions.
Similar Tests Are Producing Very Different Results
Other published tests show how difficult it is to draw conclusions from CPC alone.
Common Thread Collective shared results from a high average order value (AOV) ecommerce advertiser that scaled from $7 per day to more than $1,000 per day over roughly a month. After $9,620 in spend, the campaign averaged a $4.41 CPC, and 0.94% CTR across 136,000 weekly impressions.
CTC reported between $19,000 and $38,000 in attributed revenue, depending on the attribution model, with estimated ROAS ranging from 3.3x to 6.8x. The team relied on Triple Whale alongside ChatGPT Ads reporting to evaluate downstream performance.
Floyd Blaikie found something different when her team looked beyond the metrics available in Ads Manager.
Her B2B campaign spent roughly $7,000 CAD and averaged a $9.29 CPC, $64.34 CPM, and 0.7% CTR. Her team then used visitor deanonymization to identify 146 organizations behind 336 paid clicks.
The crazy part? Only five matched its ideal customer profile.
Blaikie’s external analysis raises a bigger concern than the $9.29 CPC itself. Very few of the identifiable visitors matched the advertiser’s ICP, and Ads Manager offered no way to uncover that on its own.
Synter’s test shows another source of variation: geography.
Across $4,428.84 in spend, Synter averaged a $9.89 CPC overall. But that number changed considerably by market, from $5.10 in the U.K. to $10.62 in the U.S. CPC reached $17.59 in Australia and $22.89 in New Zealand, although both markets had considerably less volume.
Across these tests, CPCs vary widely by advertiser, and market. The available reporting still doesn’t provide enough information to explain those differences or establish a reliable expectation for what advertisers should pay.
OpenAI’s own $3 to $5 starting maximum CPC recommendation doesn’t necessarily solve that problem, either.
The $3 To $5 Recommendation Isn’t A CPC Benchmark
One number has already started showing up frequently in discussions about ChatGPT Ads: $3 to $5 per click.
The figure comes directly from OpenAI, but what it represents is important. OpenAI recommends a starting maximum CPC bid between $3 and $5. Ads Manager may also provide bid-strength guidance indicating whether a bid is competitive or could limit delivery.
That recommendation doesn’t represent an average CPC across the platform. OpenAI explicitly states that ChatGPT Ads doesn’t yet have performance benchmarks across advertisers, industries or campaign types.
The early advertiser results we’ve already looked at reinforce how little the $3 to $5 recommendation tells us about what an advertiser will ultimately pay.
The risk is that repeated references to the recommendation eventually strip away the word “recommended.” A $3 to $5 starting bid becomes a $3 to $5 “benchmark,” even though OpenAI has published no such benchmark.
The risk is that repeated references to the recommendation eventually strip away the word “recommended.” A $3 to $5 starting bid becomes a $3 to $5 “benchmark,” even though OpenAI has published no such benchmark.
That framing is already narrowing. In August, OpenAI made Maximize results the default bid strategy for eligible new ad groups, which sets and adjusts bids automatically. OpenAI’s documentation states it does not guarantee delivery against a specific CPA, CPC or ROAS target. Advertisers who want a hard ceiling have to opt back into manual bidding.
So the number is a starting bid, for the advertisers still setting one.
The audience eligible to see ChatGPT Ads is not the same as ChatGPT’s total user base, though the gap is about composition more than size. Ads reach Free and Go users, who make up the bulk of ChatGPT’s audience, but every higher-priced tier is excluded.
See also: The High CPC Paradox: When Expensive Clicks Are A Sign Of Success
ChatGPT Ads Only Reach Part Of The ChatGPT Audience
Advertisers also need to be careful when using ChatGPT’s overall audience size to evaluate the advertising opportunity.
Ads can appear to users on Free and Go plans only. Plus, Pro, Business, Enterprise, and Edu accounts remain ad-free, and OpenAI doesn’t show ads to accounts identified as belonging to people under 18.
That means the total number of people using ChatGPT isn’t the same as the audience currently eligible to see ads.
Go is itself a paid subscription, so it’d be inaccurate to characterize the ad-supported audience as people who don’t pay for ChatGPT. Still, several of OpenAI’s higher-priced subscription tiers are excluded from the advertising audience.
That raises an interesting consideration for luxury brands, high-ticket products, and other advertisers with a narrower customer profile.
There isn’t enough public data about the demographics or purchasing power of ChatGPT’s ad-eligible audience to know how closely it aligns with those customers. Broader statistics about ChatGPT users aren’t necessarily representative either, since they include people on plans where ads don’t appear.
That doesn’t mean high-ticket advertisers should stay away. Common Thread Collective, for example, has already reported positive results from a high-AOV ecommerce campaign.
But, it may give those advertisers another reason to be cautious about applying broader ChatGPT audience assumptions to their media plans.
Geography narrows the comparison further. ChatGPT pilot launched in the U.S. in February alongside Canada, Australia and New Zealand. OpenAI announced expansion to the UK, Japan, South Korea, Brazil and Mexico in May, with Brazil and Mexico going live in August. Self-service Ads Manager access reached 52 countries on August 31, and OpenAI says ads are buyable in over 40 markets through its sales team and partners.
Between plan eligibility, geography, and limited audience data, advertisers are still working with a fairly incomplete picture of who they can reach.
That makes the advertiser’s own test design and success criteria even more important.
How To Approach ChatGPT Ads Right Now
The lack of established benchmarks shouldn’t keep advertisers from testing ChatGPT Ads, especially those looking to expand beyond their existing channel mix.
Just go into that test knowing there are still limits to what the platform can tell you.
Set expectations before spending begins based on what would make the channel worth continuing for your business. Someone else’s CPC or CTR offers little guidance when we still know so little about the competition, audience, and conversations behind those results.
Measurement outside Ads Manager should also be part of the plan from the beginning. Some of the early advertiser tests found their most useful insights elsewhere, including revenue attribution, engagement data, and identifying which companies actually visited the site.
Early results also deserve more scrutiny than they might on a mature platform. With fewer diagnostic signals available, it can be difficult to separate a true performance problem from differences in delivery, matching or the conversations available during the test period.
Advertisers may need more evidence before deciding why performance moved in one direction or another.
ChatGPT Ads will continue changing, and the data available to advertisers should improve with it. OpenAI has already said additional metrics, reporting views, and insights are planned as Ads Manager develops.
For now, there is no universal number that determines whether a ChatGPT Ads campaign is performing well.
“Good” has to start with what the channel produces for your own business.
More Resources:
- When Advertising Shifts To Prompts, What Should Advertisers Do?
- The Gap Between AI’s Heartwarming Ads And Its Real Goal: Your Job
- Surviving The Impression Squeeze: How Agentic Commerce Is Changing Google Ads In 2026
Featured Image: Roman Samborskyi/Shutterstock
