We Tested ChatGPT Ads for Our Clients. Here Is the Data We Got.

When OpenAI opened ChatGPT advertising to brands in 2026, we decided to test it properly rather than guess. We ran campaigns across multiple client accounts, tracked every metric from impressions to actual conversions, and here is the honest account of what happened — the good numbers and the disappointing ones.

The pitch for ChatGPT advertising is compelling on paper. Hundreds of millions of people use ChatGPT every month, many of them asking research-intent questions that put them right at the start of a buying journey. Getting your brand in front of someone the moment they are asking about a problem you solve sounds like a performance marketer’s dream.

So when the ad product became available, we did not wait to see what others found. We allocated test budgets across several client campaigns, set up tracking as carefully as the platform allowed, and ran it for long enough to draw conclusions. Here is exactly what we found.

Why We Decided to Test ChatGPT Ads

Every new ad channel gets a wave of hype before the real performance data emerges. We have been through this with TikTok Ads, with Performance Max, with Demand Gen. Our approach is always the same: set a defined test budget, build the cleanest possible tracking we can, and let the data decide — not the product announcements.

ChatGPT specifically felt worth testing early because the user intent signal is different from any other channel. When someone is in a ChatGPT conversation asking about a specific problem, they are often further along in their awareness than a cold social media audience, but approaching the question differently than a Google searcher who is already in buy mode. That middle ground felt like it could work well for certain types of clients — particularly those in considered-purchase categories where the research phase is long.

We also wanted to understand the platform before our clients started asking about it. Better to have real answers from our own testing than to give advice based on press releases.

How ChatGPT Ads Actually Work

Before we get into results, it helps to understand what the product actually is, because it is quite different from Google or Meta in ways that affect how you interpret the performance data.

ChatGPT ads are currently served only to Free and Go tier users — paying Plus and Pro subscribers do not see ads. This is a meaningful constraint on reach, and it also says something about the mindset of users seeing your ad: they are the cost-sensitive segment of the ChatGPT audience, which has implications for conversion rate on higher-ticket products.

The targeting mechanism is what OpenAI calls context hints — you select topics or query types that trigger your ad when a user’s conversation matches that context. There are no keywords in the Google Ads sense, no demographic targeting filters, no behavioural audience segments. You are matching to conversation context, not to user profile. This is a fundamentally different targeting logic and it produces fundamentally different traffic quality than most performance marketers are used to.

The pricing model is CPM — cost per thousand impressions — not CPC. You pay to appear, not to get clicked. This shifts the way you should think about campaign economics: a low CTR is less damaging than it would be on a CPC model, but it also means your cost per click is a function of both CPM and CTR, and if your CTR is low, your effective CPC climbs fast.

One thing that immediately stood out to us: there is no negative keyword functionality. On Google Search, we routinely filter out hundreds of irrelevant query types with negative keywords to tighten traffic quality. On ChatGPT, once you set your context hints, you cannot exclude the conversations that loosely match but represent completely different intent. That lack of control over traffic quality is something you feel in the conversion data.

How We Set Up the Campaigns

We ran campaigns across three distinct client categories: a B2B SaaS tool in the productivity space, a direct-to-consumer brand in the home goods category, and a professional services firm offering financial consultancy. Each client had a defined test budget, UTM tracking set up for click attribution, and conversion tracking via their existing pixel or CRM integration.

For each campaign we created multiple ad variations to test copy approaches. We learned quickly that the creative review process on ChatGPT has stricter requirements than we expected — more on that below. The campaigns ran for between two and four weeks per client, which we considered the minimum needed to draw any directional conclusions while being honest about the data volume limitations.

We set context hints as specifically as we could for each client — for the SaaS client, targeting productivity and workflow automation conversations; for the home goods client, targeting home improvement and interior design conversations; for the financial services client, targeting personal finance and investment planning conversations. In each case we went as narrow as the platform would allow rather than broad-matching to a wide topic area.

The Numbers: What We Got

We will be direct about these results. Some of the numbers are more promising than we expected. Others were sobering. We are sharing both because a cherry-picked case study is not useful to anyone trying to make a real budget decision.

Overall Campaign Metrics

Across our test campaigns, total impressions ran into the tens of thousands per week per active campaign, which confirmed that volume is not the bottleneck — the platform delivers impressions at scale quickly once campaigns are live. CPMs ranged from approximately $28 to $38 depending on the client category and the specificity of the context hints. More specific, higher-competition context hints pushed CPM toward the top of that range.

Blended CTR across campaigns came in at 0.3% to 0.5%. For comparison, display advertising typically delivers 0.05% to 0.1% CTR, so ChatGPT ads are performing noticeably better than standard display on click engagement. But search advertising CTRs are typically 3% to 8% for well-matched queries, so if you are coming to ChatGPT from a search advertising background and expecting search-like CTR, you will be disappointed.

Our effective CPC, calculated from CPM and CTR, ranged from about $6 at the low end to $18 at the high end. The variance was driven almost entirely by CTR differences across context hint categories — the same CPM produces very different effective CPCs depending on how well the ad creative matched the conversation context.

Where the Impressions and Clicks Actually Came From

Geography made a larger difference than we anticipated. UK placements consistently delivered better CTR and lower effective CPC than other English-language markets. Our best-performing geo delivered roughly double the CTR of our worst-performing geo within the same campaign, same creative, same context hints. This is consistent with what other advertisers running on the platform have reported — the UK ChatGPT user base appears to be more engaged with ads at this stage, possibly because the product is earlier in its rollout there and novelty is still a factor, or because the user composition is different.

We also noticed that the time of day matters, though the platform’s dayparting controls are limited. Conversation-driven ad impressions are more concentrated in certain windows, and campaigns that burned budget quickly in the early hours of the day often exhausted spend before the higher-engagement windows later in the day. Budget pacing was less controllable than we would like, and this is a real operational friction for accounts where budget management is tight.

The Conversion Story — This Is Where It Gets Honest

Clicks were the easier win. Conversions were where the platform revealed its current limitations most clearly.

For the home goods D2C client, we tracked click-through to product page visits, add-to-cart events, and purchases. Traffic from ChatGPT ads had a bounce rate significantly higher than traffic from Google Shopping or Meta prospecting — in the range of 70% to 80% compared to 45% to 55% from those channels. Average session duration was short. Users were landing on the page, looking briefly, and leaving. The purchase conversion rate from ChatGPT clicks was a fraction of what we see from Google Shopping clicks on equivalent budgets.

For the SaaS client, we tracked free trial signups. CTR was reasonable. Click-to-trial conversion rate was poor — in the low single digit percentages where our Google Search campaigns on equivalent spend were delivering 8% to 12%. The traffic was arriving, but it was not converting at the rate you need for SaaS economics to work at a reasonable CPA.

For the financial services client, we tracked form submissions and consultation bookings. Here the results were better in relative terms — not because the conversion rate was high, but because the conversations the ads appeared in were more specifically relevant to the service, and the users arriving had a clearer sense of what they were clicking through to. Cost per lead was still significantly higher than our Google Search cost per lead for the same client, but not so dramatically as to rule out a top-of-funnel role.

The pattern across all three was consistent: ChatGPT users click with curiosity, not with purchase intent. They are in a research mindset, not a buying mindset. This is the fundamental tension with using ChatGPT advertising for direct response objectives. The intent signal is real — someone asking about a relevant topic in ChatGPT is genuinely interested in that topic — but they are earlier in their journey than a Google searcher, and they convert at lower rates as a result.

Ad Approvals and Rejections: What We Learned the Hard Way

We lost time to the creative review process, and we want to flag this clearly because it affects how long it takes to actually get campaigns live.

OpenAI’s ad policy has strict restrictions on claims about AI autonomy. Any copy that frames AI as doing things autonomously — “let AI handle your X”, “our AI manages everything for you”, “AI-powered automation takes care of it” — gets flagged and rejected. For our SaaS client in the productivity automation space, this meant our first set of ad creatives were rejected entirely. We had to rewrite them to lead with user outcomes rather than AI capability: “save three hours per week” rather than “our AI does it for you”.

The rewritten creatives were approved and actually performed better from a CTR perspective once live. Specificity around user outcomes is better creative discipline regardless of the platform policy — it turned out the constraint was a useful forcing function. But the rewrite cycle added days to our campaign launch, and if you are on a tight timeline, factor that in.

We also found that image-heavy creative performed differently from text-forward creative in ways we did not fully anticipate from our experience on other platforms. The context in which the ad appears — alongside an AI-generated text response — means users are primed for reading, and ads that led with a strong text hook before any image element had better engagement than ads that led with the visual.

What Actually Surprised Us

A few things we did not expect going in.

First, the organic citation comparison. While running paid ChatGPT ads for one of our clients, we noticed that the client’s existing content was being cited organically within ChatGPT responses — not as paid placements, but as recommended sources when the AI answered relevant questions. The organic citations drove more qualified traffic than the paid ads did. Users who arrived from an organic ChatGPT citation had longer session durations, lower bounce rates, and higher conversion rates than users who clicked through from a paid placement. This tells us something important: being cited by ChatGPT in its organic responses is more valuable than being shown as an ad alongside those responses. The credibility of a recommendation is higher than the credibility of an ad, which is not a new truth but it is a meaningful one for how you think about ChatGPT as a channel.

Second, budget pacing was erratic in ways we did not fully control. Daily budgets burned out faster than expected on some days with no obvious surge in query volume, and on other days, spend came in well under the daily cap. This unpredictability made it harder to project campaign performance over a multi-week flight and complicated reporting for clients used to the more predictable pacing of Google Ads campaigns.

Third, there is no ad library. On Google and Meta, we routinely look at competitor creative as part of our campaign planning. On ChatGPT, that visibility does not exist. You are testing blind relative to the competitive landscape, which increases the number of creative iterations you need before finding effective copy and formats. For clients who are used to informed creative strategy, this is a real workflow gap.

Our Verdict: Is ChatGPT Advertising Worth It Right Now?

Here is our honest assessment based on what we ran and what we saw.

If you want a direct response channel that delivers cost-efficient conversions against a target CPA today, ChatGPT ads are not there yet. The conversion rates we saw across all three client categories were too low relative to the CPM cost to make the economics work for pure performance objectives. Google Search remains the better choice if your primary goal is cost-per-acquisition efficiency, and our Google Ads management work consistently outperforms what we saw from ChatGPT ads on that metric.

If you want to test a new channel with a defined experimental budget — somewhere in the range of $500 to $2,000 — and your goal is to understand how your audience behaves in the ChatGPT environment, it is a worthwhile test. You will learn something about your creative and messaging that carries value beyond the channel itself, and you will be ahead of competitors who wait until the platform is more proven before engaging with it.

If you are in a considered-purchase B2B or professional services category where the sales cycle is long and the research phase is extensive, the top-of-funnel awareness argument is stronger. Users asking ChatGPT about financial planning, enterprise software selection, or professional services options are in exactly the kind of exploratory mindset where a well-placed, credible brand impression can influence a decision weeks or months later. Attributing that to a conversion is hard with current tooling, but the interaction is real.

The piece of advice we would give any client considering ChatGPT ads right now: before allocating paid budget, make sure your content is the kind of thing ChatGPT cites organically. If ChatGPT is recommending your competitors in its responses and not you, that is a content and GEO optimisation problem to solve first. The organic citation is worth more than the paid placement, and getting the organic foundation right will make your paid ads more effective when you do run them.

We will keep testing as the platform develops. OpenAI is clearly building this out, and ad products at this stage of development can change significantly over the first twelve to eighteen months. The data we have today is a starting point, not a verdict on what the channel will become. If you want to talk through whether it makes sense for your specific situation, our team is happy to have that conversation — reach us via the free audit form or the contact page.

Frequently Asked Questions