The data you can and cannot get from ChatGPT Ads

Ads that appear in the ChatGPT chat window look similar to search ads. Yet the data that lands in an advertiser's hands is a world apart from a Google Ads report. Will you simply burn this money as ad spend, or put it to work as research spend that guides your content strategy? That decision only gets made once you've drawn a line between what you get and what you never get.

On ChatGPT Ads, standard metrics like impressions, clicks, CTR, CPM, and CPC, plus conversions captured via pixel or the Conversions API, flow in from the moment you switch the ad on. Once impressions accumulate to around 100, anonymized 'Conversation Insights' open up as an addition. Inside them you find what topics users ask about, what objections they hold, and in what tone they ask. Conversely, you cannot get the raw conversation text, a 1:1 link between a single click and a single conversation, or how many times that question actually came up within a category (absolute frequency). This boundary line was drawn directly by OpenAI in its Ads Manager beta guide and its advertising terms.

Standard metrics that arrive the moment you switch it on

Six numbers arrive from the moment you switch the ad on, in exactly the same shape as on any other channel. Impressions, clicks, CTR (click-through rate), spend, average CPC, and average CPM. Anyone who has ever run an ad platform has almost nothing new to learn here.

Why these numbers are useful is simple. They count, exactly as they are, how many times a user saw a sponsored answer or ad card within a conversation and how many times they actually clicked. The Ads API quickstart in OpenAI's developer documentation lets you choose CPC (recommended price of $3–5) or CPM bidding at the ad group level.

For the first two days of a campaign, look only at CTR and average CPC. When CTR sits at the bottom of the category, you'll want to overhaul the ad copy first, but that's the wrong order. ChatGPT attaches ads by conversational intent instead of keywords. So the one- or two-sentence natural-language signal you put into an ad group (context_hint in the documentation) is essentially the targeting itself. You need to fix this signal before the copy.

For example, if you want your brand's answer to appear when a user asks about "a serum for winter skin with flaky patches," you put the language of that situation and concern into the signal. If this is off, then no matter how good the copy is, the ad gets attached to the wrong conversation.

Conversions only show up once you install the pixel

The four metrics (conversions, CVR, CPA, and conversion value) are only captured after the advertiser embeds a pixel or the Conversions API on their own site, and even then only on a last-click basis. Within the scope OpenAI has disclosed, you can define your own events on top of ten kinds of standard events. The known pixel function takes the form oaiq().

Attribution is last-click only, which you need to account for from the start. If a user who came over from a ChatGPT conversation completes a purchase a few days later via search or Instagram, that conversion does not appear in the ChatGPT Ads dashboard. Mistaking this number for the channel's true performance will throw off your ad-spend decisions.

That's why, in practice, teams also feed server events from GA4 or their own CRM through the Conversions API. The point is to separately verify the revenue that the ChatGPT channel actually added (the incremental effect) within the marketing mix. It's better to set up both measurements from the start. Then there's no later dispute over "did this channel actually produce results."

Conversation Insights: no raw text, only patterns

The data that makes ChatGPT Ads decisively different from any other ad platform is 'Conversation Insights.' This data, which only opens up once impressions have accumulated past 100, is not the raw conversation text. It comes in as depersonalized topic clusters, the objections users hold, the tone they frequently use, and short paraphrased sentences that represent them.

Three metrics are reported to become available alongside it.

• Conversation depth: the number of follow-up questions a user throws out after seeing the ad.
• Contextual relevance: how well the ad fit into that conversational flow.
• Follow-up response: whether the user asked additional questions about the brand.

OpenAI's help center ad user management documentation explains that this data is opened up at the level of advertiser permissions.

Read the three numbers layered together and they become not a report card but a diagnosis. If, within the same topic cluster, CTR is high but CVR is low, that's a signal that the promise the ad made and the trust of the landing page have drifted apart. If follow-up response is lively but clicks are few, the user is at the stage of digging up brand information within the conversation rather than clicking out through the ad. The question sentences that emerge then get transcribed as they are and moved into your FAQs and content headings.

The data you can't get, and why you can't

There are five kinds of data you can't get. Four of them are permanently blocked by privacy policy, and the last one is different in nature.

The fifth item causes the most confusion. OpenAI isn't hiding absolute frequency. It's a number that simply can't be extracted in the first place, because of how the auction works. The per-cluster impression count an advertiser sees is roughly determined by 'auction win rate × contextual relevance × bid price.' Even if a cluster's impressions read 100, that doesn't mean people asked about that topic 100 times.

Even for the same question, if relevance or bid price is low, it loses the auction and impressions register as 0. ChatGPT Ads data alone cannot tell you how often that question is asked across the market. Only when you cross-check with external tools like Otterly, Semrush, and Adthena does the size of market demand finally come into view.

How an agency turns this data into content

PION treats ChatGPT Ads not as a media channel but as a content research channel. Ad spend is the price of buying insight into which intent attaches to which topic, and the signals that come out of it flow directly into content work for organic search and AI visibility. PION runs this cycle as an agency that handles ChatGPT Ads operations on behalf of brands with a U.S. entity.

Here is how we put the data to use.

• User question phrasing and tone move directly into FAQ questions and H2/H3 headings.
• Objection phrasing is used as the backbone of comparison and rebuttal content.
• Topics that generate longer conversations are published first as long-form guides.
• High-CVR clusters are put into organic search content work as the top priority.
• For clusters with high CTR but low CVR, we reinforce the landing page's trust materials (brand materials and case studies) before the ad copy.
• If the ads keep appearing in irrelevant topic clusters, we narrow the targeting signal further.

That said, we use it with the limits clearly drawn. Because absolute frequency has to be measured separately outside OpenAI, ChatGPT Ads shows not 'questions you have not reached yet' but 'the intent behind questions your ads currently reach.' The full range of questions across the market must be measured separately with external tools to avoid gaps.

Ad placements and answer citations operate separately, so agencies may handle them as separate services. How to divide the two and what to check in a proposal is laid out in What a ChatGPT Marketing Agency Actually Does.

Frequently asked questions

What data can you get from ChatGPT Ads?

You get six standard metrics (impressions, clicks, CTR, spend, average CPC, and average CPM) from the moment you switch the ad on, and if you install a pixel or the Conversions API, you get conversions, CVR, CPA, and conversion value on a last-click basis. Once impressions exceed 100, Conversation Insights, made up of anonymized topic, objection, and tone expressions, open up as an addition, along with three metrics: conversation depth, contextual relevance, and follow-up response.

What data can an advertiser not get from ChatGPT Ads?

You cannot get five kinds of data: raw conversation text, 1:1 mapping between clicks and conversations, breakdown at the individual level such as gender, age, and region, behavioral profiles, and absolute frequency of questions within a category. The first four are permanently blocked under OpenAI's privacy policy, and absolute question frequency cannot be inferred from those impression counts because per-cluster impressions are determined by 'auction win rate × contextual relevance × bid price.'

What information does ChatGPT Conversation Insights give you?

It gives you anonymized topics, objection expressions, user tone, and short paraphrased sentences that represent them, at the cluster level. The raw conversation text is not included, and proper nouns or specific personal information may disappear in the anonymization process. The conversation depth, contextual relevance, and follow-up response metrics provided alongside it show numerically how deeply a user interacts with the brand after seeing the ad.

How does ChatGPT Ads attribution work?

It's a single last-click model connected to a pixel or the Conversions API. If a click that came over from a ChatGPT conversation converts a few days later via another channel, it is not captured in the ChatGPT Ads dashboard. PION separately verifies channel contribution with its own marketing mix model and incrementality tests, handling the gap between the last-click number and actual performance as a separate matter from the start.

How do you use Conversation Insights data in content marketing?

We use it as input for converting ad spend into content research cost. User question phrasing is adopted directly into FAQs and headings, objection phrasing is moved over as the backbone of comparison and rebuttal content, and high-CVR clusters are put into organic search content as the top priority. That said, because ChatGPT Ads data does not show the question volume of the entire market, absolute frequency has to be measured separately with external tools.