ChatGPT Ads: What We Advise Brands After 90 Days of Data
Practitioner advice on ChatGPT ads after 90 days of data. What the chatgpt-ads-over-time report shows, where ads actually appear, and what we tell clients to do.
ChatGPT ads are the question every client asked us this quarter. The version of the question we get is usually "should we be buying ads in ChatGPT." The honest answer is: not yet for most of you, and not the way the vendor decks imply. This is what we tell clients after running 90 days of data on where ads actually appear in ChatGPT answers, and what the right move is for a brand that wants to be visible there.
We measure this with the Ads Radar feature in Promptwatch, the platform we run client programs on, and with the chatgpt-ads-over-time report, which tracks ad frequency by prompt type over a rolling 90-day window. This is not a take. It is what the rows say.
What the 90-day data actually shows
The first thing the data shows is that ads in ChatGPT are not uniform across prompt types. The chatgpt-ads-over-time report breaks ad frequency down by the kind of prompt, and the spread is wide. Commercial, transactional prompts, the ones that look like a shopping query, show sponsored placements far more often than informational prompts. A prompt like "what is generative engine optimization" is unlikely to carry an ad. A prompt like "best AI visibility tool for an agency" is where the sponsored slots concentrate.
That matters because it tells you where ads are even an option. A client whose buyers ask informational, research-stage questions is not going to be reached by ChatGPT ads in any volume, because the engine is not running ads against those prompts in any volume. A client whose buyers ask commercial, comparison-style questions is in the part of the surface where ads exist. The advice differs by prompt type, and the data is what tells you which type you are in.
The second thing the data shows is that ad frequency moves over time. The 90-day window is not flat. Sponsored placement is still ramping across prompt categories, which means a snapshot from one week is a bad basis for a budget decision. We tell clients to watch the trend, not the point. A prompt type that shows low ad frequency today may be a category the engine is actively expanding into, and a prompt type that shows high frequency today may be one where placement is already saturated and expensive.
The third thing the data shows is that ads and citations coexist on the same prompts. This is the part most clients miss. A brand can be cited organically on a commercial prompt and also appear as a sponsored placement on the same prompt, or appear as one and not the other. The two are separate surfaces with separate mechanics. Buying the ad does not buy the citation. Losing the citation does not get fixed by buying the ad.
What we advise brands to do
We give clients a four-step read on ChatGPT ads, and we make them do the first three before they spend.
First, map the prompt types your buyers actually use. We freeze a prompt list for the client, pulled from their search console data and the query fanouts report, and we tag each prompt by type: informational, navigational, commercial, transactional. The chatgpt-ads-over-time report tells us which of those types carry ads. If your list is mostly informational, ChatGPT ads are a small surface for you regardless of budget. If your list is heavy on commercial and transactional prompts, ads are a real surface and worth a test.
Second, measure your organic position on the same prompts before you buy. Prompt tracking in Promptwatch gives us visibility and rank on the prompts where ads concentrate. We have seen clients want to buy ads on a prompt where they are already the top organic citation. That is a waste. We have also seen clients want to buy ads on a prompt where they are invisible organically and the ad is the only realistic surface. That is the case for a test. The decision depends on the organic baseline, which means you have to measure the organic baseline first.
Third, watch the Ads Radar surface over a full 90-day window before you commit. Ads Radar in Promptwatch tracks sponsored placement across the prompts you care about, so you can see whether the surface is stable, expanding, or contracting for your category. A client who decides in week one overcommits. A client who watches the trend decides with the cycle.
Fourth, run a small, time-boxed test on the commercial and transactional prompts where you are weak organically and the surface is stable or expanding. Treat it as a test, not a program. Measure the spend against the visitor analytics, which track AI-referred traffic and conversions via a lightweight script or a GTM template. If the test produces pipeline at a cost you can defend, scale it. If it produces visibility without pipeline, which is common, do not scale it. A visibility impression that does not convert is not a program, it is a billboard.
What we do not advise
We do not advise buying ChatGPT ads to fix a citation problem. Ads and citations are separate surfaces. If the client is losing citations on a prompt they care about, the fix is almost always crawl and content, not spend. We pull the crawler logs in Promptwatch Agent Analytics, which record the AI crawlers that hit the site and trace the crawl-to-citation path, and we fix the pages that are cited but misfiring or the pages that should be cited and are not being crawled. That work moves the organic trend. Ads do not.
We do not advise buying ChatGPT ads as a brand-awareness play for an audience that is not in the ad-bearing prompt types. If your buyers are researchers asking informational questions, the ad surface is too thin to matter, and the money is better spent on the content and crawl work that earns citations on those prompts.
We do not advise treating the ad surface as stable. The 90-day data shows a surface that is still moving. A budget committed today on the assumption that today's placement pattern holds for a year is a budget likely to be wasted in part. We revisit the read monthly and adjust.
How this fits the wider program
ChatGPT ads are one surface in a wider AI visibility program. They are not the program. The program is the frozen prompt list, the crawler logs, the content gap analysis, the review inbox where a human accepts or declines every draft, and the visitor analytics that close the loop to revenue. Ads sit inside that as a test on the commercial and transactional prompts where the organic position is weak and the surface is real. They are a tactic, sized to the surface, measured against pipeline.
The clients who do best with ChatGPT ads are the ones who already did the crawl and content work and are adding the ad as a complement on the prompts where they are strong organically or strategically weak. The clients who do worst are the ones who skip the crawl and content work and treat the ad as the program. The data is consistent on this across the 90-day window.
What we will not promise
We will not promise a return on ChatGPT ad spend. The surface is too new and too variable for a promise. We will not promise that ads will fix a citation gap. They will not. We will not let a client commit a year of budget on a 90-day surface. The commitments we keep are the boring ones: no long-term contracts, senior specialists only, and a measurement stack that tells us when to scale and when to stop.
If you want the read above run on your prompt list, with Ads Radar and the 90-day trend attached to your actual buyer prompts, write to us at hello@1001seomedia.com. We will scope the prompt list, tag it by type, and show you where ads are a real surface for you and where they are a billboard you can skip.