Track Ads in Your Prompts with Promptwatch

If you came here from a search about tracking ads in your ChatGPT prompts, the short version is this: load the prompts you care about into Promptwatch, run them on a plan that includes Ads Radar, and read the sponsored units the platform captures against each prompt. The longer version is the setup, because a sloppy prompt set is the thing that makes ad tracking useless, and most of the work is before you open the ads view at all.

One fact to hold onto before the steps. There is no official ChatGPT ads API and no vendor has an OpenAI partnership for this. Ads Radar gets its data by monitoring the real ChatGPT interface and storing what it sees. Any tool that claims an official integration is misrepresenting how it works. Promptwatch is upfront that the source is UI monitoring, which is the only honest source.

Pick the plan before you pick the prompts

Ads Radar is a platform feature, not a standalone SKU. The commerce reports it sits beside are on Professional at $245/mo and above, and on the self serve agency plans (Kick-off $199, Growth $399, Scale $799). Essential at $95/mo covers mentions and citations but not the commerce views, so a buyer who stops at Essential will not see the ads report at all. Explore is free with 10 ChatGPT prompts and is enough to confirm the surface exists in your category before you pay.

The plan choice changes what you can do with the data, not whether the data exists. A roster with many brands wants an agency plan because it carries unlimited prompts and 10 seats, and ad tracking across clients only works if the prompts from each client sit in their own project. A single brand can run Professional and get the commerce views plus the rest of the visibility stack.

Build the prompt set the way a buyer types

The prompt set is the part that decides whether ad tracking pays off. Write 15 to 25 prompts the way a person would type them into ChatGPT, not the way your category pages are titled. A category page is titled "AI visibility platform." A buyer types "what is the best tool to see if my brand shows up in ChatGPT." Both are valid prompts, and only the second one is the prompt where a sponsored unit is likely to appear, because the second one is commercial intent.

Mix three kinds. Branded prompts, where someone types your name, are the ones to watch closest because a rival buying those is buying your own demand. Category prompts, where someone types a generic need, are where you find rivals you did not know about. Comparison prompts, where someone names you next to a competitor, are where the ad and the organic citation can both appear and you need to tell them apart.

Tag each prompt by intent when you load it. Promptwatch stores intent as BRANDED, INFORMATIONAL, NAVIGATIONAL, COMMERCIAL, or TRANSACTIONAL, and prompt type as ORGANIC, BRAND_SPECIFIC, or COMPETITOR_COMPARISON. The tags are not decoration. They are what let you later filter the ad list so a COMMERCIAL row on a BRAND_SPECIFIC prompt is not blended with an INFORMATIONAL row on an ORGANIC one.

Confirm the surface by hand first

Before you trust a dashboard, run five of the prompts in ChatGPT yourself. Note which answers show a sponsored unit and where the unit sits in the answer. This does two things. It calibrates how often the surface appears in your category, and it gives you a ground truth row to compare the dashboard against. If your category almost never returns a sponsored unit, ad tracking is a metric about a surface that barely exists for you, and the prompt set is what needs to change first, not the plan.

What you get back from Ads Radar

Once the prompts run, Ads Radar stores each captured ad with the creative, the advertiser name and root domain, the landing page, the source response it came from, and the position of the ad inside that answer. It keeps the prompt string, so the row is tied to the query that produced it. A library without the prompt is just a folder of ads. A library with the prompt tells you which of your queries a rival is buying into, which is the question a brand team can act on.

The report is queryable, not a flat export. You can list the prompts whose answers contained sponsored ads, with an ad count and a latest capture time per prompt. You can list advertiser root domains ordered by ad count. You can pull the top advertiser domains with daily counts for a 90 day share of ads trend. Filters cover model, prompt type, and intent, so the rows stay comparable.

The two sorts that matter

Sort by position in the response to see who buys the top slot, because the top sponsored unit and the third one are not the same buy. Then filter by intent. A rival buying your branded prompt is a ticket. A rival buying a broad category prompt you do not target is context. The two need different responses, and a single "did a rival appear" view hides that.

What this is not

This is not bid management. ChatGPT sponsored units are not Google Ads, the auction is different, and importing a Google campaign mental model breaks here. Ads Radar is a monitoring tool for the ChatGPT surface, not a management tool for a Google one. Treat the output as observation with capture gaps, not as a complete feed, because no feed exists. Site: promptwatch.com.

What to do this week

1. Pick a plan that includes Ads Radar, Professional or an agency plan.
2. Write 20 buyer prompts, branded and unbranded, tagged by intent.
3. Run five by hand in ChatGPT and confirm the surface exists in your category.
4. Load the set into Promptwatch and let the captures build.
5. Open the advertiser domain list and find the rival on your branded prompts.