Field research · Kansas City BBQ prompts · ChatGPT Ads
ChatGPT has no ad library. Your AEO tool is the closest thing.
No transparency center. No searchable archive. But if you’re already capturing full responses on a schedule, you’ve been recording competitor ad data for months without reading it.
In case you missed it: ChatGPT runs ads now. Sponsored cards at the bottom of answers, matched to whatever you happen to be talking about, roughly the way Google runs ads against searches.
Roughly. Hang onto that word. The gap between “roughly like Google” and “actually like Google” is where everything interesting in this post lives.
Here’s what nobody mentions: there is no ad library. No transparency center, no searchable archive, no place to type a competitor’s name and scroll through what they’re running. Google gives you one. Meta gives you one. OpenAI gives you nothing.
So if you want to know who’s advertising against your category, which prompts trigger ads and which don’t, and what the targeting actually looks like, you have two options.
Open ChatGPT and start typing. Log what you see. Do it again tomorrow, because it changes. And the day after. Forever, basically — and you still won’t have a sample you can trust.
Let software fire the prompts every day and log the results while you sleep. If you already run AEO tracking, you have this now. You just haven’t read the ad layer yet.
The software screen-captures every result daily, so the archive builds itself. Pages with ads can be filtered by day and by exact prompt, which turns what used to be hours of manual competitive research into a filter click.
Why this is worth doing now
Every ad channel has an era where the people paying attention hold an unfair advantage over everyone waiting for the tooling to mature. Google had one. Facebook had one.
ChatGPT ads are in that era right now, and the barrier to entry is nothing. While most teams are still arguing about whether AEO is real, you could be showing up organically and holding the sponsored slot underneath your own answer.
Why an ad library probably isn’t coming
ChatGPT ads serve inside private, one-to-one conversations. That’s why OpenAI hasn’t built an archive, and probably won’t build the transparency tool you’re picturing.
Regulation might eventually force something thin. The EU’s Digital Services Act already requires large platforms to maintain ad repositories. But a compliance archive tells you who ran an ad and roughly when. Not which prompt triggered it, or why.
Your organic monitoring stack is already an ad probe
AEO and GEO platforms work by running a fixed set of prompts through the AI engines on a schedule and capturing the output. The intent is organic: brand mentions, domain citations, share of voice.
But the capture doesn’t discriminate. It grabs the whole response. And since February 9, 2026, a growing share of those responses have carried a sponsored card at the bottom.
Which means your organic monitoring has been running a controlled experiment against OpenAI’s ad system for months. Every scheduled capture is a probe. Every ad that lands in one is a data point about how the system behaves.
What you can actually infer
-
Who’s advertising
The obvious first pass, but don’t stop at names. Read the headline, the offer, the call to action. Compare advertisers against each other and you start to see which copy is built to win the click and which was pasted over from a search campaign.
-
What ChatGPT considers monetizable
Which prompts pull ads, and how often. The system is telling you which parts of your category it has classified as commercially viable — before you spend a dollar finding out. That sets realistic budget expectations and keeps campaigns from sitting dormant.
-
How well the targeting actually works
Advertiser names and display URLs against your prompts show you how good the match is. Relevance is the whole game in a channel with no keyword list to negotiate with.
-
Where the category boundary sits
Who shows up that shouldn’t. This is the highest-value inference in the set. Think of it as the search terms report in Google Ads, and how much liberty Google takes with a keyword list on phrase and broad match.
-
How the system rewrites the question
The fan-out query ChatGPT generates internally to go retrieve results — the machine’s own translation of the human input, exposed. Retrieval and ad matching happen against that rewrite, not the phrasing your customer typed.
-
How paid and organic interact in your category
The prompts where you have no citation and no ad, while a competitor has both. Your AEO tool has always shown you the first half. It has been capturing the second half all along.
Reading the ad layer
The filtered view below shows how often a given prompt pulls an ad along with the answer. Any AEO or GEO platform capturing complete responses on a schedule can produce this.
Two things stand out. No prompt runs ads 100% of the time. And the more commercial the intent, the higher the odds an ad gets attached.
Some prompts never pull ads at all. Others trend over time and are worth watching. When you’re monitoring a hundred prompts a day, the filter that narrows the list to only the ones running ads is what makes this a ten-minute job instead of an afternoon.
The early results
Remember that word — roughly? Here’s where the targeting looks rough.
Kansas City was a host city for the World Cup, so we’d been monitoring Kansas City BBQ prompts to see how ChatGPT ranks and mentions local restaurants. That started as an organic AEO task. The same prompts turned out to trigger a lot of ads.
The data doesn’t look great for advertisers. The audience and keyword targeting are worse than expected often enough that we’ve told clients to slow down and keep more budget in Google and Meta for now.
Four sampled ads, scored on match quality
A sample of four, not a rate. Pulled from Kansas City BBQ prompts we were already monitoring for organic visibility.
PromptBest KC barbecue restaurant for catering large events?
Survey and form software, served against a restaurant catering question. Both the prompt and the ad headline contain the word “catering,” so there’s something to correlate on. It’s hard to imagine Jotform is thrilled about where that budget went.
PromptKansas City BBQ recommendations
A “BBQ tonight?” headline from a print shop. This one is a stretch in any direction you read it, and it may well be the advertiser’s doing rather than the matching. Hard to tell from the outside — which is the whole problem with a channel that has no ad library.
PromptBest BBQ rubs and sauces
The strongest placement in the sample. The prompt wants rubs and sauces. They sell rubs and sauces. If every match looked like this, the channel would be an easy yes.
PromptWhere to find burnt ends in Kansas City
Category-relevant and still wrong. This is an online retailer selling grills and equipment, answering someone who wants to know where to eat tonight. The topic matches. The intent doesn’t.
Four ads is a sample, not a study. But it shows why tooling and research matter before you invest in the new thing rather than after.
In fairness to OpenAI: it’s early, and relevance should improve as more advertisers enter the auction and the platform matures.
What we’re seeing beyond relevance is aggressive bidding from large national brands. They aren’t afraid to pay several times more per click on ChatGPT than they do on Google. The CPM model doesn’t hold up either — we’ve seen CPMs at roughly double what you’d expect to pay for a CTV programmatic video ad.
Which is the real argument for reading the ad layer before you buy into it. The cost of finding this out by running campaigns is considerably higher than the cost of reading data you’re already collecting.
Frequently asked questions
Can you actually see ChatGPT ads without an ad library?
Not directly, but you can capture them. ChatGPT serves ads inside private conversations, so there’s no public surface to index. The workable substitute is sampling: send realistic prompts on a schedule, capture complete responses, and read the ones that came back with a sponsored card.
How does AEO tracking software help with ChatGPT ads?
AEO and GEO platforms already run a fixed prompt set through the engines on a repeating schedule and capture the full response. Sponsored placements come along for the ride. Tools that expose the ad as a filterable element, rather than burying it in the response body, make the audit practical instead of tedious.
Is there an official ChatGPT ad library?
No. OpenAI publishes no searchable archive of ChatGPT ads. Given that ads serve inside private conversations, an equivalent to the Meta Ad Library or Google’s Ads Transparency Center is unlikely in the form marketers expect.
Can I see competitor ad spend on ChatGPT?
No. You can observe how often an advertiser appears against a given prompt in your sample. Running your own ads on the same prompts will tell you what it takes to win the placement.
What is a query fan-out, and why does it matter for ads?
It’s the retrieval query ChatGPT generates from a user’s prompt — the system’s internal rewrite of the human question. It matters because retrieval and ad matching happen against that rewrite, not the original phrasing. It’s the closest observable analog to a keyword this channel has produced.
Why is a company that isn’t my competitor advertising on my prompts?
Because ad matching is contextual and currently operates at the category level, not the intent level. A national retailer targeting a broad topic lands on local service intent inside that topic. Check for adjacent-category drift specifically — it’s often the softest inventory in your set.
Who can run ChatGPT ads right now?
Since the May 5, 2026 beta launch of OpenAI’s self-serve Ads Manager, US advertisers of any size can register at ads.openai.com. No minimum spend, CPC bidding, conversions API. Managed and agency-partner buying still exists alongside it.
Do ChatGPT ads influence the organic answer?
OpenAI’s stated position is no — ads appear below the response in a labeled unit, separate from the answer. Worth noting that’s OpenAI’s claim about its own system, not an independently verified finding.
How often should I re-run this?
Your AEO tool is already running on a schedule. Read the ad layer every two weeks. The value isn’t any single capture. It’s watching who enters, who exits, and which prompts get more contested over time.
AI Search Optimization
See who’s advertising against your prompts
We run the prompts your customers actually type, capture the full response every day, and read both layers — where you get cited, and who bought the space underneath. Before you put budget into a channel with no ad library, it’s worth knowing what’s already there.
Book an AEO CallYour prompts, your category, your competitors. Not a generic sample.