AttributionAttribution / journal ATTRIBUTION.COM.BR
ChatGPT Ads·Aug 21, 2026·6 min

You do not start advertising on ChatGPT from zero

Almost nobody has real numbers on advertising on ChatGPT yet. The channel is weeks old for most accounts, the documentation is thin, and the case studies going around are mostly screenshots with the interesting parts cropped out. So take everything below for what it is: first readings, not laws.

What we can share is what happened when our system started advertising on ChatGPT without starting from zero. It already operated the Google Ads side of the same accounts. It knew which searches converted and which ad text actually pulled. The open question was whether any of that knowledge would survive the trip to a channel where people write prompts instead of queries.

It survives. More than we expected.

Do Google Ads keywords work on ChatGPT Ads?

Not directly. There is no keyword field to paste them into. GPT Ads targets contexts, roughly the situation and intent a conversation is in, and a context is a different animal from a keyword. The person is mid-conversation, half decided, asking a machine to think with them. But a keyword never converted because of its string. It converted because of the intent underneath, and intent travels.

So the system took each account's top-performing Google Ads keywords and translated the intent behind them into ChatGPT Ads contexts. Those contexts converted 2.1× more than the account's own baseline in the channel. Same budgets, same weeks. The only difference was that one set of contexts was guessed fresh and the other was carried over from years of search evidence.

Context groups the system built on its own

The second reading surprised us more. The system does more than copy winners. It hunts for patterns in converting searches: the worry that keeps repeating, the shape of the question people ask right before they buy. On several accounts, the system that already runs the search side turned those patterns into new GPT Ads context groups nobody had requested, and ROAS rose 1.8× once they went live.

One pattern showed up almost everywhere: price. People ask ChatGPT about price differently than they search for it. Less product plus cheapest, more is this worth it for someone in my situation. Prompts about price behaved like their own cluster, so the system gave them their own context group, built overnight while everyone involved was asleep. We found it in the morning, already spending.

A caveat we owe you: volumes are still small. A young channel means young data, and a 2.1× measured on early traffic can soften as spend scales. We will report the follow-up either way.

See what your search data already knows about ChatGPT Ads

Start with Google

Ad text crossed the channel line too

Winning ad copy is the other asset an account quietly owns. Phrasings that kept beating their rivals in Google Ads carry real information about what customers respond to. Those patterns crossed into GPT Ads on their own. Nobody exported a spreadsheet, nobody briefed a writer. The system had already learned which claims won, and it wrote the ChatGPT side with that knowledge, adjusted to how a sponsored answer reads inside a conversation rather than on a results page.

A new channel also invites junk. So every GPT placement sits under a sponsorship guard that watches performance and retires the weak ones. No sentiment, no sunk cost. Early channels reward whoever prunes fastest, because the platform itself has not learned yet what bad looks like.

Everything above happened without a confirmation click. The contexts were created, the copy crossed, the weak placements retired, and nobody was asked to review a draft first. On a channel this young that matters twice over: nobody has the experience to review these decisions well anyway. The account's own data does.

What advertising on ChatGPT rewards right now

Memory. That is the honest summary of our first weeks. Clever prompt theories have mattered less than we assumed they would. What has mattered is showing up with an account that already understands its own customers, and moving that understanding into the new channel before the crowd arrives. Years of Google Ads evidence turned out to be the best onboarding a new channel can get.

We will keep writing as the numbers mature. Some of these early multiples will hold, some will soften. The core reading feels durable: you do not start a new channel from zero when your system already understands why people buy from you.

New channelsContextsROAS
Read next