ChatGPT Ads: An Honest Guide for Performance Marketers

What ChatGPT Ads actually are

ChatGPT Ads are labelled ad units that appear below a ChatGPT response when the system decides the conversation is a relevant match. Each unit carries an advertiser name, a favicon, a headline, a description, an image, and a landing page link. Structurally it looks closer to a search ad than a display banner.

What sits behind the placement is different from anything you have bought before. According to OpenAI's own documentation, the selection system weighs the context and intent of the live conversation, your landing page, your title and copy, the context hints you supply, and, where a user has ads personalisation enabled, signals from their broader ChatGPT history. There is no keyword list. There is no placement report in the sense you are used to.

Who sees your ads, and who never will

This is the part most launch coverage buries, and it is the part that should shape your decision.

• arrow_forwardFree and Go plans only. Users on Plus, Pro, Business, Enterprise, and Edu accounts do not see ads at all.
• arrow_forwardAdults only. Accounts identified as belonging to users under 18 are excluded, using account age data and age prediction.
• arrow_forwardNine markets. The United States, Canada, Australia, and New Zealand at launch, with the United Kingdom, Mexico, Brazil, Japan, and South Korea added on 11 August 2026.

Read that first bullet again. You are buying the free tier. For consumer categories, local services, and considered purchases with a broad audience, that is a perfectly reasonable population to reach. For B2B, it is a structural problem: the senior buyers you want are disproportionately sitting on company-paid Business and Enterprise seats, which are exactly the accounts that never see an ad unit.

Practitioners running the beta have been blunt about this. The recurring complaint in paid media communities is not that the ads perform badly in absolute terms, but that the audience available to buy is systematically the one with the least budget authority. If your ideal customer profile skews senior or enterprise, treat that as a reason to keep the test small rather than a reason to skip it entirely.

What ChatGPT Ads cost

Two buying models are live. You pick an objective, and the objective determines how you pay.

The auction is relevance-weighted and second-price. Your creative quality and landing page relevance affect whether you win, not just your bid. If that sounds familiar, it should. The mechanics borrow heavily from search, which means the discipline that works there carries over: tighter message match beats a higher bid more often than people expect.

Reporting in Ads Manager Beta currently gives you impressions, clicks, spend, click-through rate, average CPC, average CPM, and conversions. That is a thin column set by the standards of mature bidding platforms, and it is one reason the next section matters more than the pricing does.

Context hints are not keywords

At the ad group level you supply context hints: descriptions of the conversations, topics, or situations where your product is relevant. OpenAI is explicit that these are not exact-match keywords and do not guarantee delivery in any specific conversation. They guide matching. They do not control it.

If you have managed Performance Max, the feeling will be familiar. You describe intent and the system decides. The difference is that a conversation carries far more context than a query does, so the hints that work best are the ones written as situations rather than product nouns.

• checkWrite the moment, not the product. "Comparing accounting software after outgrowing spreadsheets" beats "accounting software".
• checkInclude the disqualifiers. Describe who the product is for so the system has something to narrow against.
• checkSplit hints across ad groups. One hint theme per ad group is the only way to read which situation actually delivers.

Because you cannot see the conversations that triggered delivery, ad group structure is your only reporting dimension. Build it as though it were the search terms report you are not going to get.

The measurement gap nobody puts in the launch deck

Here is the thing that will decide whether your test is worth running. In the beta, the click counts reported in Ads Manager routinely run well ahead of the sessions that show up in analytics. Advertisers comparing the two have reported gaps of an order of magnitude on small budgets: a dashboard showing over a hundred clicks against single-digit sessions from the matching geography.

Some of that is normal. Every platform counts clicks more generously than an analytics tool counts sessions, because of bounce-before-load, prefetch behaviour, and different definitions of a valid interaction. But a gap that size is not a rounding difference, and it makes every derived metric on the platform side untrustworthy. If clicks are overstated, your CPC is understated and your conversion rate is deflated. Every budget decision built on those numbers inherits the error.

This is not a reason to avoid the channel. It is a reason to instrument before you spend.

Four things to set up before you launch

• arrow_forwardStatic UTM parameters on every landing page URL. OpenAI confirms these persist through an ad click, which makes your own analytics the arbiter. If you do not have a convention already, fix that first with a consistent UTM naming scheme.
• arrow_forwardDynamic URL macros. Ads Manager populates campaign, ad group, ad, and ad account IDs at delivery time. Those IDs are the join key between the platform's rows and yours.
• arrow_forwardA reconciliation ratio. Platform clicks divided by analytics sessions, tracked weekly. Above two, stop quoting platform CPC in any report. Above five, the channel is unmeasurable at your current spend and the test should be paused.
• arrow_forwardA deliberate pixel decision. Automatic advanced matching is now the default for new web pixels and was switched on for existing ones. It hashes supported form data in the browser before sending it. If that needs a privacy review at your company, make the call knowingly rather than discovering it in an audit.

The broader principle applies well beyond this one platform. Every ad surface reports its own contribution generously, and the only defence is a single view where spend across every channel is reconciled against one source of truth you control. Adding an experimental channel is precisely when that discipline stops being optional.

How to run a ChatGPT Ads test that gives you an answer

A test is only useful if it can fail cleanly. Most early-channel tests cannot, because nobody agreed in advance what a failure would look like. Seven steps, in order.

Where this channel fits right now

For most mid-market advertisers in August 2026, ChatGPT Ads is a test line, not a core channel. The audience is capped at the free tier, the reporting is immature, and conversion volume on modest budgets is thin. Retailers with a clean product feed are the exception, and they should be moving now.

The reason to test anyway is not volume. It is cheap learning while the auction is uncrowded and the operational knowledge is still scarce. The same window existed on Meta in 2013 and on TikTok in 2020, and in both cases the advertisers who instrumented properly during the messy phase were the ones who could scale confidently when the inventory got expensive.

There is a wider shift underneath this too. As more of the discovery journey happens inside assistants rather than on results pages, the click you used to buy is being quietly redistributed. Understanding how paid placement behaves on a conversational surface is going to matter regardless of whether this particular platform is the one that wins.

Frequently Asked Questions