10 September 2026 · 6 min read
Written from OpenAI's own documentation for ChatGPT Ads, read in a browser on 10 September 2026. Every claim about what the product does links to the page it came from. Avid has no measured read of how the format performs, so none is offered here.
Self-service ChatGPT Ads is now listed as available in both Australia and the United Kingdom. A business in either market can register, set a budget and a maximum bid, upload a headline and an image, and start appearing below ChatGPT's answers. What none of that money buys, on OpenAI's own account of its own product, is the answer sitting above the ad.
That is the design rather than an oversight, and OpenAI states it in the documentation and in its announcements. It also settles which line of the plan the money leaves.
The ad appears below the end of a response. It carries the advertiser's name, a favicon, a headline, a line of copy, an image and a landing page. The help centre says ads are clearly labelled as sponsored, visually separated from ChatGPT's response, and not an endorsement by OpenAI. Ads may appear for people on the Free and Go plans. Plus, Pro, Business, Enterprise and Edu accounts do not see them, nor do accounts OpenAI identifies as belonging to people under 18. During the current phase they also stay out of Temporary Chats, logged-out sessions, image generations and the Atlas browser. Go is a paid plan, so this is not only the free tier. It is the free tier plus the cheapest paid one, which is the audience every higher subscription removes from the inventory. Read the advertiser basics, and check the country list.
It prices like paid media, because it is paid media. Reach is bought on cost per thousand impressions and Clicks on cost per click. A third objective, Conversions, arrived after launch: it bills on impressions or on clicks while delivery is aimed at one conversion event you choose before the campaign starts and cannot change afterwards. The advertiser sets a maximum bid at ad group level, and OpenAI picks between eligible ads using what it calls a relevance-weighted second-price auction, so a more relevant ad can take a placement off a higher bid. Reporting runs to impressions, clicks, spend, click-through rate and conversions, with a pixel, a conversions API and tracking parameters that survive the click. Read how the buying works, and read what the conversions objective does.
Targeting is worth reading twice. An advertiser supplies context hints at ad group level: descriptive phrases for the conversations, topics or scenarios where the product is relevant. OpenAI is explicit that these are not exact-match keywords and do not secure delivery in any particular conversation. Its ads system makes the delivery call itself, on the context and intent of the live conversation, the ad's own copy and landing page, and, where the person has ads personalisation on, selected signals from their broader use of ChatGPT. So it is a placement you influence and do not control, closer to a social buy than a keyword buy. OpenAI publishes no performance benchmarks for the format, across advertisers, industries or campaign types. Read the advertiser FAQ. If a partner quotes you a benchmark, ask where it came from.
"Ads do not influence ChatGPT's answers. Ads run on separate systems from our chat model, and advertisers have no ability to shape, rank, or alter ChatGPT's responses."
That is the help centre article on ads in ChatGPT. OpenAI lists answer independence as one of its stated principles for the format, says the model does not see the ads shown to a person by default, and says an ad is not an endorsement. Read the article, and read the post that launched the format.
Now the narrow question. OpenAI makes that statement about answers and responses. The ads documentation does not restate it in terms of sources or citations, so on whether a paid placement can affect which pages the model retrieves and quotes, the wording covers the response and does not name the source list. That is ambiguous, and we would rather say so than read a conclusion into it either way.
The nearest explicit statement sits on the shopping side of the product, where OpenAI says product results are selected independently, are not ads, and are not influenced by any of its partnerships, and that merchants are ranked on factors like availability, price, quality and whether they are the maker or primary seller. A product feed keeps that data accurate; OpenAI does not describe it as a way to rank. Read how product results are chosen. Product results are not the citations behind a written answer, so that covers the shop next door rather than the retrieval the answer is built from. Nothing in the documentation offers a route from an ad buy to the source list, and nothing carves the source list out of the independence claim either. That is a thinness in the wording, not a loophole.
A sponsored card buys a labelled placement, a click and something measurable after it. It does not buy the paragraph above it. That paragraph is written out of whatever the model retrieved, which is a different market with a different price.
Foundation Marketing and AirOps, May 2026, found that 90% of the sources AI cites when answering are ones the brand does not control. The market the sample covers is not stated. Read the study. And most of that ground is unclaimed. Semrush with Kevin Indig, in Search Engine Land, July 2026, on United States data, found that 15.2% of categories have a brand the models clearly recommend. The other 85% are unowned. Read the analysis. The second finding is the more urgent one. Ground nobody holds is open to take, and hard to take back once a rival holds it.
So the card and the recommendation are two purchases. The first is settled in an auction and reported in clicks. The second is won by commissioning the pages the model reads: a topic buyers ask about, a set of unaffiliated titles that cover it, one publishing window. One half of this work is a site job and the other is a media buy. The ad is a third thing, and none of the three does the other two's job. Buy the card and report it as AI visibility and you end up with a click count and no position in the answer.
The case for the ad line is real. People arrive at an assistant mid-decision, saying what they want and what their constraints are, and OpenAI says an ad can appear after a single high-intent prompt rather than only after a long exchange. The click is measurable, the conversion is measurable, and the reporting is the kind a performance team already reads.
The catch for an Australian or British advertiser is the category list. In these early phases OpenAI limits ads to household and consumer goods, local services, travel and entertainment, digital products and education. Financial services, health services and legal services are restricted, and the restriction is geographic: they run for approved advertisers in the United States, health services outside the United States are in OpenAI's own words generally prohibited, and legal services outside it are prohibited. Adult content and dating, gambling and betting, alcohol, tobacco, illegal drugs, cryptocurrency, credit repair and debt settlement, and political content are disallowed outright, and no ad is eligible to appear alongside a sensitive or regulated conversation. Read the ad policies. So for a bank, an insurer, a health fund, a wagering brand or a law firm in Sydney or London, the ad line is not open today. The answer still is. Nobody in those categories can buy the card, and everybody in them can compete for the sources it sits under.
The sponsored card belongs on the paid media line, priced against clicks and conversions, held to the standard of any other auction buy, and worth opening if your category is allowed. The answer belongs on the publisher line, because it is decided by pages you brief rather than pages you own, and it reports on whether the answer moved on the questions you agreed to be judged on before anything ran.
Keep one eye on the boundary. OpenAI says it plans to support further formats, objectives and buying models, and the conversions objective shows how quickly that list moves. Read the latest update. Answer independence is restated in each announcement we read, including that one. It is a decision about a product, not a law of physics, and if it changes it will change on the record. Until then, you can buy the placement and you cannot buy the answer.
So get the source list first. A free AI Visibility Report runs the questions your buyers ask across the engines and shows where you get named and which publications the answer is written from. Those sources are rarely your own site. If the ad line is open to your category, run it. Just do not put it on the row where the answer was supposed to go.
10 September 2026 · 6 min read
Before any engine can quote your title, the page has to clear conditions that have nothing to do with the writing. The six that decide it, the two halves of a quotable page, and the crawler distinction most robots.txt files have never made.
9 September 2026 · 6 min read
The largest kind of source in an AI answer is often one nobody can buy. Here is the subtraction that shows what is actually left to commission in your category.
9 September 2026 · 5 min read
Someone has probably told you to publish an llms.txt file. It is cheap, and it has real uses. It is also a file the largest answer engine says it ignores, so here is the difference between what reads it and what decides your visibility.
See how your brand appears in AI answers, and where trusted publisher content can strengthen its visibility.
