The theory of LLM advertising is simple: attention has moved from the web page to the AI answer, so brand presence must follow. The practice is messier. In May 2026, OpenAI opened a self-serve advertising platform with CPC bidding and no minimum spend. Perplexity, which had been the most aggressive early mover on AI advertising, pivoted away from ads entirely and towards subscriptions. And OpenAI briefly ran promotional messages inside chat conversations before suspending them after user backlash and regulatory scrutiny. What exists today is a channel in rapid, uneven development - parts of it live and testable, parts of it still being designed, and the economics very different depending on where in the pipeline you buy.
The three models of LLM advertising
Platform-native ads
Platform-native ads are sold and served by the AI company, sitting alongside the generated answer as clearly labelled units. The clearest example is OpenAI's Ads Manager, which launched in beta on 5 May 2026. It is open to US businesses of all sizes, uses CPC bidding, carries no minimum spend, and places ads on the Free and Go tiers of ChatGPT in the US, Canada, Australia, and New Zealand. Campaigns run through a three-level structure (campaign, ad group, ad) and use conversational targeting based on prompt context rather than keywords. Early agency partners include Dentsu and Omnicom; technology integrations include Adobe and Criteo.
Google has been running ads inside AI Overviews since 2024 and has since extended them to AI Mode. Microsoft serves ads through Bing AI and has built a Publisher Content Marketplace that lets publishers license content to AI products on usage-based terms, with partners including the Associated Press, Condé Nast, Hearst, and Vox Media.
The common principle is that platform-native ads are labelled, visually separated from the generated answer, and controlled by the AI company rather than the advertiser. Advertisers target by context and intent; the platform determines adjacency and format.
Product-feed and shopping ads
Shopping represents a distinct and fast-growing format within platform-native advertising. OpenAI launched product-feed ads inside ChatGPT on 12 May 2026, with a ceiling of one million SKUs per advertiser. When a user asks a shopping-intent question, sponsored product cards appear beneath the answer showing the brand logo, a "Sponsored" label, the product name, price, stock status, and estimated delivery. Stackline data from the headphones category shows ChatGPT generated 19.4 million shopping queries over the past 52 weeks - more than Amazon's Rufus at 10.68 million. For e-commerce advertisers, this is now a live, self-serve channel.
Content-layer ads
Content-layer advertising works upstream of the answer, placing a brand inside the publisher content the assistant reads before it writes its response. Because the assistant builds its answer from that content, a brand fact present in the source material can enter the answer as context. This is the model blankspace operates. An advertiser defines accurate brand facts, targets by content category, market, and prompt cluster, and when a Live Search Agent retrieves a matching publisher page, the brand fact is injected into the content at the CDN edge before the agent finishes reading. The publisher earns the revenue; the advertiser gets presence in the answer.
The practical difference from platform-native is timing and control. Content-layer advertising runs before the answer is generated rather than alongside it, which means the brand is part of the source material the model draws on. It also generates revenue for publishers regardless of whether users ever visit the page, because it monetises the read itself rather than the click.
Why advertising entered the LLM at all
Advertising follows attention, and research intent is increasingly handled inside AI assistants rather than on the open web. OpenAI projects $2.5 billion in ad revenue for 2026, and ChatGPT accounts for an estimated 82.6% of all generative AI traffic. Google saw this shift early and moved ads into AI Overviews before competitors had a product. The underlying pressure is that the mid-funnel moment - when a user researches a product category and forms a shortlist - is now happening inside a conversation, which neither a display ad nor a paid search listing can reach.
What happened when OpenAI pushed too hard
Not every format has landed well. OpenAI suspended a test of promotional messages inside chat conversations after users pushed back on placements for brands including Peloton and Target that surfaced when they were not relevant to the conversation. On 12 June 2026, New York's attorney general issued OpenAI a subpoena covering advertising, user engagement and retention, and data practices. The episode illustrates the tension at the centre of LLM advertising: an assistant's authority depends on users trusting it, and placements that feel like the assistant is steering rather than informing erode that trust quickly.
Perplexity's pivot away from ads
Perplexity, which was among the first AI assistants to announce advertising and launched a publisher revenue-sharing programme tied to ad inventory, has since pivoted its monetisation model towards subscriptions. Its formal 2026 programme, Comet Plus, allocates a $42.5 million payout pool with publishers receiving 80% of the $5 per month subscription revenue. Average publisher earnings under the model run at roughly $8 to $15 per 1,000 citations. The revenue source is subscriptions, not advertising. The Perplexity case is a useful data point: the economics of AI advertising are still being worked out, and not every AI platform will follow the same path as Google or OpenAI.
Brand safety and control in LLM advertising
Brand safety changes shape in a conversational medium. With platform-native ads, advertisers depend on the AI company's controls over placement and adjacency. The context changes with every prompt, so traditional page-level vetting does not apply. The brand safety question shifts to whether the platform can guarantee that ads appear near appropriate conversation types, and whether the labelling is sufficiently clear that users do not attribute sponsored content to the assistant's own reasoning.
Content-layer advertising has a different control profile. The advertiser defines exactly what is said, because a brand fact is a specific, accurate statement rather than a creative that could render against unknown content. Targeting by content category and prompt cluster controls context directly. Publishers can exclude sensitive editorial content and conflicting categories. The risk of a placement looking as though it is steering the answer is reduced because the brand presence is part of a source document the model reads, not a claim attributed to the assistant itself.
What this means for publishers
For publishers, the LLM advertising landscape in mid-2026 presents two paths. The platform-native path - accepting terms from OpenAI, Microsoft, or Perplexity - gives access to licensing revenue or citation-based payouts, but on terms set by the platform and with eligibility typically skewed to larger, branded publishers. The content-layer path monetises every read by a Live Search Agent regardless of platform deals, because the monetisation sits at the CDN edge before the content reaches any assistant. The two are not mutually exclusive, but they solve different problems: platform deals compensate for content use, edge monetisation captures AI traffic that bypasses every existing ad format.
Frequently asked questions
Are there ads in ChatGPT now?
Yes. OpenAI's Ads Manager launched in beta on 5 May 2026 and is available to US businesses of all sizes without a minimum spend. Ads appear on the Free and Go tiers in the US, Canada, Australia, and New Zealand. Separately, product-feed shopping ads (with sponsored product cards) launched on 12 May 2026. Google has had ads in AI Overviews since 2024. Microsoft serves ads through Bing AI.
How is content-layer advertising different from a ChatGPT ad?
A ChatGPT ad is served by OpenAI alongside its answer, labelled as sponsored. Content-layer advertising places a brand inside the publisher content the assistant reads to build the answer, so it appears as part of the source material rather than as a separate unit. The advertiser controls the message; the publisher earns the revenue.
Do LLM ads change what the assistant tells users?
They should not, and when they do it causes problems. Platform-native ads are meant to be labelled and separate from the answer. OpenAI suspended an early test of in-conversation promotional messages after user backlash and regulatory scrutiny, because the placements felt as though they were steering conversations. In content-layer advertising, the placement is a factual statement added to source content; it can influence what the assistant includes, but it is designed to be accurate and relevant rather than to distort.
Who gets paid in LLM advertising?
With platform-native ads, the AI company is paid by advertisers. Some, like Perplexity's Comet Plus programme, share revenue with publishers whose content is cited. OpenAI has stated it does not currently plan to share ad revenue with publishers. With content-layer advertising, the publisher whose content is read earns the revenue directly.
What happened to Perplexity's advertising model?
Perplexity launched an ad revenue-sharing programme with publishers in 2024 but has since shifted its primary monetisation to subscriptions. Its 2026 Comet Plus programme pays publishers from a $42.5 million pool based on citation volume rather than ad impressions, earning publishers roughly $8 to $15 per 1,000 citations. It is a subscription-funded model, not an advertising one.