Agent Ads in 2026: How Brands Began Buying Advertising Seen by AI Bots, Not People

Digital advertising has traditionally been built around human attention: a person sees a banner, watches a video, reads sponsored content or clicks a search advertisement. In 2026, an unusual alternative began moving from theory into paid experiments. Some advertisers started buying messages designed primarily for AI crawlers and agents that collect information from websites and use it when answering questions, comparing products or assisting with purchasing decisions. The emerging format is commonly described as Agent Ads. It remains a small and experimental part of advertising rather than an established replacement for search, display or social media campaigns. Still, the idea is significant because it changes the immediate audience of an advertisement. Instead of persuading a person directly, an advertiser supplies structured information that software can process before that information potentially reaches a human through an AI-generated answer.

How Agent Ads Moved From an Idea to a Paid Advertising Experiment

One of the clearest real examples appeared at TIME in 2026. The publisher had begun creating simplified Markdown versions of its webpages, removing much of the visual design and leaving text in a format that automated systems could process efficiently. By late July, TIME was working with advertising technology company Mobian to place sponsored material inside these machine-readable pages. Ally Bank and the Project Management Institute were reported among the first advertisers. Rather than traditional banners, the advertisements appeared as text organised around questions and answers. They were designed to provide an AI system with understandable statements about a company, its services and subjects associated with the advertiser.

The buying logic is different from ordinary display advertising. A banner is valuable when a person notices it, while an Agent Ad is intended to become readable input for automated systems retrieving information from the web. Mobian and TIME described a process in which information supplied through an advertiser brief could be turned into an FAQ-style advertisement, reviewed by people before publication and identified as sponsored material. The commercial objective is therefore not simply to generate another impression. Advertisers are testing whether paid, clearly structured information placed on a recognised publisher’s site can improve the accuracy, visibility or context of their brand when AI services subsequently research related subjects.

This experiment sits within a much larger change in advertising. A McKinsey survey conducted among 182 US advertising and marketing leaders in February 2026 found that more than 90% were already using AI for activities such as media planning, budget allocation, targeting optimisation or creative work. More than half of respondents also reported investment in advertisements embedded in AI-generated answers. That broader category is not identical to Agent Ads: advertisements shown inside an AI answer are still presented to a human, whereas Agent Ads are created for software consumption further upstream. The figures nevertheless show why advertisers are paying attention. AI is no longer being used only to produce campaigns; it increasingly participates in the process through which consumers receive commercial information.

What AI Bots Actually Read When They Encounter Advertising

AI agents do not necessarily respond to advertising in the same way as people. Human advertising can depend heavily on photography, typography, animation, humour, celebrity recognition or emotional associations. An automated agent working through webpage content is more likely to rely on accessible text, clear labels, product attributes, prices, names, descriptions and relationships between pieces of information. A 2025 academic experiment involving several AI agents found that purely visual calls to action were frequently overlooked, while semantic labels and clearly identifiable interactive elements made advertising easier for agents to detect. The study was conducted in a controlled environment, so it should not be treated as proof that every commercial AI system behaves identically, but it illustrates the basic change advertisers must consider.

For an advertiser, this shifts attention from visual impact towards informational clarity. An Agent Ad can contain straightforward questions such as what a company provides, who a particular service is intended for, how an offer works or what distinguishes one product category from another. Dates, conditions, eligibility requirements and other factual attributes become more important because an AI system may need to compare them with information from competing sources. Brand language that sounds persuasive to a person may provide little value if it does not contain verifiable facts. For the same reason, vague superlatives are risky: an AI service may ignore unsupported claims, contradict them with another source or reproduce them in a context the advertiser did not intend.

This also explains why machine-readable advertising should not be confused with secretly writing instructions for a chatbot. The more defensible model is closer to structured sponsored publishing: the advertiser pays for identifiable commercial information, the publisher makes its status clear, and an AI system remains free to use, ignore or challenge the material. No advertiser can reliably guarantee that buying such an advertisement will produce a favourable answer in ChatGPT, Gemini, Perplexity or another service. Retrieval methods, ranking systems, source selection and model behaviour can all change. Agent Ads therefore sell an opportunity to make verified brand information available to automated readers, not ownership of the final answer produced for a consumer.

Why Brands Are Paying for Machine-Readable Visibility in 2026

The commercial interest becomes easier to understand when the scale of automated web activity is considered. DataDome reported that its network processed 17.7 billion AI-agent requests during the second quarter of 2026, up 45% from the previous quarter. The company based the report on traffic across more than 400 businesses, so the figures describe its own network rather than the entire internet. Even with that limitation, the growth demonstrates that automated visitors now represent a substantial audience for many websites. At the same time, the relationship between crawling and referral traffic is uneven. An AI service can consume information without sending a visitor to the original source, meaning that publishers and advertisers need measures beyond ordinary website sessions.

For brands, the attraction is strongest when customers frequently ask AI systems questions before making a decision. Financial services are a useful example because people may ask about account features, fees, eligibility or terminology rather than searching directly for a company homepage. Professional organisations face a similar issue when people ask about qualifications, career development or industry standards. The participation of Ally Bank and the Project Management Institute in the TIME experiment reflects these kinds of information-heavy customer journeys. In such situations, appearing accurately in an AI-generated explanation may matter even when the consumer never sees the sponsored material from which some of the underlying information originated.

Retail is moving in the same general direction, although through different advertising formats. In January 2026, Google announced the Universal Commerce Protocol, intended to help AI agents interact with retailers during product research, purchasing and after-sales activity. Google also introduced a Direct Offers pilot in AI Mode, where relevant commercial offers such as discounts can appear to shoppers at moments when the system determines that the user is close to making a purchase. These sponsored deals are human-facing and should not be classified as Agent Ads. Their importance lies elsewhere: they show that AI-assisted product research is moving closer to transactions. If agents increasingly compare products, interpret offers and help complete purchases, the commercial value of supplying them with accurate information rises as well.

How Advertising Measurement Changes When the Immediate Audience Is Software

Traditional advertising metrics do not disappear, but they become less useful on their own. A campaign created for an automated reader cannot be judged primarily by viewability, banner click-through rate or time spent looking at an advertisement. Marketers instead need to study whether relevant AI answers mention the brand, describe its products correctly, cite the intended source and retain important qualifying information. They can also track subsequent referral visits and conversions where these occur. These indicators should remain separate. A higher frequency of brand mentions does not automatically mean stronger commercial performance, while an increase in referrals does not prove that an Agent Ad caused the change.

A practical measurement process requires repeated testing rather than a handful of manually selected prompts. A company could build groups of realistic customer questions covering brand terms, generic product categories, comparisons and informational queries, then record responses from several major AI services over time. The same questions should be repeated because AI answers can change as web indexes, retrieval methods and models are updated. Useful measurements include share of relevant answers, accuracy of factual attributes, citation frequency, sentiment or favourability, referral quality and eventual conversion behaviour. A control group of questions or markets without the advertising treatment can make the results more meaningful.

Attribution remains one of the hardest problems. An AI-generated answer may be assembled from a company website, publisher articles, product feeds, reviews, structured databases and other sources at the same time. The advertiser may never know precisely why one source was selected and another ignored. McKinsey’s 2026 survey found that 42% of advertisers considered dependence on opaque optimisation systems a key AI-related media risk. Agent Ads add another layer of uncertainty because the advertiser pays for information to be available to an AI reader but does not control the reader’s retrieval rules. Reporting therefore needs to separate measurable changes from assumptions about what caused them.

The Trust, Policy and Business Risks Surrounding Agent Ads

The most immediate problem emerged only weeks after the TIME experiment became public. In August 2026, Perplexity said it was preventing TIME’s Agent Ads from influencing its systems and characterised the approach as cloaking because automated visitors were being directed towards content that differed from the normal human-facing version of a webpage. This is a serious issue because cloaking has a long history in search policy. Google currently defines it as presenting different content to users and search engines with the intention of manipulating rankings or misleading users, and its spam rules also address attempts to manipulate generative AI responses in Google Search. That does not mean every machine-readable advertisement automatically violates Google’s rules, but it makes transparency and intent central questions.

Supporters of Agent Ads argue that clearly labelled sponsored material containing cited and verifiable facts is different from hidden spam. Critics respond that disclosure inside a bot-oriented page does not fully solve the problem if ordinary readers receive different information. Both arguments matter because the long-term future of the format will depend less on what advertisers call it and more on how AI companies, search engines, publishers and consumers judge the practice. A cautious publisher should therefore make the commercial relationship explicit, apply normal advertising standards, check every factual statement and avoid material designed to impersonate independent editorial coverage.

There is also a basic commercial dependency that advertisers cannot eliminate. An AI company can change its crawler rules, exclude sponsored blocks, alter source-ranking methods or stop reading a particular machine-readable format. Publishers are simultaneously gaining more control over automated access. Cloudflare, for example, has been testing Pay Per Crawl, which allows participating site owners to charge AI crawlers for access rather than treating every automated request as free traffic. These developments point towards a web in which automated access, content licensing and advertising may be negotiated more deliberately. They also mean that an Agent Ad campaign working today could lose reach after a policy or technical change tomorrow.

What a Sensible Agent Ads Strategy Looks Like in 2026

For most brands, Agent Ads make more sense as a controlled experimental budget than as a major replacement for established media. The starting point should be the company’s own public information. Product names, prices, terms, support documentation, company descriptions and important policies need to be accurate, internally consistent and easy for both people and automated systems to interpret. Paying to distribute structured claims while the company’s own website contains outdated or contradictory information creates a measurement problem and a reputational risk. Brands should first establish which questions customers ask AI services and where existing answers contain genuine gaps or errors.

It is also useful to separate several activities that are often grouped together under AI advertising. Improving owned content so that answer engines can interpret it is one activity. Buying machine-readable sponsored material from a publisher is another. Purchasing advertisements that appear visibly inside AI-generated answers is a third, while supplying structured product and commerce data to shopping agents is a fourth. Each has a different audience, cost structure and measurement method. Treating all of them as a single tactic makes it difficult to know what is working and can encourage unrealistic expectations about a brand’s ability to influence AI-generated responses.

The strongest approach in 2026 is built around verification, disclosure and repeatable measurement. Every sponsored statement intended for automated consumption should also be defensible when read by a journalist, customer or regulator. Advertisers should record what was published, when it was available, which AI services could access it and how relevant answers changed afterwards. Policies from search and AI companies also need regular review because the rules are developing quickly, as Perplexity’s response to the TIME experiment demonstrated. Agent Ads may become a recognised advertising category, evolve into a different form or remain a specialised experiment. What is already clear is that marketing is beginning to address a new intermediary between brands and customers: software that can research, compare and increasingly act before a person makes the final decision.