Jonah Goodhart on Mobian Agent Ads

We sat down with Jonah Goodhart, CEO and co-founder of Mobian, to talk about a problem advertisers are just starting to grapple with: as AI agents take on more of the research behind our decisions, reading pages and synthesizing answers on our behalf, how do you reach an audience that doesn't scroll, click, or respond to creative?

Goodhart's answer is Agent Ads, machine-readable blocks of brand-verifiable, advertiser-approved facts served to AI agents as they read publisher pages in real time. He walked us through how it works, what it means for targeting and attribution, and how he thinks about the manipulation question that comes with advertising to machines instead of people.

Curious what he has to say? Dive into his answers below.

In plain terms, what exactly is an “Agent Ad,” and how is it different from the display, search, and programmatic inventory buyers work with today?

An Agent Ad is a machine-readable block of brand-verifiable facts, advertiser-approved and source-cited, served to AI agents on the agent-readable version of a publisher page. The first line is a sponsorship disclosure and every fact carries a clear citation. 

The world has split into two audiences, humans and AI agents. Traditional display, search, and programmatic ads are built for people. They use creative, placement, and storytelling to shape attention and preference. An Agent Ad is built for an AI system that is gathering information. The atomic unit of advertising for a human is the brand and the story; for an AI agent, it is credible, trusted, and properly sourced facts. 

Give us a concrete example using your launch partners. How would this actually play out with TIME and Ally Bank? Are you relying on the agent to discover and read the ad in real time as it answers a question?

Suppose someone asks an assistant which bank might fit their needs. To build the answer, the assistant reads relevant pages. When it reads a TIME page, it receives the agent version, and on relevant pages that version carries an Agent Ad, for instance a set of brand-verifiable, approved facts about Ally in question-and-answer form with a disclosure and citations. 

The agent reads that unit in real time as part of its research; agent platforms also read these pages when indexing content. In both cases, the brand facts are present in the information layer the agent is using. 

How does targeting work when the “audience” is an AI agent? Can buyers target by query intent, publisher, vertical, or agent type?

Targeting is contextual. A buyer can align Agent Ads with topics, genres, personas, publishers, franchises, and date ranges. Reporting is available at the content level as well as by agent and agent type.

Can a brand control which agents or which kinds of questions their ads show up against?

A brand controls the creative itself and the context it appears in. The advertiser approves every word in the markdown file, which functions as the creative, and nothing serves without approval. The buyer can choose the content environments it wants to run against; advertisers can also target by agent if they want to test different messaging or campaigns across different agents. 

Brands do not have direct control over the user questions that lead to the consumption of Agent Ads, just like a brand doesn’t control what a human user chooses to read or watch. Just like with human experiences though, the brand facts get exposure to an agent at the time of consumption on a publisher page, and that is a fundamentally new concept.

What does attribution look like? How do you connect an Agent Ad exposure to a downstream outcome like consideration or a purchase?

Agent Outcomes combines two complementary forms of measurement: Causal Ad Impact and Model Impact.

Causal Ad Impact measures whether a single Agent Ad exposure changes an assistant’s answer. We continuously run randomized paired trials using two identical versions of a real publisher page, one with the Agent Ad and one without it. After reading one version, the assistant is asked a campaign question separately and is free to answer from everything it knows. Live search is disabled, so exposure to the ad is the only informational difference between the two groups. Question-level results are reported as Causal Ad Lift, and combined across the campaign’s questions and assistants into a statistically tested Causal Ad Impact score. This measures the effect of exposure at answer time; it does not claim that the underlying model has changed.

Model Impact tracks whether the brand’s representation across AI assistants changes over the course of the campaign. Before launch, we establish a baseline by asking the campaign questions multiple times each day across leading assistants. We continue that measurement during the campaign, tracking visibility, favorability, accuracy, average ranking, and how often the brand is named first or elsewhere in the response. This shows when results move beyond their normal baseline range and whether a shift is sustained or statistically significant. Because other factors can also affect model behavior, these changes are not automatically attributed to the campaign.

For downstream outcomes, we also report which pages were read and by which agents. Advertisers can then connect agent-referred traffic with their own site activity and sales data to evaluate consideration and purchase behavior.

Early lab tests suggest agents do not ignore these ads. What is the most convincing performance evidence you can share so far?

In terms of lab tests there is a study that examined whether AI agents interact with online ads (arXiv 2504.07112, https://arxiv.org/abs/2504.07112). It found that the tested agents neither ignored nor systematically avoided ads, and that they favored features such as keywords and structured data.

In our campaigns, the early readings are positive, including statistically significant movement in the brand outcome measures we track. It is early but the signals are strong, which is why many brands are engaging.

The AI platforms themselves, including OpenAI/ChatGPT, Perplexity, and Google, are building their own advertising businesses. How do Mobian Agent Ads fit alongside those? Complementary or competitive?

They are completely different concepts and products, just like organic vs. paid. A platform’s native ad is generally designed for the person using that platform and appears in or alongside the answer. An Agent Ad is delivered upstream, when an AI agent is reading relevant publisher content to research the answer.

One model monetizes the platform interface. The other makes approved, cited brand facts available in the contextually appropriate content the agent consumes. These are different moments, and multiple models are likely to coexist and matter to brands in these environments.

If ChatGPT or Perplexity sells ad placements directly inside their answers, why would a brand also invest in Agent Ads on the publisher side? What does the publisher-page approach give buyers that a platform’s native ads do not?

The publisher-side approach has two advantages. First, it can reach an agent during consumption, regardless of where the final answer is presented. Second, it supports the publishers whose content models use to train, index, and answer questions.

It puts disclosed, cited brand-verifiable facts into the source environment and creates a way for publishers to monetize their agent traffic. A platform-native ad may reach the user at the end of the process. An Agent Ad can shape the information set earlier without guaranteeing or dictating the answer.

You are currently live with TIME. How do you think about publisher supply and scale, and what does inventory growth look like from here?

We are starting with premium publishers and content environments where agents already do meaningful consumption. TIME is the first live publishing partner, stay tuned for more news. Inventory grows as publishers make more pages agent-readable and as we expand across franchises, verticals, and relevant contexts. The goal is high-quality supply where the brand facts are relevant to what the agent is trying to understand.

On the demand side, Ally’s experience illustrates why brands are paying attention. Andrea Brimmer told Business Insider she was quick to experiment with advertising to AI agents because “the signals are impossible to ignore.” Consumers visiting Ally’s site from AI platforms were 3.5x more likely to open an account than those from traditional search, and Ally’s AI-driven traffic is up 9x year-over-year. Brimmer said she would “rather be in the room helping shape what comes next than standing outside it,” and we’re excited to be shaping what’s next alongside her. 

Your own lab study found that agents favor structured data over content made for humans. Does that create built-in bias? If an AI prefers reading the Agent Ad, will it end up recommending the brands inside those ads over competitors that do not run them? How do you keep that from tilting the playing field?

The format is designed to be inspectable, not privileged. Sponsorship is disclosed on the first line, every fact has a citation, and anyone can see exactly what the agent received. An agent remains free to weigh those facts against other sources. There are no hidden instructions telling it what conclusion to reach.

The playing field is already uneven because agents often assemble brand information from third-party sources that may be outdated or inconsistent. In a Mobian study of more than 750 brands, roughly 17 percent of cited sources conflicted with the brand’s own facts or positioning. Publishing verified, cited information gives brands a transparent way to correct that gap. It is available to any advertiser, and brands can also make their own sites more machine-readable.

Advertising to machines invites obvious manipulation concerns. What guardrails keep this honest, and how do you see standards or measurement currencies developing for the category?

Manipulation involves concealment. This format does the opposite: the disclosure is on the first line, each fact is cited, the advertiser signs off on every word, and the whole unit is inspectable. The core guardrails are transparency, source provenance, and a clear separation between facts and instructions. In addition, the pages that agents crawl today without this solution are filled with sponsored content as well as a huge variety of paid content, some disclosed, some less disclosed. Many are arguing this is far more transparent as it cleans up the page and has a single, clearly labeled block instead of what exists today.

We support  industry standards for disclosure, exposure, and outcomes. The IAB recently released its principles on Measuring Visibility in the AI Era and we have incorporated them into our approach. We look forward to continuing to collaborate with industry bodies like the IAB as this space evolves. 

What do you predict will be the biggest challenge to getting brands and publishers to invest in this today versus waiting?

The biggest challenge is habit. Budgets, workflows, and measurement currencies are built around human audiences. Agent-facing media is new, so buyers and publishers will look for benchmarks, pricing experience, and a history of measured outcomes before it becomes a standard line item.

The counterweight is that the behavior is already here. Brands and publishers that test now will build the baselines and operational knowledge others will eventually need. Waiting may feel safer, but it also means learning later, after agents are already shaping more discovery and purchase decisions.

If agents keep taking on more buying decisions, how should media buyers be reallocating budget and building capability over the next few years? What do brands that ignore this stand to lose?

There are two audiences, humans and agents, and media plans today are only built for one of them. Buyers should start by building out their agent-facing capabilities: audit the facts AI engines use, make approved information machine-readable, test agent-facing inventory, and add AI visibility, favorability, and accuracy to the measurement plan alongside human media metrics.

Brand investment should follow the evidence. As agents take on more research and buying decisions, brands that are present and accurately represented in the sources agents read will have an advantage. Brands that sit out leave their story to third-party sources, and our data shows those sources do not always match the brand’s facts or positioning.