The young business of selling ads to AI agents has run into the oldest question in advertising: how do you prove it worked? Digiday’s 20 August media briefing lays the problem out plainly. An agent-facing ad is retrieved by a bot, folded into a generated answer, and may shape a purchase decision — without the person ever seeing the original creative. There is no impression in any conventional sense, no viewability, and usually no click. Agents draw on dozens of sources per answer, so even when a recommendation lands, tracing it back to the ad that seeded it borders on impossible. As Brendan Norman, founder of ad tech firm Classify, put it: “There’s a lot of things to consider with this because it’s not as straightforward as a clickthrough or a checkout.”
Last week we covered the governance half of this fight, when Perplexity blocked Time’s markdown ads — the sponsored copy Time embeds, with Mobian, in the stripped-down page versions agents actually read — and called the format deceptive. This week’s story is the commercial half of the same trendline: even where platforms permit agent-facing formats, nobody yet agrees on how to count them, verify them or price them. A market can survive a format war. It cannot survive without a currency.
The workarounds now emerging are inventive and mutually incompatible. OpenAds inserts unique referral codes — a 10% discount, for instance — into pages as ChatGPT scrapes them; if the model judges the code relevant enough, co-founder Steven Liss says, it will “survive the retrieval” and surface in an answer, where a click becomes trackable. His summary of the buy side’s mood doubles as the industry’s problem statement: “The best currency is currency.” Oasy, co-founded by Choy Travers, is pursuing cost-per-click and cost-per-referral models through dynamic ad insertion at the CDN layer, via integrations of the Cloudflare kind. Time measures its markdown programme through Mobian’s AI-visibility scores. And the IAB is drafting attribution frameworks for agent-mediated advertising, having already released standards for measuring AI visibility and disclosing AI usage. Until something converges, Digiday reports, advertisers are funding agent-facing formats from experimental budgets only.
01No currency, no market — and no verifier either
Every advertising medium that scaled did so on a unit somebody neutral could count: the audited circulation figure, the ratings point, the measured, viewable impression with third-party verification layered on top. Agent-facing advertising currently has none of that stack. The metrics on offer are proprietary — one vendor’s visibility score, another’s surviving referral code — which means every seller is grading its own homework. Publishers should recognise the shape of this moment, because they have lived it: early programmatic also ran on unverifiable claims, and the publishers who prospered were the ones who backed standards and audits early rather than riding the ambiguity. The vacuum is also an invitation. There is, as yet, no MRC of the agent web — no accreditation body blessing how a machine-read placement is counted — and whoever fills that role will set terms publishers live with for a decade.
02The publisher’s stake is the denominator
There is a second, quieter reason this fight matters for publishers: every attribution scheme in play runs through the publisher’s page. The referral code is inserted into publisher content at scrape time; the CDN-layer ad rides publisher infrastructure; the visibility score is computed over publisher markdown. Publishers are the substrate of agent-facing advertising — which means they hold more leverage over measurement design than they did in the programmatic build-out, where the counting happened in other people’s servers. Data access, log rights and audit rights are negotiable now, while vendors need publisher inventory to prove their models. They will be far less negotiable once a winner emerges.
Agent-facing advertising has real advertisers, real formats and real disputes — what it lacks is a countable, auditable unit that a buyer, a seller and a neutral third party can all agree on.
03Why this matters for publishers
| Revenue stays experimental until measurement matures | Advertisers are explicit that agent-facing spend comes from test budgets until bottom-funnel attribution improves. Publishers should size this line accordingly and resist booking it as durable. |
|---|---|
| Proprietary metrics mean vendor lock-in risk | A programme measured only in one intermediary's visibility score is a programme you cannot benchmark, cannot audit and cannot take to a competing partner with evidence. |
| The counting happens on your pages | Codes injected at scrape time, CDN-inserted units, markdown scoring — every current method touches publisher infrastructure, which is negotiating leverage most publishers have not priced. |
| Standards are being written now | The IAB's attribution drafts and visibility standards are open for shaping at exactly the moment publisher input is cheapest and most influential. |
04What publishers should do
05The bottom line
Agent-facing advertising has real advertisers, real formats and real disputes — what it lacks is a countable, auditable unit that a buyer, a seller and a neutral third party can all agree on. Until that exists, the money stays experimental and the metrics stay proprietary. The Perplexity–Time fight showed that platforms will contest who governs agent-facing formats; the measurement gap shows that even the permitted versions cannot yet be priced with confidence. For publishers the assignment is the same in both halves: be the party with the best records. Instrument your own agent traffic, contract for raw data, corroborate every vendor claim, and help write the standards. Currencies are being minted; publishers should be at the mint, not waiting at the exchange window.