Here is the awkward part of the news that landed in every marketing feed this week. If you buy ads for a B2B SaaS company, you will probably never open Ask Ad Manager.
Google announced Ask Ad Manager on June 18, a Gemini-built conversational agent that lives inside Google Ad Manager. Ad Manager is the sell side. It is the platform publishers use to sell their own ad inventory, the system behind the banner on a news site or the pre-roll on a streaming app. It is not the platform you use to spend money acquiring SaaS trials. So the chatbot itself is not for you.
The plumbing underneath it is a different story. That is the part worth two minutes.
What actually shipped
Ask Ad Manager is a multi-turn agent that answers a publisher’s questions about their own inventory in plain language. Google describes three jobs it does at launch. It troubleshoots a stalled line item in real time instead of making someone build reports to find the problem. It generates a custom report, table, or benchmark from a single prompt. And it navigates the platform on your behalf, loading the right filters and settings based on what you were just asking about.
The beta opened in mid-June to a mix of desktop, mobile, and connected TV publishers. It is free during testing with no query caps, though Google’s Peentoo Patel has signaled usage-based pricing may arrive at general availability later this year. Every answer is grounded in the publisher’s own first-party data using retrieval-augmented generation, so the agent is scoped to that account’s real numbers rather than a shared pool.
Useful if you run a publishing business. Beside the point if you run paid acquisition. The reason to keep reading sits one layer down.
The signal is the MCP server, not the chatbot
Buried under the launch is the line that matters. Later this year Google will ship REST APIs and an MCP server for Ad Manager.
MCP, the Model Context Protocol, is the open standard Anthropic released in late 2024 that lets any compliant AI agent plug into an external system, the way USB-C connects a device regardless of who made the cable. Google and OpenAI both adopted it. Once Ad Manager has an MCP server, a publisher’s own agent, an agency’s, or a third-party vendor’s can talk to Ad Manager directly. Yahoo is already wiring Ad Manager into custom agents that handle forecasting, line item creation, reporting, and optimization.
This is Google building the rails for AI agents for advertising to interact at scale. The chatbot is the demo. The MCP server is the product, and it is the same direction the buy side is heading. That is why the agentic ad tech shift is worth tracking even from the advertiser seat. When Google opens the same kind of programmatic access on the buy-side stack, the agencies that already know how to point an agent at an ad platform will move first.
The design tell most coverage skipped
Read the fine print and one choice stands out. The Google Ad Manager AI agent is advisory. It diagnoses, recommends, and links you to the setting, but it does not change anything until a human approves. Recommendation and execution are kept as separate steps on purpose.
That is a pointed contrast with how Google’s autonomous products already work on the buy side. Performance Max already hands Google a large amount of control over where your budget goes, with far less transparency into the why. We have written before about how Performance Max actually behaves and when it is safe to trust Google’s automation. Against that backdrop, shipping a publisher agent that asks permission before it acts is a deliberate signal about where Google thinks the trust line sits in agentic advertising. Humans approve. The model advises.
Whether that line holds as the specialized “discover, negotiate, execute” agents Google previewed actually arrive is the open question.
Three things actually worth watching
The hype framing for agentic advertising is “AI does your ad ops now.” The honest framing has caveats.
Hallucination is unsolved. Google has not published an error rate for the beta, and Patel acknowledged inaccurate answers are still a risk. An agent that confidently misreads a delivery issue costs more than a slow human who reads it right.
The benchmarks are Google’s own. The comparative numbers the Google Ad Manager AI surfaces come from Google’s platform data, not independent measurement. That matters most for a company that already holds roughly 90% of the publisher ad server market, a figure that surfaced in the DOJ ad tech case decided in 2025.
The antitrust backdrop is live. A remedies decision in that same case is expected this year and could shape how aggressively Google is allowed to build and push features inside Ad Manager.
What B2B SaaS marketers should do Monday
Skip the beta. You do not sell inventory, so there is nothing to log into.
Do one thing instead. Look at how AI is already rewriting your PPC reporting on the buy side, because the same “ask in plain language, get a report” pattern is coming to the tools you do use. The teams that get value from it will be the ones who already know which questions are worth asking. A faster answer to the wrong question is still the wrong answer.
And start treating MCP literacy as a real skill, not a curiosity. When buy-side platforms open the same kind of agent access, the advantage goes to whoever can connect their own data and their own agent to the ad stack without waiting for a vendor to build the button.
The takeaway
Ask Ad Manager is a publisher tool you will not use. The agentic ad tech architecture it sits on is the thing to file away. Google just told you, fairly plainly, that the next phase of ad platforms is AI agents for advertising talking to each other through open protocols, with a human approval gate where the stakes are high. The chatbot is this month’s headline. The MCP server is next year’s default.
When that access reaches the buy side, the question will not be whether to use it. It will be whether your data and your measurement are clean enough to point an agent at. That is the work to start now.
If your paid media is already drowning in dashboards and you are not sure which numbers actually move pipeline, that is the audit OneMetrik runs. Free, 30 minutes, no commitment: cal.com/onemetrik/30min. Or see how we think about paid acquisition on our Google Ads agency page.