Welcome to today's edition of AI Marketing Geek, your daily dose of the latest AI developments, tools, trends, and actionable insights shaping marketing in 2026. As AI shifts from experimental hype to pragmatic, ROI-driven reality, we're seeing agentic AI take center stage. Autonomous agents are handling media buying, personalization, and commerce, which is wild to watch.

Something is quietly shifting inside marketing agencies, and it deserves more attention than it's getting. A wave of shops are no longer waiting for AI vendors to build the tools they need. They're building those tools themselves, in hours, using AI coding assistants.

Agencies like Havas, Broadhead, and Supergood are vibe coding their own generative engine optimization tools on top of large language models, often in a matter of hours. Using coding assistants like Anthropic's Claude Code, teams are building bespoke applications that analyze how brands appear in AI-generated responses, track competitors, and in some cases package those tools into products sold to clients.

The specific examples are worth paying attention to. Havas built Brand Insights AI, a GEO product using Claude Code and Replit that generates prompts based on a client's brand, runs them across multiple models, and analyzes how often a brand appears in responses, including citations. The platform has been rolled out globally, covering nearly 100 countries and more than 60 languages, and is licensed to clients as a SaaS product.

At Broadhead, the timeline is even more striking. VP of product innovation Mitch Hislop said he vibe coded the first version of the agency's GEO monitoring platform in a single evening using Claude Code. One of its earliest features was a "competitive intelligence vote," where a user inputs a brand and location, and an LLM returns the competitors it is most likely to surface. Adding audience personas to simulate how different types of consumers might query tools like ChatGPT or Claude took about two hours.

The strategic implication here runs deeper than the cool factor. Agencies that build proprietary tools own something competitors can't copy with a subscription. They can license those tools to clients as a separate revenue line. And they can iterate on them at a speed no traditional software vendor can match. The agencies that figure this out first will have a structural advantage in the next generation of GEO and AI visibility work. The ones waiting for someone else to build the tools will just be buying someone else's product.

There's a billing problem quietly building inside marketing agencies, and a recent Digiday investigation surfaced just how messy it's becoming. The question is simple: when AI generates content for a client campaign, who pays for the compute?

Generative AI prompts and responses don't come for free. LLM developers like OpenAI use tokens as a means of metering AI compute, and while the tokens themselves cost a fraction of a cent, they add up fast. A recent Coca-Cola ad campaign required 70,000 prompts and millions of tokens, according to its makers.

Agencies are landing on very different answers. Chris Neff, global chief AI officer at Anomaly, was wary of passing on AI overheads to clients. "It feels like a money grab," said Neff. Others are moving the opposite direction. Brandtech's AI platform Pencil negotiates bulk rates with providers like Anthropic and OpenAI, then charges clients "generation credits" equivalent to a single chat response, an image generation, or a second of video. The more a client commits, the cheaper Brandtech can make it.

That model is raising its own set of questions. Brandtech's bulk token deals invite comparison with one of advertising's most contentious practices: principal media buying. If an agency can acquire large amounts of tokens at a discount, what stops them from passing that cost on to clients at a markup?

And this is likely just the beginning of the scrutiny. Ebiquity CEO Ruben Schreurs told Digiday he expected his consultancy to be asked, in time, to audit an agency's token usage the same way it audits media spending now.

For any marketing firm using AI tools in client work, this is a conversation you need to get ahead of. The clients who figure out to ask about token costs will start asking about them. Better to have a clear, defensible answer ready than to have a CFO discover an unmarked line item they don't recognize.

The story above is a preview of a conversation that's coming to every agency relationship. Here's how to get ahead of it now.

First, actually know what you're spending. Most AI platforms have a usage dashboard. Go look at it. If you're running campaigns through tools that use OpenAI, Anthropic, or Google APIs, you have a cost-per-run that you may never have looked at. Pull it.

Second, decide on a policy before a client asks. There are three defensible options: you absorb AI compute costs as part of your margin, you pass them through transparently as a line item, or you build them into your project rate. What you can't do is have no answer when the question comes up.

Third, if you're going to pass costs through, build a simple tracker now. Log what tool, what client, and roughly how many runs per project. You don't need a spreadsheet from hell. You need enough visibility to answer "how much did AI cost us on this campaign?" with a real number instead of a shrug.

The agencies that treat token costs like media costs, trackable and auditable, will have a much cleaner conversation with clients than the ones that have been quietly absorbing them and hoping nobody notices.

2026 is the year AI moves from "cool demos" to real and measurable business use cases. Expect more agentic tools, voice-powered targeting, and a focus on privacy-first data.

Stay ahead of the game by treating your AI reputation like your website. It would behoove you to make your brand easy for agents to understand and cite.

What AI experiment are you running this week? Reply and let me know. I'll feature top stories in future editions! (Everybody loves a good story)

Stay sharp,
Jon
@mistersterling
Chief AI Marketing Geek