Ad Agencies Are Vibe-Coding Their Own GEO Tools with Claude Code
Generative Engine Optimization (GEO) has become the most urgent capability gap in advertising — and ad agencies are closing it overnight with vibe coding and Claude Code, Anthropic's agentic coding tool.
On March 4, 2026, Adweek reported that agencies including Havas, Broadhead, and Supergood are vibe-coding bespoke Generative Engine Optimization (GEO) monitoring platforms in a single evening. No developer hired. No six-figure software contract. Just a non-programmer, a chat interface, and Claude Code.
This is what happens when vibe coding meets real enterprise need — and it has implications for every AI development studio.
What Generative Engine Optimization (GEO) Is and Why Agencies Need It Now
Search has fractured. According to Gartner's 2025 forecast, traditional search volume will decline 25% by 2026 as users shift to AI-powered alternatives. Millions of people now open ChatGPT, Perplexity, or Claude before they open Google. They ask: "What's the best project management software for a remote team?" The AI answers — and if your brand doesn't appear, you effectively don't exist.
Generative Engine Optimization (GEO) is the practice of ensuring a brand appears favorably in AI-generated responses. The term was formalized by Aggarwal et al. in their 2023 Princeton research paper, which demonstrated that specific content optimization strategies can increase source visibility in AI-generated responses by up to 40%. Where traditional SEO optimizes for keyword rankings in a web index, Generative Engine Optimization targets brand salience in language model outputs: how often a model cites a brand, in what context, and compared to which competitors.
The tooling to measure Generative Engine Optimization barely existed twelve months ago. Off-the-shelf GEO platforms exist — Profound, Bluefish, and Emberos are all competing in this space — but agencies found that generic solutions don't map to how they actually work. Multiple client portfolios, proprietary brand frameworks, specific prompting strategies. The Generative Engine Optimization tools they needed didn't exist. So they built them with Claude Code.
How Vibe Coding with Claude Code Makes This Possible
Mitch Hislop, VP of Product Innovation at Broadhead, built the agency's entire Generative Engine Optimization monitoring platform in one evening using Claude Code. One evening. One person. Zero lines of traditional code written by hand.
His Claude Code-built tool analyzes how different AI providers rank a brand and its competitors, runs a "competitive intelligence vote" where a user inputs a brand and location and an LLM returns which competitors it would most likely surface, and layers in audience personas to simulate how different consumer types query AI tools.
That persona-layering upgrade — transforming a basic brand tracker into a full-spectrum competitive intelligence system — took about two hours of vibe coding with Claude Code.
Claude Code enables this because it holds enough context to build a complete application architecture, not just snippets. As detailed in Anthropic's Claude Code documentation, the tool operates as an agentic coding system that reads files, runs commands, and iterates autonomously. You describe the Generative Engine Optimization tool you need, Claude Code generates working code, you iterate through natural language. For someone with domain expertise but limited coding background, Claude Code is transformative.
Havas scaled further. Their Brand Insights AI — also built on Claude Code and Replit — generates custom prompts based on a client's brand, runs them across multiple AI models simultaneously, and tracks how frequently the brand appears in responses. The platform covers nearly 100 countries and more than 60 languages, and Havas licenses it to clients as a SaaS product. Dan Hagen, Havas' global chief data and technology officer, says Brand Insights AI has become part of the agency's core pitch strategy.
Supergood signed an enterprise agreement with Anthropic and uses Claude models as core infrastructure across multiple applications — including organizing internal knowledge graphs and building self-evaluation loops where a model generates a response, evaluates it against criteria, assigns a score, and repeats until it hits a quality threshold. Mike Barrett, founder and CSO of Supergood: "Everybody's making software right now. In two years we are going to be delivering more software than actual documents."
What This Means for AI Development Studios
Here's the uncomfortable question for anyone running a development studio: if a marketing VP can build a working Generative Engine Optimization SaaS platform in one evening with Claude Code, what exactly is a dev agency selling?
The honest answer: engineering judgment. What vibe coding with Claude Code doesn't replace is the work that happens after the first version ships.
The cost numbers tell the story. Claude Code API access runs roughly $50-200/month for the kind of usage Hislop describes. A comparable custom development contract for a GEO monitoring platform would start at $50,000. That's a 99% cost reduction for the MVP — but the MVP is not the product.
What still requires engineering:
- Security hardening, authentication, and data residency compliance for client competitive intelligence
- Database schema design, rate limiting, and deployment pipelines at scale
- Multi-model architecture to avoid single-vendor lock-in (Hagen explicitly flagged this risk — enterprise Anthropic agreements can run "multiple millions" annually)
- Maintenance when the AI model API changes, new providers need integration, or edge cases surface
At Context Studios — one of Berlin's leading AI development studios — we've shipped Claude Code-built tools in hours that previously took days. We build GEO-adjacent tooling into our client workflows. Our experience mirrors what Broadhead reports — for scoped, well-defined tasks, Claude Code genuinely delivers 10x productivity gains.
But we've also hit the walls. Vibe-coded tools struggle when the problem domain shifts mid-build (Claude Code's context window doesn't hold cross-session architectural memory), when integrating with complex legacy infrastructure, and when security constraints require explicit design rather than pattern-matched code. We rebuilt a Claude Code prototype three times before the architecture stabilized for production — a common pattern that the "one evening" narrative doesn't capture.
Our view: Claude Code is an exceptionally capable junior developer who ships fast on well-defined tasks but needs senior oversight for anything that needs to last. That oversight — engineering judgment — is where studios like ours focus. The market for simple custom dashboards is shrinking. The market for production-grade AI systems is growing.
Caveats: What Vibe-Coded Generative Engine Optimization Tools Don't Do Well
Maintenance is harder than creation. A Generative Engine Optimization tool vibe-coded in an evening with Claude Code is easy to build and hard to maintain. When the AI model API changes or an edge case surfaces, the code Claude Code generated needs to be understood by a human. If nobody on the team can read it, that's a liability.
Security is not implicit. AI-generated code from Claude Code doesn't automatically implement authentication, rate limiting, input validation, or data residency compliance. A GEO monitoring tool handling client competitive intelligence needs real security design.
Hallucination risk compounds. Generative Engine Optimization tools built on top of LLMs are measuring systems that are themselves probabilistic. Without careful prompt engineering and output validation, a GEO tracker produces confident-sounding but inconsistent results.
Vendor lock-in is real. If you build your entire Generative Engine Optimization offering on one model provider and that provider stops being frontier enough, you have a problem. Multi-model architectures add complexity but reduce risk.
Frequently Asked Questions
What is Generative Engine Optimization (GEO)?
Generative Engine Optimization (GEO) optimizes how a brand appears in AI-generated responses from ChatGPT, Perplexity, and Claude. Unlike traditional SEO targeting keyword rankings, GEO focuses on brand salience, citation frequency, and competitive positioning in large language model outputs. The practice emerged in 2025 as AI-powered search tools captured significant market share from traditional search engines.
How did Havas build its Brand Insights AI platform?
Havas built Brand Insights AI using Claude Code and Replit in an iterative vibe coding process. The platform generates custom prompts per client brand, runs them across multiple AI models, and analyzes citation frequency. It covers nearly 100 countries and 60+ languages, and Havas licenses it as a SaaS product to clients.
Is vibe coding with Claude Code reliable for production enterprise tools?
Vibe coding with Claude Code produces functional tools quickly, as Broadhead and Havas demonstrate. Production-grade tools additionally require security hardening, error handling, scalability architecture, and ongoing maintenance. The "one evening" MVP is real; the durable SaaS product behind it still requires engineering judgment and ongoing development investment.
Why are agencies building GEO tools instead of buying off-the-shelf?
Off-the-shelf Generative Engine Optimization platforms don't map to agency workflows — multiple client portfolios, proprietary brand frameworks, specific prompting strategies. Building with Claude Code allows full customization at a fraction of enterprise software contract costs, with flexibility to iterate as AI technology evolves rapidly.
What does vibe-coded agency tooling mean for development studios?
The barrier to building domain-specific, low-complexity tools has dropped dramatically. This compresses the market for simple custom software. Complex production-grade systems requiring security design, architectural judgment, and long-term maintainability still require real engineering. AI-native studios focused on that tier are well-positioned.
Which Claude Code features enable vibe coding GEO tools?
Claude Code's extended context window holds entire application architectures in memory, generating coherent interconnected components rather than isolated snippets. Its agentic capabilities — reading files, running commands, iterating on errors — let it build and test complete Generative Engine Optimization workflows without constant human intervention.
Related Reading
- Claude Code Best Practices: Workflows, Tasks & Tutorials — Going deeper on what Claude Code can do in production
- Polsia: How a Solo Founder Hit $1M ARR in 30 Days With AI Agents — Another case study in AI-native velocity
- The AI Model Reset: Most Important Releases of February 2026 — The model landscape enabling all of this