Ruder Finn has launched a new version of ruderfinn.com that puts a custom large language model at the center of the agency's digital presence. The site, built on Anthropic's Claude models via Amazon Bedrock, lets visitors ask questions in natural language instead of navigating static pages and search bars.
The platform uses retrieval-augmented generation, or RAG, to ground responses in a curated knowledge graph that maps the agency's capabilities, expertise, and engagement history. Humans built and maintain that graph, and client mentions are anonymized to honor confidentiality agreements. Traditional web pages remain available for visitors who prefer them.
Kathy Bloomgarden, CEO of Ruder Finn, said the redesign reflects where digital behavior is heading. "People don't browse anymore, they ask," she said. "If that's how the world is communicating, then our website shouldn't describe the future of communications. It should operate like it."
The agency is among the first communications firms to make an LLM the primary interface for its own site, though it has plenty of company when it comes to building such systems for clients.
A closed-loop system with human oversight
The architecture runs on what Ruder Finn calls a "closed-loop" LLM, meaning inputs, outputs, and updates are optimized with AI but pass through human-in-the-loop workflows. A moderation and review layer gives experts oversight, protects client confidentiality, and keeps responses grounded in approved internal content.
"The differentiator isn't the model, it's the system design, the data strategy, and the human expertise layered on top," said Tejas Totade, CTO and Head of AI at Ruder Finn. "We aren't the only agency doing this kind of work for clients, but we are just about the only one walking the technology, not just talking the technology."
The company has made human approval a permanent part of the system. That reflects the privacy and compliance requirements of communications data, where client work can be too sensitive for a fully automated response layer.
The roadmap includes an integrated CMS-to-vector pipeline, where approved content will feed both traditional web pages and the index that powers the retrieval layer. That means the site's conversational responses will stay current as new content is published.
A tech-company mindset inside an agency
The move mirrors a broader shift among enterprise teams adopting LLMs, particularly for AI for PR & Communications. For Ruder Finn, the bet is that its own site can demonstrate what AI systems can do for other clients.
Using Claude as the engine gives the system a strong language base, but the agency's custom layers handle the specifics: the knowledge graph, the moderation workflows, and the interfaces visitors actually use.
Why this matters for PR professionals
For comms teams, the notable shift is grounding-the idea that an AI assistant can answer questions with approved internal content managed through a system rather than left fully to a generic model. Ruder Finn has built guardrails around the technology, and it expects client work to push in a similar direction.
The other takeaway is practical. A public-facing AI tool that must protect client confidentiality and maintain factual accuracy faces the same challenges that internal knowledge systems will face in agencies and corporate affairs teams. Watching how Ruder Finn handles the balance between responsive answers and human review might offer useful lessons for anyone building similar tools. And the site's continued use of traditional pages suggests the transition from browser to chatbot won't happen all at once.
Your membership also unlocks:
