No one can deny that the AI revolution has moved at breakneck speed. Still, there are dozens of horror story that prove that success depends more on thoughtful integration than speed.
A low barrier to entry is incorporating AI chatbots on service-focused sites to streamline the user experience and reduce search time and friction when our end users are seeking information. A process GW began in 2025 with Cloudforce’s NebulaOne AI agent tool.
I was brought into the project to provide client strategy, brand guidance and product configuration.
The problem
While there is a big appetite across GW for AI chatbot integrations, we quickly learned that our older sites, in particular, require extensive content audits to review every published page for timeliness and accuracy.
Additionally, configuring the tool to “stay on course” and not provide incorrect information or inappropriate language was just as critical as ensuring it met brand and accessibility requirements. While it was humorous to see it provide cake recipes in the tone of characters from HBO’s The Sopranos, it was also wildly inappropriate.
Finally, we realized that quite a bit of stakeholder education needed to happen to help them understand how their content will guide the chat conversation, but that it could not dictate it like a script.
The solution
There was quite a bit of trial-and-error in properly configuring the instructions and default prompts for our pilot chatbot. We also had to navigate design hurdles to ensure accessibility with limited admin tools for branding. Eventually, I was able to overcome these hurdles by configuring accessibility guidelines into the instructions.
I also worked with the Drupal team to create on-demand content audit exports of all published Drupal website content, so that stakeholders could review what information the AI agent might view as a reliable source when responding to prompts. After all, as a colleague pointed out, “Garbage in, garbage out.”
Finally, I worked with our Digital Support Strategist to create live and on-demand training resources – both evergreen and client-specific – to education our stakeholders on how AI works, what to expect, and where there may be limitations.
The result
Our pilot chatbot launched in early 2025. User adoption was slow, but steady, and it now sees ??? monthly usage. We also launched about two dozen public and provide AI agents across GW websites to meet a variety of needs.
Our goal remains the same, quality over quantity.
We’ve launched agents serving as a resource for frontline customer service workers who need fast access to information, to guiding undergraduate students through questions about housing, dining, financial aid, and commencement.
Every request is treated like a standalone, unique need. We draw on what we’ve learned and what we know works, but we customize and tweak configurations and instructions to make the tool work for that particular stakeholder’s needs.
As Cloudforce interates and improves the NebulaOne platform, we to expand on how we can improve the GW Chat experience for everyone.