How Descript Keeps Its Help Center Accurate at a Company that Ships Daily

Before Solo, Stephanie's mornings looked like this: log in, open Slack, and work through somewhere between 60 and 100 unread messages across engineering channels to figure out what had changed in the product. On a busy day, that alone could take an hour and a half.

Even after she knew what changed, finding every place in the help center that needed updating was a tedious process. She'd search Zendesk for any term she could think of that might be related to the product change, trying to catch any article that would have been impacted.

"I would search every phrase I could possibly think of that might be a place that needed an update. We'd always find something we overlooked later. When you have an ecosystem of 400 articles that link to each other, it matters."

The margin for error was real. Because Descript's in-app AI and internal support tools pull from the help center as a source of truth, outdated documentation didn't just confuse customers—it could send the AI in the wrong direction too.

Before landing on Solo, Descript evaluated a few alternatives.

Notion AI was an early candidate. It's already embedded in how Descript's team works, and connecting it to Slack and GitHub seemed promising. But in practice, Notion's historical context worked against them. Because Notion holds years of project history—old features, deprecated flows, past naming conventions—the AI tended to conflate what exists now with what used to exist. Even when they tried to constrain the context, the hallucination rate was higher than they could accept for something customer-facing. In addition, Notion was missing context on Feature Flags and Linear, which meant it missed a lot of important context.

They also considered building something internally. The technical capability was there—they had access to capable AI models and the right data sources. But the conversation kept coming back to the same sticking point.

"Sure, we can build this internally—but a team of people is going to build the same tool better than one or two people. And then it always comes back to: who owns it? Who manages the upkeep? Who corrects the hallucinations?"

Building and maintaining a custom tool would have required ongoing ownership, which isn't a good use of their developers' time. They needed something that worked well out of the box and would keep getting better over time.

What set Solo apart was that it maintains documentation based on what's actually in the codebase—not just what engineers say or has been written down in a doc somewhere. Since the product's code is the source of truth on what has changed, she doesn't have to worry about lines of communication breaking down.

Two integrations sealed the decision: feature flags and Zendesk. Descript's engineers use feature flags extensively, so understanding what's live in the product at any given moment means understanding the flag state—not just the code.

"If we don't know about the feature flags, knowing what's in the codebase is almost useless."

The Zendesk integration meant Stephanie could go from spotting a change to updating the relevant article without ever leaving Solo. That closed the loop in a way no other tool had.

The Solo team itself was also a factor. Descript's use case has some specific nuances—feature flags, a large cross-linked help center, an AI that depends on the docs—and the Solo team engaged seriously with those details rather than treating them as edge cases.

"Their team seems to have a genuine, real interest in making Solo better for Descript, even with what feels like very nuanced or narrow scenarios that might only apply to our organization. That openness to feedback has been fantastic."

Now, Stephanie starts her day by opening Solo's release notes. Instead of parsing dozens of Slack threads, she gets a clean, readable summary of what changed in the product—with citations. Every change links out to either the feature flag controlling it or the GitHub pull request behind it, so she can verify things herself rather than taking anyone's word for it.

"I love that with those release notes, I see all of the changes in one place and can click directly into whatever docs were impacted."

When something does need updating, Solo surfaces every article that mentions the relevant term—no more searching Zendesk manually and hoping she caught everything. And because the Zendesk integration lets her make edits directly from Solo, the process that used to mean opening Zendesk, finding the article, and tracking down the right sentence is now a single click.

The improvement isn't just in time saved. It's in confidence. With a major product update coming, Stephanie described going into it feeling prepared rather than anxious—knowing that after the change goes live, Solo will help her catch anything that still needs a touch-up.

"My confidence that the help center is accurate is just at a different level now. Before, it was me trying to think of every possible search term. Now I know something is actually checking."

• Reduced time chasing product updates by 78%
• 30% reduction in questions sent to engineers
• Increased confidence that the help center is up-to-date