We kicked off this September series with a promise: a new theme every week, straight from everything we’ve shipped these past months. Last week was Heatmaps, now open to everyone. This week, we’re turning to AI — because seeing what’s happening is only half the job. The other half is understanding it fast enough to act.
Across Screeb, AI is increasingly doing that legwork for you: tagging what your agent conversations are actually about, telling human traffic apart from AI agent traffic, pinpointing exactly where your agent is letting users down — and now, letting you query all of it from the AI tools you already use. Here’s what’s new.
Agent Conversation Topic Tagging
📈 Spot the Recurring Themes in Your Agent Conversations
Every conversation your users have with your AI agent — on your website or in your mobile app — is automatically tagged by topic. No manual reading, no sampling: the full volume gets classified as it comes in.
You then see how each topic evolves over time. Is a subject coming up more than last period, or less? Did the change you shipped on your agent last week actually shift what people ask about? The trend answers that.
Benefits:
- Automatic topic tagging across 100% of agent conversations
- Volume and share for each topic, so you know what matters most
- Period-over-period evolution to catch emerging and fading subjects
- A direct feedback loop: measure how agent changes affect what users talk about
AI Agent Detection
🤖 Tell Humans and AI Agents Apart in Your Traffic
AI agents now browse and interact with websites and mobile apps just like real users — and their share of your traffic keeps growing. If you can’t tell them apart from humans, they end up in your survey responses, your session recordings, and every number you use to make decisions.
A dedicated dashboard shows you what share of your users are actually agents, and how that evolves. From there, every targeting rule in Screeb — surveys, product tours, session recording triggers — lets you decide whether to include or exclude agent traffic.
Benefits:
- See the real split between human and agent traffic in your product
- Keep agents out of your NPS, CSAT and analytics — or study them separately
- Exclude agents from survey, product tour and session recording targeting, rule by rule
- Filter and analyze your data by traffic type
Agent Problem Detection
🔍 See Exactly Where Your Agent Is Failing Your Users
We’ve rebuilt problem detection from the ground up. Screeb now surfaces three distinct kinds of breakdown in your agent conversations: answers that technically worked but frustrated the user, requests the agent simply couldn’t handle, and technical failures in how the conversation itself unfolded.
Each one is grouped by topic and, more importantly, tied to what it actually costs you. For every issue, you see its impact on agent retention — do people come back after hitting it? — and on post-conversation conversion. You’re no longer looking at a list of complaints, you’re looking at a ranked list of what to fix.
Benefits:
- Friction on answered questions — topics where the agent replied, but the user walked away unsatisfied
- Unsupported requests — the share of conversations where the agent couldn’t do what was asked
- Technical issues — problems in how the conversation ran, independent of its content
- Business impact on every issue — retention and post-conversation conversion, so you fix what matters first
Screeb MCP Server
🔌 The Screeb MCP Server Is Live
Screeb now has its own MCP server. Any tool that supports the Model Context Protocol — Claude, Cursor, and a fast-growing list of others — can connect to it and query your Screeb data directly.
That means asking questions about your surveys, responses, users and analytics from wherever you already work, in plain language, without exporting a CSV or building an integration. Your product feedback stops being locked in one interface and becomes something you can pull into any AI workflow.
Benefits:
- Ask questions about your Screeb data from the AI tool you already use every day
- No export, no API integration to build — connect once and query
- Bring product feedback into work you’re already doing elsewhere: specs, analysis, reporting
- Works with any MCP-compatible client, so it follows whatever tools you adopt next