What you are looking for
Some questions will always come back unanswered — a topic nobody wrote up, a detail that changes too often for an article, a way of phrasing something your content never uses. Those are knowledge gaps, and they are the cheapest improvements available to you: each one you close removes a whole category of unanswered questions, not just the one that was asked.
Where to look
Two screens, both under your brand:
- Brand → Vex → Reports — the figures, plus a browsable list of every individual answer the agent has given, with filters and per-answer detail. If you remember a separate "AI Responses" screen, this is where it went; it is now the bottom half of this one page.
- Brand → Vex → Suggestions — the pattern analysis, where the AI reads through the failures and tells you what keeps coming up.
Reading the figures on Reports
Pick a date range and you get three cards:
- Successful responses — of the real questions asked, the share the AI handled: answered from your content, connected a person who was asked for, or ran a workflow.
- Neutral responses — messages that asked nothing at all: greetings, thanks, acknowledgements. These are shown against all responses and deliberately kept out of the success and failure rates.
- Failed responses — real questions the AI could not answer: missing knowledge, a handoff it decided on itself, or a generation error.
Below them is a table of AI-handled conversations — the ones that needed no human at all — and below that, the full list of answers.
Worth knowing if you have written the number down before: the success rate is calculated differently now. Small talk used to count against you, so a friendly bot that got a lot of "hi" and "thanks" looked worse than it was. It is counted separately today, which means the figure reads noticeably higher than the one you may remember. Nothing about the AI changed — only the arithmetic.
Getting the pattern analysis
Open Brand → Vex → Suggestions and press Generate suggestions. It reads up to the last 500 unanswered or escalated questions since your previous analysis, groups them by theme and writes up the ones that recur most. It needs at least 10 new ones since last time, so there is genuinely something to find. Questions where the visitor simply asked for a person are left out on purpose — that is routing, not a gap in what you know.
Each past analysis stays on the page with the date it was produced, so you can see whether a gap you closed has actually stopped coming up.
What to do with a suggestion
- A missing topic → write the article and publish it. Publishing indexes it immediately — no chatbot to sync, no widget involved, nothing to re-import. It is answerable from the moment you press publish.
- A topic that exists but is too thin → deepen the article. These show up as questions where Vex cited the right page and the visitor still escalated.
- A question about live data → add an API tool. Order status, stock, an account balance: no article can stay accurate. API tools live on Brand → Knowledge → API tools, and the description you write is what the AI reads to decide when to call the tool, so describe the situation rather than the endpoint.
Then leave it a week or two and run the analysis again. A category that disappears is a category you fixed.
The short version on Brand → Overview
You do not have to go looking. The brand's Overview page carries a Couldn't answer card listing the questions from the last 30 days that nothing on your site answered, with how many times each was asked, and a Vex card showing how the last 30 days split between answered, unanswered and small talk. It is the fastest way to notice a new gap opening up.
This history belongs to the agent
Conversation starters, suggestions and report history all belong to the AI agent, so deleting the agent removes them. Your knowledge is untouched — knowledge bases, documentation, files, website pages, question-and-answer entries and products belong to the brand and stay exactly as they are.
Next step
Once you have written new articles, check that Vex is pointing people at them properly: Vex citations — how it points users to source articles.