New to skills? Start with the AI Skills Library for how to add one.
Reads what people actually asked your Adoption Agent over a time window and turns it into a shareable report, with topics named the way a colleague would say them out loud rather than as raw conversation logs.
What it does
The report splits into three sections, each a ranked list of topics you can expand to read the real conversations underneath:
Couldn't answer—where the Agent said it wasn't able to find an answer.
Negative feedback—topics where people gave a thumbs-down, with their written feedback shown word for word.
Asked most often—by volume, with internal and test traffic kept visibly separate from real customers.
A topic can appear in more than one section, and often does—the thing people ask most is frequently the thing that fails most. When that happens it's badged so you can tell it's the same topic rather than two similar ones.
After the report you get a few paragraphs on what it means: which gaps are missing documentation versus something missing from the product, what the thumbs-downs are really about, and one or two things worth doing next.
What you'll need
The Userflow MCP connected
An Adoption Agent with conversations in the window you're asking about
How to use it
"What are people asking our Adoption Agent?"
"What is the Agent failing to answer?"
"Where is the Agent getting negative feedback?"
You'll get a picker listing your Adoption Agents so you can choose a specific one. That step exists because accounts tend to accumulate test copies with identical names, and picking by name alone could quietly analyse an abandoned draft. Time windows are 7, 30, or 90 days, or a range you set. The default is 30 days.
Things worth knowing
"Couldn't answer" is not an accuracy score. It counts the times the Agent said it couldn't find a match in its knowledge. An answer that was confidently wrong counts as answered, so it won't appear here. Read the section as a map of gaps, not as a measure of correctness.
Thin windows produce thin reports. Under about ten conversations, there isn't enough to find patterns in, and the skill will suggest a wider window before building anything.
People show as shortened internal IDs. There's currently no way to turn those into names or email addresses.
Grouping conversations into topics involves judgement. When a conversation could sit in two places, it goes where its opening question points, and the original wording stays visible so you can disagree.
What to do with it
The first section is usually the most useful. Anything the Agent couldn't answer is either missing from its sources or written in a way it can't match to the question—both fixable. See Adoption Agent settings and advanced customizations.