For creators
Use instruction-load analytics to find better questions
Read the dashboard as evidence of retrieval, then investigate what users need from the workflow.
Quick answer
Start with the event the dashboard actually records: successful instruction retrieval through MCP. Compare loads by skill, distinct accounts, returning users, and daily activity to choose questions worth investigating. None of those metrics proves task completion, quality, retention, or financial return; combine them with specific feedback and reviewed examples.
Read each number at its actual scope
The creator dashboard shows total loads, unique users, returning users, loads by skill, and recent daily activity in UTC. Returning users means accounts that loaded the creator’s skills more than once. It is a creator-level count, so two different skills can contribute to the same account’s repeat activity.
The ledger counts a successful first load once. Continuation chunks belonging to that load do not add extra uses. A long skill should not appear more popular simply because it takes several responses to deliver. Historical purchases also do not manufacture load events for a period that was never measured.
Work through a small fictional dashboard
Suppose a hypothetical research skill shows 18 loads from 10 accounts. That supports a narrow observation: the workflow was retrieved 18 times by 10 distinct accounts in the measured data. It does not mean 18 research projects were finished, eight customers returned, or a particular percentage of people succeeded.
The distribution matters. One account could have made nine loads and the other nine accounts one each. Different distributions can produce the same totals. Use the dashboard’s actual returning-user figure for its stated definition, and do not invent a cohort retention rate from subtraction.
| Observation | Useful next question |
|---|---|
| Several loads after a new release | Did users understand the changed input requirements? |
| One skill draws most retrievals | Does its narrower job make the offer easier to recognize? |
| Few recorded loads | Is the intended audience finding and connecting the workflow? |
Turn a question into one controlled improvement
If feedback says users do not know which interview notes to supply, revise the input checklist. Test a normal note set and an incomplete one, then submit that change through review. Keep a dated record of the issue and the approved release so later feedback can be compared with a specific intervention.
Treat any change in loads as an observation, not proof that the edit caused it. Promotion, new buyers, the frequency of the underlying job, and troubleshooting can also change retrieval activity. Ask users about the concrete step that confused them instead of treating repeated loads as automatic satisfaction.
Keep sales and outcomes separate
Recorded net sales are after the platform fee and adjustments, before Whop processing fees. They are not bank payouts and do not measure the value a buyer obtained. Use provider balances for withdrawals and evidence of actual output quality for claims about results.
When sharing a figure, include its period and definition. If there is no activity yet, say so or leave the metric out. A truthful small sample is more useful for improving a workflow than a fabricated popularity number.
Common questions
Can I divide returning users by all users and call it retention?
Not without defining a cohort and return window. The dashboard’s returning-user count is repeat retrieval, not a ready-made retention study.
Does a failed access request count as a load?
A load records successful instruction retrieval. An access failure or a continuation chunk should not be treated as a new successful load.
Can analytics tell me whether an email was sent?
No. Skillset instruction-load analytics do not verify external actions. Check the relevant system and authorized evidence for that separate event.
Put a workflow to work.
Connect your library to your AI, or turn your method into a skill.