Install the skill and attach the example JSON to your assistant. Ask it to use sourcevine-audience-insights to analyze the supplied fictional evidence offline, with no URL fetching or MCP calls. Label the result a demo. Your assistant’s own usage charges may apply.
The fixture is marked illustrative: true and uses non-resolving example URLs. Its charge fields describe fictional cached readings, not spending during this exercise.
Workflow
How the skill works.
01
Connect your assistant
Set up Sourcevine MCP and install the skill in Codex or Claude Code.
02
Add the posts and your question
Provide post urls, research question or business context and a credit limit. The skill estimates the cost before reading.
03
Confirm, then answer
Check each suggested answer against your verified product facts. The skill never posts a reply.
Install this skill
Add Audience Questions & Objections to your assistant.
Choose your client and run its command on macOS or Linux. For live data, connect Sourcevine MCP once. The fictional examples work offline.
1. Confirm your account and save a key.Start setup →
After setup, paste this prompt into your assistant and replace the placeholders.
Use the Sourcevine audience insights skill on these public posts:
[paste up to three TikTok, Instagram or YouTube post URLs here]
Our research question: [what do you want to understand?]
Use at most 24 credits and one comment page per post.
Prioritize questions and objections, then draft answers with evidence IDs.
Mark unknown product facts as placeholders and propose a content test.
Save the HTML and Markdown brief and JSON evidence.
Limits
What the comment sample covers.
Themes describe the comments analyzed, not every viewer or customer.
Duplicate IDs and unusable text are accounted for. Theme counts link to distinct evidence IDs.
No commenter profiles, contact enrichment, demographics or invented purchase-intent scores.
Suggested answers require your verified product facts. The skill does not post replies.
3 posts · 1 page each · 24-credit estimated limit. Requests ask for up to 20 items per page.