--- id: creatorland-playbook-09 version: 1.0.0 title: Turn a call transcript into a creator shortlist category: casting plan: free skills_pack_min_version: "0.1.0" tested_clients: - claude-web - claude-desktop requires_local_filesystem: false requires_scheduling: false can_create_side_effects: false --- # Turn a call transcript into a creator shortlist Turn a client call, kickoff transcript, meeting summary, or voice-note dump into a normalized campaign brief and a ranked, client-ready creator shortlist. ## Before you begin 1. Check whether the Creatorland MCP tools and the `transcript-to-shortlist` skill are already available. If they are, do not reinstall them. 2. If the Creatorland MCP is not connected, direct the user to https://mcp.creatorland.com/skills and help them connect it in their current AI client. 3. If the client supports the Creatorland skills plugin but the skill is missing, install the current Creatorland skills package using the supported method for that client. 4. Confirm Creatorland authentication before making a paid tool call. The first tool call may open a one-time OAuth sign-in. 5. If the client cannot install companion skills, follow the workflow below directly with the available Creatorland MCP tools. ## What to ask for Ask the user for: - The transcript, meeting notes, or an accessible transcript link. - Desired shortlist size. Default to 8 when they do not specify one. Do not ask the user to re-enter information already present in the transcript. Ask one batched set of no more than four questions only when missing information would materially change the search. Prioritize: - Required platform - Priority geography or market - Creator size or follower tier - Budget posture Also ask for competitor brands only when a conflict review is needed and the transcript does not name them. ## Run this workflow ### 1. Read and normalize the call Use the `transcript-to-shortlist` skill when it is available. Extract: - Brand or client - Product or campaign - Objective - Target audience - Geography and language - Required platforms and content formats - Creator category, style, and audience - Deliverables - Compensation model and budget - Timing - Usage rights and exclusivity - Brand-safety requirements - Hard requirements - Preferences - Exclusions - Competitor brands - Creators named as examples or lookalike seeds - Open questions and assumptions Label each extracted item as `stated`, `inferred`, or `missing`. ### 2. Confirm the working brief Show a compact section titled `What I heard on the call`. - Quote back the important stated constraints. - Keep inferred requirements visibly labeled. - Ask for confirmation before spending credits. - If the user has already told you to proceed and there is no material ambiguity, state the assumptions and continue. Never silently relax a hard requirement. ### 3. Estimate the run Give the user a concise estimate of the expected Creatorland credit use before a high-cost profile fan-out or whenever the expected run exceeds roughly 30 credits. Offer a lower-cost version when practical: - `Thrifty`: smaller candidate pool, up to two lookalike seeds, and fewer profile calls. - `Thorough`: larger candidate pool, all useful named seeds, and fuller profile validation. ### 4. Build the candidate pool Use tight-precision brief search for an actionable outreach or client slate. Preserve useful language from the call because aesthetic and tone words improve fit rationale. - Search for 2 to 3 times the requested shortlist size. - Apply explicit platform, niche, geography, tier, audience, and freshness filters when the transcript supports them. - Do not broaden a failed search without explaining which constraint is limiting results and asking permission. For each creator named on the call as an example, run a lookalike search when a resolvable handle or profile is available. - Thorough mode: use each useful named seed. - Thrifty mode: use at most the two most-emphasized seeds. - Dedupe the combined result set before profile enrichment. - If a named creator cannot be resolved, include that in the caveats. ### 5. Validate and rank Follow the current `brief-to-shortlist` workflow for the combined candidate pool: 1. Enrich the strongest candidates with available Creatorland profile data. 2. Apply the freshness gate and flag aging or stale information. 3. Run conflict checks using competitor brands named in the transcript or confirmed by the user. 4. If no competitors were supplied, skip the conflict check and say so. 5. Add one aggregate market-rate benchmark when the campaign category, deliverables, or budget supports a meaningful comparison. 6. Rank candidates against the confirmed brief. Do not present a category-level rate band as a creator's personal quote. ## Deliverable Return a section titled `What I heard on the call`, followed by a ranked table: | Rank | Creator | Platform | Fit for this brief | Relevant signals | Risk or missing data | Rate context | Recommended next action | |---|---|---|---|---|---|---|---| For each finalist include: - Creator name, handle, and Creatorland profile URL when available - Primary platform - Concise rationale tied to the confirmed brief - Relevant audience, performance, brand-affiliation, or freshness signals - Conflict, data-quality, or missing-data flags - Aggregate rate context when available - Recommended next action When a finalist came from a creator named on the call, add: `Entered the slate as a lookalike of @, who was named on the call.` After the table include: - `Open questions and assumptions` - `Re-verify before outreach` - `Data freshness and provenance` - `Credits used` ## Privacy and honesty rules - Do not expose private contact information. - Use `available for Creatorland outreach` when that is the accurate next step. - Do not imply that a named seed creator endorsed a result. - Do not turn an inferred requirement into a client-stated fact. - Distinguish missing data from a negative finding. - Do not promise that any creator will accept a campaign. - Do not initiate outreach from this playbook. ## Empty and partial results - If the transcript lacks enough information, ask the smallest useful set of questions. - If no strong matches exist, explain the binding constraint and propose one change at a time. - If Creatorland lacks fresh data for a candidate, keep the creator only when otherwise useful and clearly flag re-verification. - If a tool refuses or times out, report the failure plainly and preserve the confirmed brief so the user can retry without repeating intake.