3RD

View saved version →

---
name: citation-sources
description: Show which websites and pages AI assistants cite when they answer questions a brand monitors in 3RD, and how much of that is the brand's own site. Use when the user asks which sources, domains or pages the assistants rely on or cite.
---

Use this skill for questions about the sources behind AI answers.

1. Call `list_brands` and take the `brandId`. Ask which brand if there are several and the user named none.
2. Call `get_citation_sources` with the `brandId`. Useful arguments:
   - `fromDate` / `toDate` (YYYY-MM-DD) for a period; the default is the last 30 days;
   - `kind` to keep only `owned` or `earned` sources;
   - `domain` to keep one website;
   - `promptType` to keep one prompt category (must-win, core, foundation);
   - `limit` and `page` to walk a long list. The result says how many pages there are.
3. For the brand's own share by assistant, call `get_attribution_share` for the same period.
4. Answer with the top sources: URL, citation count, share, owned or earned, and whether the page mentions the brand. Say which period and which filters the list reflects.

Report counts and shares exactly as returned. A source being cited is not the same as it recommending the brand; only say a page mentions the brand when the result says so.

View saved version →

---
name: competitor-trend
description: Show how a brand's AI visibility has developed over time and how it ranks against the competitors it tracks in 3RD. Use when the user asks about trends, development, momentum, or who is ahead of them.
---

Use this skill for questions about change over time and position against competitors.

1. Call `list_brands` and take the `brandId`. Ask which brand if there are several and the user named none.
2. Call `get_visibility_trend` with `days` set to the period the user asked for (default 30). It returns one visibility score per day.
3. Call `get_competitor_ranking` with the `brandId`. Pass `fromDate` and `toDate` (YYYY-MM-DD) for the same period as the trend; without them it covers the last 30 days.
4. Answer with:
   - how the series moved: the first and last value, the highest and lowest day, and any sharp change;
   - the brand's rank and score, and the competitors ranked above and just below it;
   - the period each part covers.

Report scores exactly as returned. Only compare against competitors that `get_competitor_ranking` returns; to see or change the tracked list, the user can call `list_competitors` or use the 3RD app. Do not explain a movement unless a tool result supports the explanation.

View saved version →

---
name: content-gaps
description: Find the questions a brand lacks content for, using 3RD's content recommendations, and what kind of page would close each gap. Use when the user asks what content to write, which questions they are missing, or where their pages fall short.
---

Use this skill to find content gaps.

1. Call `list_brands` and take the `brandId`. Ask which brand if there are several and the user named none.
2. Call `get_content_gaps` with the `brandId`. Pass `topicId` to look at one topic, and `limit` to shorten the list. The list is ordered by opportunity, highest first, as on the Content page in the 3RD app.
3. For each prompt you report, give:
   - the question;
   - how well the brand's own pages cover it (coverage % and matching pages);
   - the recommended action: optimize an existing page, create a new one, or defend;
   - the recommended content type and target URL when there is one.

Coverage describes the brand's pages as they are now; it has no date range, so do not attach another tool's period to it. If the user wants a brief for one of the prompts, `brief_content_gap` writes one with AI generation and saves it for the team, so confirm with the user before calling it.

View saved version →

---
name: next-actions
description: Tell a brand what to do next to improve its AI visibility, from the plan in 3RD and the recommended tasks not yet in it. Use when the user asks what to do next, what to prioritize, or what is in their plan.
---

Use this skill to answer "what should we do?" from the brand's own plan.

1. Call `list_brands` and take the `brandId`. Ask which brand if there are several and the user named none.
2. Call `get_playbook` with the `brandId`. These are the tasks already in the plan, with a full brief and checklist each.
3. Call `list_suggested_tasks` with the `brandId`. These are recommended tasks that are not in the plan yet.
4. Optionally call `get_opportunity_radar` to show which of the five drivers (content coverage, content quality, crawlability, authority, outreach) is weakest. A driver with `dataAvailable: false` is not measured; never call it a weakness.
5. Answer in two clearly separate parts:
   - **In your plan:** the open tasks, most relevant first, each with its link (`url`).
   - **Suggested next:** the recommended tasks, presented as suggestions the team can add, not as planned work.

`expectedLift` is a model estimate, not a measured effect; say so. When it is null, the effect is unknown, not zero. Do not add tasks or change a task's status unless the user asks; `update_task_status` changes the plan for the whole team.

View saved version →

---
name: visibility-status
description: Report how visible a brand currently is in AI assistants' answers, using 3RD. Use when the user asks how visible they are, what their AI visibility score is, or how it changed against the previous period.
---

Use this skill to answer "how visible are we?" from 3RD's measurements.

1. Call `list_brands` and take the `brandId`. If the account has several brands and the user did not name one, ask which brand they mean. Never guess a brand from a name or a URL.
2. Call `get_visibility_breakdown` with the `brandId`. It returns the overall visibility score and the scores per prompt type (must-win, core, foundation).
3. Call `get_highlights` with the same `brandId` for sentiment, citation share and the direction against the previous period.
4. Answer with:
   - the overall score, exactly as returned;
   - the must-win, core and foundation scores;
   - the direction against the previous period, only when `status` is up, down or neutral (when it is unknown, say there is nothing to compare with);
   - the period the numbers cover. Without dates both tools report the last 30 days; say so.

Do not average, round differently or recompute any score. A score of 0 with no measured prompts means "not measured yet", not "invisible". If the user names a period, pass it as `fromDate` and `toDate` (YYYY-MM-DD) to both tools.