Skill instructions
aeo-visibility-report2.25 KB
View saved version →
---
name: aeo-visibility-report
description: "Build an AI search visibility report for a brand: how often it appears in ChatGPT, Perplexity, Gemini, Claude, Google AI and other AI answers, how it compares to competitors, which sources cite it, and what to do next. Use when the user asks about AI visibility, brand mentions in AI answers, share of voice, competitor comparison, citation analysis, or a weekly or monthly report. Not for website technical health (use site-audit-triage) or for writing content (use content-gap-to-article)."
---
# AI visibility report
## Steps
1. **Find the project.** `get_context` lists the projects you can reach. Use the only one, or the one the user named; if there are several and they didn't say, ask.
2. **Read the brief.** `get_brief` with `dateRange` 7d for a weekly report, 30d for a monthly one. One call returns visibility with its `sampleSize`, mentions, citations, each AI platform, the competitor ranking, the prompts that moved, the most cited domains, and prompts added or paused in the period.
3. **See what changed.** `get_changes` with `since` set to the start of the period: prompts that moved, competitors that gained, citations won and lost.
4. **Go deeper only where something moved.** `query_visibility` with the ids the brief gave: `groupBy: ["date"]` for when a drop happened, `["prompt", "competitor"]` for where a competitor wins, the `cited_urls` preset for which pages get cited. Use `query_sql` for anything else, after `explain: true` for the schema.
5. **Pick the next steps.** `get_recommendations` gives one ranked list; take the top two or three.
## Writing the report
- Lead with visibility and its change, then the one change that matters most.
- Then platforms (where the brand is strong and weak), competitors (who leads and by how much), citations (which sites AI trusts where the brand is missing), and the next steps.
- Give numbers as returned. When a platform has `thinSample`, or a visibility range (`visibilityLow` to `visibilityHigh`) is wide, say the number rests on few answers instead of calling it a trend.
- Before explaining a drop, check the brief's tracking section: prompts added or paused move the numbers on their own.
- If the period has no data, say so and suggest checking that prompts are being tracked.
content-gap-to-article3.43 KB
View saved version →
---
name: content-gap-to-article
description: "Find the prompts where competitors get mentioned in AI answers and the brand doesn't, turn the best into article ideas, then write, edit and publish them to WordPress, Webflow, Ghost, Framer, HubSpot, Sanity, Wix or the ChatFeatured CMS. Use when the user wants content that improves AI visibility, asks what to write next, wants to fill citation gaps, or wants to write, edit or publish an article. Not for analytics-only questions (use aeo-visibility-report)."
---
# Content gap to published article
## Find the gaps
1. **Find the project.** `get_context`.
2. **Find the gaps.** `query_visibility` for 30 days, `groupBy: ["prompt"]`, then `["prompt", "competitor"]`. A gap is a prompt where the brand has 0% visibility and competitors are mentioned often. `get_recommendations` lists the content gaps ChatFeatured has already found; use those first.
3. **Check what's already covered.** `articles_read` (operation `articles.list`) for published and in-progress articles, so you don't propose a duplicate.
4. **See what wins.** For each top gap, `get_answers` with its `promptId` lists the answers; `get_answers` with an `answerId` shows one answer with the pages it cites. Read two or three. The idea should beat those pages.
5. **Propose ideas.** For the top three: a title, the angle, the prompt it targets, and the pages it has to beat. Show them to the user.
6. **Save the ones they pick** with `save_articles` as ideas. Don't set `write` unless the user asks to start writing: writing uses the plan's monthly article allowance.
## Write, edit and publish
1. **Start writing.** `write_articles` with the idea ids. Each article uses one from the plan's monthly allowance, so say how many it will use and get a yes first. It returns a job per article.
2. **Wait.** `wait_for_job` with the job ids; each call waits up to 50 seconds, so call it again while they run. Writing takes 2 to 5 minutes. If the user doesn't want to wait, tell them to come back later; the drafts stay in review.
3. **Review.** `get_article` returns the outline; read only the sections you need with `sectionId`. Summarize or quote it for the user.
4. **Edit.** `edit_article` with the `revision` you read: replace exact text, insert, or rewrite whole sections. Leave the rest alone. If it answers `revision_conflict`, someone changed the article; its details have the changed sections, so re-target your edit and send the new revision.
5. **Images.** `generate_images` makes images into the project's library; follow the job with `wait_for_job`.
6. **Publish, only when the user confirms.** `integrations_read` (operation `integrations.list`) shows the connected sites; if there are several, ask which. `publish_article` publishes now, or later with `publishAt`. If they published it themselves, `publish_article` can record the URL instead.
## Boundaries
- Ask before saving, and never start writing without the user saying so.
- Gaps on a platform with a thin sample may be noise; prefer gaps that show up on several platforms.
- Writing spends the allowance and publishing changes their public site: both need the user's explicit yes in this conversation.
- A published article is locked; the error says to reopen it before editing. Reopening keeps the live copy up until it's published again.
- If no site is connected, finish the draft and tell the user to connect their CMS under **Integrations** in the ChatFeatured dashboard; the credentials go in there, not in the chat.
site-audit-triage1.54 KB
View saved version →
---
name: site-audit-triage
description: "Turn a brand website's audit into a short, ordered list of fixes that help AI assistants crawl, understand and cite it. Use when the user asks to audit their site, asks why AI answers don't cite their pages, or mentions llms.txt, structured data, robots or thin content. Not for visibility analytics (use aeo-visibility-report) or for writing articles (use content-gap-to-article)."
---
# Site audit triage
## Steps
1. **Find the project.** `get_context`.
2. **Overview.** `website_read`: operation `sites.list` for the site, then `issues.list` for open issues and `pages.list` for page scores (lowest first).
3. **Rank the fixes.** Site-wide problems first (llms.txt, robots, structured data), then the lowest-scoring pages that AI answers cite or that target tracked prompts. `get_recommendations` already ranks audit fixes alongside other work.
4. **Explain each fix** in a sentence: what's wrong, on which pages, and what to change. Quote the check's details rather than guessing.
5. **Check the fixes.** For site-wide fixes (robots.txt, sitemap, llms.txt, HTTPS), `website_write` (operation `siteAudits.start`) re-runs the site checks and is free. For page fixes, `website_write` (operation `pageAudits.start`) re-audits those pages at one page audit credit each; say how many credits before starting.
## Boundaries
- Report what the audit found; don't invent issues it didn't flag.
- If the site hasn't been audited, say so and offer to start a site audit (free), then page audits for the pages that matter (one credit each).