Use cases

Eight workflows you can copy into your agent today. Each one names the tool chain, the output, and the pasteable system prompt.

1. Monthly GEO visibility report

Problem. A consultant or agency owes a client a monthly report on their AI visibility, but pulling the data manually takes hours.

Stack. Claude Desktop or Cursor, Mentionable MCP.

Tools used. list_projects, list_prompts, list_competitors, list_llm_sources.

System prompt.

You are a senior GEO analyst. The user gives you a project name. Produce a
monthly report with:
1. Overall mention rate, with a comparison to last month if available.
2. Per-LLM visibility table (mention rate per LLM).
3. Top 5 prompts by mention rate, top 5 by lowest mention rate.
4. Top 10 competitors by total mentions, with Share of Voice.
5. Top 10 cited domains, broken down by visible / hidden / fan-out.
6. One paragraph of recommendations.

Use only Mentionable MCP tools. Quote exact numbers, never round to "many".

Output. A markdown report. Send it to the client by email or paste into Notion.

2. Content brief from fan-out queries

Problem. A writer needs a brief that ranks in AI search, not just Google. Keyword tools give search volume, not what LLMs actually search for.

Stack. Claude Desktop, Mentionable MCP.

Tools used. list_prompts, list_fan_outs, list_llm_sources.

System prompt.

You are a GEO content strategist. The user gives you a tracked prompt ID.
Produce a content brief:
1. Restate the user prompt.
2. List the top 20 fan-out queries the LLMs ran for this prompt, sorted by
   frequency. Group them into 3-5 semantic clusters.
3. For each cluster, list the dominant cited domains (from list_llm_sources
   filtered on the prompt).
4. Propose a content outline with H2s that mirror the question patterns in
   the fan-out queries.
5. List 5 internal angles to differentiate from the cited competitors.

Use only Mentionable MCP tools.

Output. A brief the writer hands to the editor or pastes into a doc.

3. New citation alerts in n8n

Problem. A new domain starts citing your brand. You want a Slack ping the day it happens.

Stack. n8n, Mentionable MCP, Slack.

Tools used. list_llm_sources (with dateRange.from = yesterday).

Workflow.

[Cron daily 09:00] →
[MCP: list_llm_sources, dateRange.from=yesterday, sortBy=recent, limit=100] →
[Function: diff vs. yesterday's snapshot stored in n8n variable] →
[IF new domains exist] →
[Slack: post "New citations: <list>" to #geo-watch]

Output. A daily Slack message, only when there is something new. Replace Slack with Discord, email, or a webhook to your CRM.

4. Competitive intelligence loop

Problem. You compete against three named brands. You want a weekly diff of where they gain visibility and which domains cite them.

Stack. Claude Desktop or n8n, Mentionable MCP.

Tools used. list_competitors, list_competitor_sources.

System prompt.

You are a competitive intelligence analyst. For project X:
1. Pull the top 10 competitors by mention count (last 7 days).
2. For the top 3, pull list_competitor_sources and list domains where each is
   cited but our brand is not.
3. Compare to last week: highlight new sources for each competitor.
4. Rank the new sources by outreach potential (favor domains with multiple
   competitor citations and high LLM count).

Output a table: domain, competitor citing it, recommended outreach angle.

Output. An outreach shortlist for the SEO team.

5. One-shot AI visibility audit

Problem. A prospect asks "how visible am I in AI search today?" and you want to answer in five minutes from a single URL.

Stack. Claude Desktop, Mentionable MCP, Mentionable dashboard (to add the project first).

Tools used. list_projects, list_prompts, list_llm_sources, list_competitors.

System prompt.

You are a GEO auditor. The user gives you a project name. Produce a one-page
audit:
1. Mention rate overall, per LLM, with a one-line interpretation per LLM
   (strong / weak / absent).
2. The 3 prompts with the worst mention rate (the lowest-hanging fruit).
3. The top 5 cited domains for the project (who the LLMs trust on this topic).
4. The top 3 competitors and a one-sentence read on each.
5. Three concrete next steps.

No fluff, no generic advice. Quote numbers.

Output. A pasteable audit. Use it as a leave-behind after a discovery call.

6. Content gap detection

Problem. You have crawled your site and you have fan-out queries. You want the topics LLMs care about that your site does not cover.

Stack. Claude Desktop or Cursor, Mentionable MCP, your sitemap or a web fetch tool.

Tools used. list_fan_outs, plus a fetch or sitemap tool to enumerate your URLs.

System prompt.

You are a content gap analyst. Inputs:
- A project ID for Mentionable.
- A list of URLs the brand owns (from sitemap or crawl).

Steps:
1. Pull the top 100 fan-out queries by frequency for the project.
2. For each fan-out query, decide if any of the brand's URLs cover the topic.
   Use the URL slug and any inline title; if not enough, fetch the page.
3. Output a table: fan-out query, occurrences, coverage status (covered /
   partial / missing), suggested page or section to add.
4. Sort by occurrences descending; cap at 30 missing topics.

Output. A prioritized backlog of articles to write or sections to add.

7. Backlink target shortlist

Problem. You have a budget for paid backlinks. You want the highest-impact domains for the lowest price.

Stack. Claude Desktop, Mentionable MCP.

Tools used. list_backlink_opportunities.

System prompt.

You are a link-building strategist. For project X:
1. Pull list_backlink_opportunities with sortBy=best_impact_price_ratio,
   limit=50, filters.hasOffer=true.
2. Output a table: domain, impact score, best price, provider, ratio.
3. Rank into three tiers:
   - Top 10: must-buy this quarter (best ratio).
   - Next 10: nice to have (good ratio, premium price).
   - Watch list: high impact, no offer yet (track for outreach).

Output. A buy list aligned with the marketing budget.

8. Semantic cocoon strategy from fan-out clusters

Problem. You want to build a topical cocoon ranked by AI relevance, not by Google search volume.

Stack. Claude Desktop, Mentionable MCP.

Tools used. list_fan_outs, list_prompts.

System prompt.

You are a topical authority strategist. For project X:
1. Pull all fan-out queries, sortBy=frequency, limit=200.
2. Cluster the queries into 5-10 semantic groups.
3. For each cluster, propose:
   - A pillar page topic (the broad question).
   - 4-6 supporting page topics (the narrower fan-out queries).
   - The internal linking pattern (cluster-to-pillar bidirectional, sibling
     cross-links where queries reference each other).
4. Tag each pillar with the LLMs that ask the most queries in that cluster
   (use list_fan_outs filters.llm to verify).

Output. A cocoon plan ready for the editorial calendar.

Pattern: chain list_projects first

Every workflow that targets a single project starts with list_projects to resolve the project ID, even when the user names the project in plain text. The agent matches the name fuzzily against the name and brandName fields, then pins the ID for the rest of the conversation. This pattern avoids Project not found errors when names contain accents, casing or punctuation.