Recipes

These prompts work in any client connected to the AmICited MCP server. Replace example.com with your domain. Under each one you will find what the agent does, so you can tell a good run from a shortcut.

A few habits make every recipe better:

• Name the domain. Most tools need a domain_id, and a workspace often tracks several domains. Naming it saves the agent a guess.
• Name the window. “Last 28 days against the 28 before” gives comparable numbers. See the report-windows skill.
• Ask for links. Agents can call report_link to give you a working link to the report behind a finding.

Reporting#

Weekly AI visibility briefing#

Give me a weekly AI visibility briefing for example.com: the last 7 days
against the 7 days before. Headline KPIs with changes, the prompts we
gained and lost, which engines moved, the competitors whose share of voice
grew, and the sites cited most often. Note any annotations in the period.
End with the three things worth doing next, each with a link to the report.

What the agent does: list_domains to resolve the domain, get_dashboard_metrics for the KPIs, deltas, platform breakdown and competitor share of voice, get_prompts_metrics for per-prompt wins and losses, list_top_cited_domains for the most cited sites, annotation_get_overlay for changes logged in the period, and report_link for the links. It should read the recurring-reports skill so next week’s briefing uses the same windows and tools.

Monthly competitor snapshot#

For example.com, compare our AI share of voice with each tracked competitor
over the last 30 days against the previous 30. Which competitor gained the
most, on which engines, and on which prompts are they named when we are not?

What the agent does: list_competitors, then get_dashboard_metrics for competitor share of voice (which covers every brand named, not only tracked ones), and get_prompts_metrics filtered by engine to find the prompts where you are missing. It follows the competitor-gap-analysis skill, which warns that raw mention counts can point at the wrong competitor.

Diagnosing#

Investigate a visibility drop#

Our AI visibility for example.com dropped this week. Use the
ai-visibility-audit skill to find out why. Check whether the prompt set or
engines changed, which engines ran but stopped citing us, and whether any
annotation lines up with the drop. Tell me what changed, not just the number.

What the agent does: reads skill://ai-visibility-audit/SKILL.md, compares get_dashboard_metrics across the two windows, then drills into get_prompt_detail for the prompts you lost, where each engine shows whether it ran and whether it cited you. It checks list_prompts for prompts added in the period (new prompts dilute the visibility score) and annotation_list for site changes. See why did my score drop? for the same checks by hand.

Can AI crawlers read our site?#

Check whether AI engines can read example.com: robots.txt rules for AI
crawlers, llms.txt, the accessibility tree and Common Crawl reach. Compare
our llms.txt with the domains that get cited for our prompts.

What the agent does: get_agent_accessibility for robots.txt, the accessibility tree and Common Crawl presence, llms_txt_get_latest for your llms.txt review, and llms_txt_get_comparison to compare against cited domains. It should present the llms.txt comparison as a correlation, not a proven cause.

Finding opportunities#

Find citation gaps#

For example.com, find the queries where we rank well in Google but AI
assistants never cite us, and the pages AI cites that rank poorly in
Google. Skip queries without enough runs to be reliable. Group the results
by intent and rank them by search volume.

What the agent does: seo_get_citation_gap_invisible_winners for good rank without citation, and seo_get_citation_gap_inverse for pages cited but ranking poorly (a promotion gap rather than a content gap). Both need Search Console data; if it is missing, the agent should say so and link Data Sources instead of reporting an empty result. See the citation gap report.

Which cited sites should we be on?#

List the domains AI engines cite most for example.com's prompts in the
last 30 days, excluding our own site and our competitors' sites. For the
top 10, tell me which prompts cite them and whether we are mentioned there.

What the agent does: list_top_cited_domains and get_sources_metrics for the most cited third-party sites, then get_source_detail for the prompts each one appears in. These are the review sites, directories and listicles worth getting onto. See Sources.

Taking action#

Write an article for a prompt we are missing#

Find the prompt for example.com with the highest search volume where we are
never mentioned and no article exists yet. Show me the fan-out questions no
page of ours answers, then generate a comparison article targeting that
prompt and those gaps. Tell me when the draft is ready.

What the agent does: get_prompts_metrics filtered to missing prompts without an article, sorted by volume, then get_prompt_coverage for the uncovered fan-out questions (prompt coverage). It calls article_generate through run_tool (this counts against your monthly article allowance and needs write access), then polls article_get until the status leaves “generating”. The draft appears under Articles. See AI articles.

Expand the prompt set from a page#

Suggest 15 buyer prompts for example.com based on https://example.com/pricing/,
skip any we already track, and show me the list before adding anything.
After I approve, add them tagged "pricing", daily, in the US.

What the agent does: reads the prompt-set-design skill, calls generate_prompts_from_url (nothing is saved), then lookup_prompts_bulk to drop prompts you already track. After your approval it calls create_prompts_bulk through run_tool with one tag, schedule and country. New prompts use credits on every run, so the agent should not add them without asking.

Log a change and measure it#

We rewrote https://example.com/features/ today. Log an annotation for that
URL and set an expectation that our AI visibility improves within 30 days.
In a month, ask me to check whether it worked.

What the agent does: annotation_get_metric_registry for valid metric keys, then annotation_create with the URL scope and a checkpoint. Later, annotation_get_progress shows the metric against its baseline, and the did-our-change-work skill covers reading the outcome. See annotations.

Related#