Recipes

Recipes are short walkthroughs — a question you ask your assistant, the tools it will reach for, and what to look for in the answer. Use them as templates: copy the prompt, swap the brand name, and run it.

1. Weekly visibility check

"How has my AI search visibility moved over the last 7 days?"

Tools called

1. list_domains — pick the domain.
2. get_visibility_time_series (days=7).

What to look for

• A clear up/down delta day over day.
• If visibility is flat near 0, your prompts aren't matching the surfaces — check Prompts coverage.
• If it spikes then drops, ask the assistant to follow up with get_mentions_time_series to see if the underlying mention count moved.

2. Who's eating my share of voice?

"Compare our share of voice against our top 3 competitors over the last 30 days."

Tools called

1. list_domains.
2. get_brand_rankings (days=30).

What to look for

• The row with is_brand: true is yours; sort by sov to rank.
• A competitor with high mentions but mid-tier visibility is winning volume from a few concentrated prompts — ask the assistant to confirm with list_prompts and a targeted get_citations lookup.

3. Where are LLMs sending citations?

"Show me the top URLs that LLMs cite when they mention us — sorted by mentions."

Tools called

1. list_domains.
2. get_citations (days=30).

What to look for

• The output is sorted by mentions desc — first row is your most-cited URL.
• page_type tells you the kind of page (Listicle, Review, Comparison…). A heavy Listicle skew means LLMs are pulling you from third-party round-ups; build for more first-party citations.
• models per row shows which channels cite each URL — a URL only cited by ChatGPT but not Claude is an opportunity for the other channel.

4. Which model surfaces me the most?

"Which AI channel mentions our brand the most, and how do positions compare?"

Tools called

1. list_domains.
2. get_model_leaderboard (days=30).

What to look for

• Sorted by visibility desc. The top row is your strongest channel.
• A row with high visibility but high avg_position (4+) means you're showing up but late in the answer — citation-worthy but not the lede.

5. Citability vs raw citations

"What percentage of our brand mentions actually come with a citation back to our site?"

Tools called

1. list_domains.
2. get_citations_time_series (days=30).
3. get_citability_time_series (days=30).

What to look for

• Citability ratio = citations / mentions. A high mention count with low citability means LLMs talk about you but don't link out — typical for unowned mentions on third-party Listicles.
• Watch the trend, not the level. A falling citability ratio while mentions hold steady means content discovery is happening off-page.

6. Debug: "Why did Claude say X about us?"

"What fan-out queries did the LLMs ask when probing our brand prompts last week?"

Tools called

1. list_domains.
2. list_fanout_queries (days=7).

What to look for

• Fan-out queries are what the LLM actually searched for to answer the prompt. A surprising fan-out (e.g. "<brand> lawsuit" when your prompt was "best CRM") explains a strange answer.
• Compare to your prompt list — if fan-outs drift far from your tracked prompts, your prompt set probably needs broadening.

7. End-to-end brand health check

"Give me a one-paragraph brand health summary: visibility trend, top competitor, top cited URL, and best channel."

Tools called (parallel)

1. list_domains.
2. get_visibility_time_series (days=30).
3. get_brand_rankings (days=30).
4. get_citations (days=30).
5. get_model_leaderboard (days=30).

What to look for

• The assistant should fan out then synthesize. If it serializes the calls instead of parallelising, you'll feel the latency — re-prompt with "call those in parallel" on capable clients.
• This is the canonical Monday-morning recipe. Save it as a shortcut in your client.

8. New prompts to add

"Look at our prompts and the fan-out queries from the last 30 days — suggest 5 prompts we're missing."

Tools called

1. list_domains.
2. list_prompts (paginate until done).
3. list_fanout_queries (days=30).

What to look for

• Compare fan-out queries against tracked prompts. Fan-outs that show up but aren't on your prompt list are candidates.
• This isn't a guarantee of monitorable prompts — review before adding.

Chaining tips

• Always start with```. Your key may scope multiple domains; passing the wrong domain_id returns a 403.
• Reuse the ********``` within a session — the assistant doesn't need to refetch the list every turn.
• Default the window to 30 days unless you have a reason, then pass the window you actually mean — any value 1–365 is honoured exactly. (Before 14 September 2026 the window was rounded up to the next bucket in 1, 7, 14, 30, 90, 180, 270, 365 — asking for 21 days answered for 30; that no longer happens.)
• Combine charts and tables. Recipes 1 + 2 read better together than alone — visibility trend without share-of-voice context can be misleading.

Stuck on a chain? See Troubleshooting.