Read the Citation Flow
The Citation Flow is a Sankey diagram that traces, step by step, the path each LLM bot (ChatGPT, Gemini, Perplexity…) takes to build its answer. Read it from left to right: from the model queried all the way to the sources it ends up citing.
1. The top-line indicators
Above the diagram, a row of headline metrics summarizes the period:
| Indicator | What it means |
|---|---|
| Web Search Rate | Share of prompts where the bot performed a web search (vs answering from its internal index alone). |
| Mention (Web Search) | When a web search happens, how often your brand is cited in the answer. |
| Mention (No Web Search) | When the bot answers from its internal knowledge only, how often your brand is cited. |
| Owned Citations | Number of citations pointing to your own pages. |
Compare the two Mention rates side by side: the gap tells you how much of your AI visibility depends on the live web vs. the model's training memory.
2. Reading the diagram, column by column
The Sankey flows through four stages:
Stage 1 — Models (left)
Each blue block is one bot queried, sized by the number of prompts played against it (e.g. Gemini, ChatGPT, Perplexity Interface, Perplexity Sonar).
Stage 2 — Web Search or internal index?
The flow splits in two:
- Web Search & Query Fan Out — the bot queries the web in real time.
- No Web Search — the bot answers from its internal knowledge only.
Stage 3 — Mentioned or Not Mentioned
In each branch, the flow splits again: whether your brand was Mentioned in the answer, or Not Mentioned.
Stage 4 — Owned or External sources
On the right, you see where the sources consulted by the bot came from:
- Owned — your own pages (your root domain).
- External — third-party sites (comparators, brokers, media, etc.).
Key point: sources exist on the "Not Mentioned" side too. The bot read pages — sometimes yours — without citing the brand in its answer. That's exactly where optimization opportunities hide.
3. Source detail (click through)
Owned Sources — your own URLs
The list of your scraped URLs, ranked by citation count. The Mentioned on page column tells you whether the brand is actually cited on that page: a dash ("—") flags an owned page that doesn't mention you. Those are the first pages to rewrite.
External Sources — third-party sites
The most-cited third-party domains. The colored pills next to each domain reveal which brands also appear on those pages — a quick way to spot competitors that already own the conversation on that source.
4. Three questions to ask
- Do our own articles mention us? In Owned Sources, find pages cited by the bots that don't mention the brand → content to rewrite to anchor the brand.
- When we're not mentioned, which sources are used? Follow the "Not Mentioned" branch → External Sources → that's your target list for PR, partnerships, and outreach.
- Are competitors doing content worth learning from? The "Mentioned on page" column highlights competitor brands present where you're absent → editorial inspiration.
5. Prompt-level granularity
Every detail view has a Prompt tab (e.g. "320 results · 5 prompts") that lets you drill down to a single query: were you mentioned, and which sources — internal or external — were used to answer. Use it to prioritize optimizations on the queries that matter most to your business.
Key takeaways
- The Citation Flow reads left-to-right: Models → Web Search? → Mentioned? → Owned vs External sources.
- Sources on the "Not Mentioned" side are the most actionable — the bot read those pages but didn't cite you.
- Owned pages without a brand mention → rewrite to anchor the brand.
- External sources on the "Not Mentioned" branch → targets for PR, partnerships, outreach.
- Competitor pills on the external sources reveal who's winning the conversation where you're absent.
- The Prompt tab lets you isolate any of the above to a single business-critical query.