Most brands track whether they appear in AI answers. Very few ask a more useful question: which websites do AI engines already trust in their category?

That second question is where the leverage sits. Citation share is a lagging indicator. By the time you see your number drop, the engine has already shifted its source preferences weeks ago. An AI citation source analysis looks upstream: which domains earn citations, why they earn them, and where your brand falls short.

Without that picture, every GEO decision is a guess. With it, you know exactly which content types, formats, and third-party placements will move your visibility, because you can see what the engine is already rewarding in your category.

What Is Citation Source Analysis And Why Does It Matter?

An LLM citation analysis is a structured audit of the domains AI engines cite when answering the queries your buyers use. It is not a ranking check. It is a map of the source ecosystem the engine draws from in your category.

Every AI Engine Has A Source Preference

The assumption that all AI engines pull from the same pool is wrong. Research consistently shows that each platform maintains distinct citation patterns. ChatGPT leans on Wikipedia and Reddit. Perplexity weights YouTube and niche publications. Google AI Overviews favours YouTube across most intent types. Claude cites institutional and brand-owned sources more heavily than any other engine.

Our research on how brands show up differently across AI platforms documents this divergence in detail. The practical consequence: a brand visible on Perplexity may be invisible on ChatGPT for the same query, because the two engines trust different source types.

An LLM visibility audit that treats all engines as one channel misses this entirely. The analysis has to run per platform to produce decisions you can act on.

How To Run The Analysis

The method is straightforward, but the details determine whether the output is useful or decorative.

Step 1: Build The Prompt Set

Start with the 20 to 40 prompts your brand needs to appear in. These should map to your highest-intent queries: "best [category] for [segment]," "[your product] vs [competitor]," "how to [solve the problem your product addresses]." Include recommendation, comparison, and evaluation prompts, because these are where citation source patterns are most distinct.

Fewer than 15 prompts tend to surface noise rather than signal. Between 20 and 40 is the range where source preferences repeat consistently enough to base decisions on.

Step 2: Capture Citations Across Platforms

Run every prompt across ChatGPT, Perplexity, Google AI Overviews, Google AI Mode, and Gemini. For each answer, log every cited URL, the domain, and whether your brand was mentioned. This is the raw dataset for the LLM content visibility scanner layer.

Run the same set weekly for at least three weeks before drawing conclusions. AI citation sources rotate 40 to 60% month over month, so any conclusion from a single run risks treating a temporary citation as a durable preference.

Step 3: Classify Sources By Type, Not Just Domain

This is where most analyses stop too early. A list of cited domains is not actionable. A classification by source type is.

Group every cited domain into categories that map to things you can influence: brand-owned content, competitor content, earned media (publications, trade press), community and UGC (Reddit, forums), reference sources (Wikipedia, institutional sites), and video (YouTube).

The classification reveals the engine's source preference for your category. If 60% of citations come from earned media and community content, optimising your own blog is necessary but insufficient. The engine is telling you where it looks for trust, and you need to be present in those source types.

What The Analysis Tells You That Citation Share Does Not

Citation share tells you how often you appear. An AI citation source analysis tells you why you appear or do not, and what would need to change.

Identifying The Domains That Anchor Your Category

In most categories, a small number of domains account for a disproportionate share of citations. Identifying those anchor domains tells you two things: what content characteristics the engine rewards (format, depth, evidence density, freshness), and where your brand needs third-party presence to enter the citation set.

If a specific trade publication anchors citations for your category across multiple engines, a placement in that publication is worth more than ten blog posts on your own site. If Reddit threads dominate comparison queries, your community engagement strategy becomes a direct input to LLM search visibility rather than a brand awareness exercise.

Spotting Shifts Before They Show Up In Your Numbers

Citation source sets are not static. Engines adjust retrieval mixes, legal disputes change access to platforms, and new content enters the pool. Running the analysis quarterly rather than once catches these shifts before they erode your position.

Our research on why AI citations may not be the best single visibility metric makes a related point: the durability of a citation matters more than its presence in any single snapshot. Source analysis is how you track that durability at the ecosystem level.

From Analysis To Action

The analysis produces a map. The map needs to become a prioritised list of moves.

If the engine trusts earned media in your category, invest in placements in the specific publications it cites. If it trusts community content, build an authentic presence in the subreddits it draws from. If it trusts brand-owned content with strong evidence density, restructure your pages with specific statistics, named examples, and structured claims.

The cleanest way to operationalise this is through Passionfruit Labs, which tracks citation share across platforms and connects source-level data to the work that moves the numbers. See how Passionfruit's GEO service builds visibility on top of this analysis, and talk to the team about running a citation source audit for your category.

Frequently Asked Questions

How Large Does A Prompt Set Need To Be Before Citation Source Analysis Produces Stable Results?

Between 20 and 40 prompts covering your highest-intent queries is the practical minimum. Below 15, individual citations create noise that looks like a pattern but is not. The prompts should span recommendation, comparison, and evaluation intent types, because source preferences vary by query type. Run the set weekly for at least three weeks before treating the output as stable enough to base content or placement decisions on.

How Do I Classify Cited Sources In A Way That Leads To An Action Rather Than A List?

Group domains into categories that map to things you can influence: brand-owned content, competitor content, earned media, community and UGC, reference sources, and video. The classification reveals which source types the engine trusts for your category. If earned media dominates, invest in placements. If community content dominates, build an authentic forum presence. If brand-owned content with strong evidence density appears, restructure your pages accordingly. A domain list is an observation. A source-type classification is brief.

What Does It Mean When One Domain Dominates Citations Across An Entire Category?

It means the engine has identified that domain as the most reliable source type for your vertical. This is common with trade publications, Wikipedia, or YouTube, depending on the category and engine. The response is not to replicate that domain. It is to understand what characteristics make it citeable (structured data, evidence density, freshness, editorial independence) and build those characteristics into your own content and third-party presence. If the dominant domain is a publication, a placement there carries disproportionate citation weight.

How Frequently Do Citation Source Sets Change, And How Often Should The Analysis Be Re-Run?

Citation sources rotate 40 to 60% month over month across most engines. Quarterly analysis is the minimum cadence to catch meaningful shifts before they erode your position. Brands actively investing in GEO should run monthly checks on their priority prompt set. Major platform events like retrieval-mix changes, data licensing disputes, or model updates can shift source preferences within weeks, so any sudden drop in citation share should trigger an immediate re-run rather than waiting for the next scheduled cycle.

Can Citation Source Analysis Be Automated, Or Does It Require Manual Work?

Prompt execution and URL logging can be automated through AI visibility tracking tools. The classification and interpretation layer still requires human judgment, because deciding whether a source type is actionable depends on your content capabilities, earned media pipeline, and competitive position. The most effective setup is automated data collection with manual strategic interpretation on a monthly or quarterly cycle.