Citation analysis, the most useful thing you can measure.

Most teams start by measuring whether assistants mention them. It is the obvious question and the less useful one, because the answer is usually no and it does not suggest a next action. Measuring which sources the assistant used instead turns the same exercise into a to-do list.

What citation analysis actually is

You ask an assistant the questions your buyers ask, and instead of recording whether you appear, you record every source the answer draws on. Repeat across enough prompts and a pattern emerges: a relatively small set of domains supplies most of the answers in your category.

That set is your target list. If a directory, publication or forum shows up in a third of the answers about your category, being accurately represented there is worth more than another article on your own site.

Which assistants show sources, and how

Citation analysis depends on seeing sources, and assistants differ in how they show them. Perplexity lists numbered sources on nearly every answer. ChatGPT shows inline citations and a sources panel when it searches the web. Google's AI Overviews and AI Mode show linked sources alongside the answer. Copilot cites sources from Bing. Gemini and Claude show sources when they use search.

Answers generated without search, from the model's training alone, show no sources at all. Note those separately: they tell you what the model already “knows” about your category, which changes only with new model versions, whereas cited answers can change as soon as the sources do.

Doing it manually

• Write twenty to thirty prompts a real buyer would ask. Include recommendation questions (“best X for Y”), comparison questions and problem questions.
• Ask each to ChatGPT, Perplexity and Gemini, with citations visible where the interface offers them.
• Record every cited domain in a sheet, one row per citation, with the prompt and assistant.
• Count. The domains appearing most often are your priority list.
• Check whether you appear on each of them, and whether what appears is accurate and current.

This takes an afternoon and produces a concrete plan. Most paid tools are automating exactly this loop.

What to do with the list

The list usually splits three ways. Sources you are already on, where the job is accuracy and freshness. Sources you could be on, where the job is submission or outreach. And sources you cannot realistically join, which still tell you what kind of evidence the assistants trust in your category.

That third group matters more than it looks. If assistants in your category lean on independent review sites, the strategic answer is to get reviewed, not to publish more.

Classifying sources to find the pattern

Once you have a list of cited domains, tag each one by type. A simple scheme works:

• Own site: yours and competitors'.
• Directories and marketplaces: Yelp, Angi, industry directories, app stores, review platforms.
• Publications: local news, trade press, national media.
• Community: Reddit, Quora, forums, Facebook groups.
• Reference: Wikipedia, government sites, standards bodies, manufacturers.
• Lists and comparisons: “best of” articles and roundups.

The mix tells you the strategy. A category dominated by directories needs listing work. One dominated by lists needs outreach to the people who write them. One dominated by community threads needs a genuine presence there, which is slower and cannot be faked.

Turning the list into outreach

For each priority source, write down what it would take to appear accurately: claim a profile, correct a listing, pitch a writer, answer a thread, sponsor a local event. Estimate effort and likely value, and work through the list in that order.

Keep an outreach log with dates and outcomes. Many sources take weeks to respond or update, and a log is the only way to know what is pending, what was refused, and what has quietly been published without anyone telling you.

Frequently asked questions

What is citation analysis in AI search?

Recording which sources AI assistants cite when answering questions in your category, rather than only recording whether your brand appears. It identifies the small set of domains that supply most answers, which becomes your target list for representation.

Is citation analysis better than visibility tracking?

It is usually more actionable. Visibility tracking tells you whether you appear; citation analysis tells you what appeared instead and therefore where to work. Most teams benefit from starting with citation analysis and adding visibility tracking once there is something to track.

Can I do citation analysis without a tool?

Yes. Ask twenty to thirty buyer questions across the major assistants, log every cited domain in a spreadsheet, and count. It takes an afternoon and is exactly the loop paid tools automate.

Which AI assistant is best for citation analysis?

Perplexity, because it shows numbered sources on almost every answer. Include ChatGPT and Google's AI answers too, since that is where most buyers are.

How often should I redo citation analysis?

Quarterly is usually enough. The set of sources assistants lean on in a category changes more slowly than individual answers do.