Citation Tracking

Citation tracking monitors when AI search engines mention your brand in their responses. It's how you measure the ROI of your GEO efforts.

How It Works

1. Define your brand - Set up the brand name and variations to track
2. Create validations - Write prompts that represent how users search for your topic
3. Run validations - We query AI platforms with your prompts
4. Analyze results - See if and how your brand was mentioned

Why We Run a Prompt More Than Once

Answer engines are probabilistic. Ask the same question twice and you can get two different sets of brands. A rate built from a single run is a coin flip reported as a fact, so prompts you mark for measurement are run repeatedly, and against alternate phrasings of the same question as well as the original.

Those alternate phrasings are generated once and then frozen for the life of the prompt. If the question changed between scans, every week-over-week comparison built on it would be meaningless.

Repeated sampling produces two honesty signals shown next to the rate:

• Repeat agreement - Whether identical runs of the same phrasing agree with each other
• Phrasing spread - How far apart the rates for different phrasings of the same question sit

Together they resolve into an answer stability label — stable, variable, or volatile — so a prompt that looks strong on paper but flips on rewording is visibly flagged rather than quietly averaged in.

Mentioned Is Not Cited

Every run records two separate facts, and we never merge them into one number:

• Mentioned - Your brand name appeared in the answer text
• Cited - Your domain appeared in the sources listed underneath the answer

They have different causes. Being named without being linked usually means the model already knew you and didn't need to fetch your page. Being linked without being named means your page won the retrieval but lost the sentence.

Did the AI Search at All?

Before an engine can cite you it has to retrieve something. We record whether retrieval actually happened on each run, which splits "not cited" into two different problems:

• The engine never searched - It answered from what it already carried. Nothing you published that week could have been read. This is an entity/authority problem.
• The engine searched and picked someone else - Retrieval happened and a competitor won the source slot. This is the one your next page can actually fix.

Rates are also split by prompt class — branded questions that name you versus discoveryquestions that don't. Engines routinely recognize a brand by name while never surfacing it for the generic queries where customers actually discover products, and one blended rate hides that gap entirely.

Validation Results

For each validation, we capture:

• Citation status - Whether your brand was mentioned, cited, or both
• Sentiment - Positive, neutral, or negative context
• Competitors - Other brands mentioned in the response
• Full response - The complete AI-generated answer
• Sources cited - Links the AI referenced (where available)
• Answer stability - Repeat agreement and phrasing spread, on sampled prompts

Supported Platforms

Best Practices for Validations

• Use realistic queries - Write prompts the way real users ask questions
• Cover different intents - Informational, comparison, and transactional queries
• Include competitor queries - Track "X vs Y" comparisons
• Monitor category terms - "Best [category] for [use case]" queries