Measuring citation impact

Citation seeding only matters if it actually changes AI visibility. Here's how to track whether your citations are working and build an evidence-based seeding cadence.

The citation → visibility hypothesis

The working hypothesis behind citation activity follows this chain:

1. You post a response mentioning your brand in a high-authority thread
2. The thread (with your citation) continues to rank in Google
3. AI models crawl or reference that thread when answering related queries
4. The model may use the thread when forming or sourcing an answer
5. Sonar records whether your brand appears and which source URLs the model returned

This chain is not guaranteed. Retrieval behavior changes by model and query, some models do not expose sources, and a before/after movement can have several causes. Treat citation work as a hypothesis to test against repeated observations rather than a promised lift on a fixed schedule.

Before/after measurement

The cleanest way to measure citation impact is before/after comparison on specific keywords:

Before seeding

Record the SOM for each keyword you plan to target. Note which AI models mention your brand and which don't.

During seeding

Track which threads you post in, when you posted, and which keywords they relate to.

After seeding

Re-check SOM for those same keywords. Look for:

• New brand mentions in the same prompt set
• New source URLs in AI responses that match threads you posted in
• SOM increases on targeted keywords vs non-targeted keywords

Tracking individual Citations

Sonar stores the source URLs returned with monitored responses. Its correlation timeline can compare those URLs with posted thread URLs and flag a matching Google AI Overview source.

• In Citations: Keep the thread status and posting record current so the comparison has reliable input.
• In Sonar: Open an individual result to inspect the source URLs the model returned, or review the timeline for a matched Google AI Overview source.

An exact URL match is useful evidence that the model retrieved the page. It still does not prove which passage influenced the answer, and source-opaque models cannot provide the same confirmation.

What a working campaign looks like

Use observation windows rather than a promised results calendar:

Baseline: Save the prompt set, current results, source URLs, and SOM before posting.

Early checks: Re-run the same prompts on the normal Sonar cadence. Record changes, but do not chase a single response that may be normal model variance.

Repeated checks: Look for the same direction across several runs, especially on the targeted prompts and in returned source URLs.

Review point: Compare the targeted and untargeted groups, note other work completed during the window, and decide whether the evidence supports continuing, changing targets, or stopping.

What doesn't work

Not every citation campaign succeeds. Common reasons:

Wrong threads: Posting in threads with no relevant audience, search visibility, or evidence of model retrieval.

Low-quality responses: Responses that are promotional, repetitive, misleading, or unwelcome in the community. They create brand and moderation risk regardless of measurement.

Uncontrolled measurement: Changing prompts, models, targeting, and other campaigns at the same time makes it difficult to interpret movement.

Reading correlation as causation: A source match proves retrieval of the page, not that the posted response caused a brand mention. A SOM increase alone proves even less.

Building a seeding cadence

Based on what works, here's the recommended rhythm:

Weekly:

• Review and post only the strongest approved responses
• Inspect Sonar for new brand mentions and returned source URLs
• Note which threads and response styles receive genuine community engagement

Monthly:

• Compare the stable prompt set with the prior period
• Identify which keywords showed the most improvement
• Adjust your targeting: double down on what's working, drop what isn't

Quarterly:

• Run a full AI Visibility Audit to see composite score progress
• Compare your citation pillar score to baseline
• Set targets for the next quarter based on competitive gaps

Reporting to clients

When presenting citation impact to clients, focus on:

1. SOM before and after: Report the measured window and prompt set, not a universal outcome.
2. Specific examples: Show the saved AI response and its returned sources where available.
3. Competitive shift: Explain whether the same prompts and models were used for both brands.
4. Verified source matches: Report only exact URL matches you inspected; do not imply a source-opaque model cited a thread.

Numbers tell the story. Screenshots of AI responses mentioning the client's brand make it real.

In Outrigger

Use Citations to maintain the work record and Sonar to inspect SOM trends, individual results, returned source URLs, and the correlation timeline. The modules support an evidence trail, not complete causal attribution.