# LLM Visibility Metrics: A Practical Measurement Guide

> Canonical source: [LLM Visibility Metrics: A Practical Measurement Guide](https://www.answer.cloud/pages/llm-visibility-metrics-the-kpi-playbook-for-seo-and-marketing-leaders-hosted/)

Define crawler activity, AI referrals, sampled mentions and citations, and on-site chat outcomes as separate metrics.

Measure AI discovery with clearly defined signals. Crawl traffic, referrals, citations, and conversions answer different questions and should not be combined into one unexplained score.

## Request-level activity

Count eligible requests by URL, date, and observed agent category. Record status codes and access problems. These logs show requests to your site, not what appeared in a private AI answer.

## Human AI referrals

Count sessions with an identifiable AI referral source and their meaningful outcomes. Missing or stripped referrers limit coverage. Compare conversions with an appropriate baseline and disclose the attribution method.

## Sampled answer visibility

Use a fixed question set and record the provider, product/model, search settings, date, region where relevant, and repeat runs. A sampled mention rate is the share of tested answers mentioning your brand. A sampled citation rate is the share of tested answers citing a defined set of your URLs.

```text
Sampled citation rate = answers citing a qualifying URL / answers tested
```

Report the denominator and test conditions. This is a result for the sample, not a measure of every user’s conversation. Bot logs alone cannot supply that denominator.

## On-site chat outcomes

Review questions asked in your own website chat, helpfulness signals where available, content gaps, and subsequent actions using an explicit attribution method. Do not assume your site can observe conversations visitors hold in unrelated AI services.

## Use changes to improve content

Investigate inaccessible pages, unclear explanations, stale claims, and missing answers. Compare the same measurements before and after a change, allowing for crawl and search delays. Avoid unsupported universal benchmarks or promises that a new file guarantees visibility.

For request-level implementation, see the [Cloudflare Worker guide](https://www.answer.cloud/docs/cloudflare-worker-agent-analytics-hosted/). For content fundamentals, see [AEO 101](https://www.answer.cloud/blog/aeo-101-a-beginner-s-guide-to-answer-engine-optimization-hosted/).