# Unmeasured citation
You can't tell whether AI engines quote your site because you don't measure it. The four other layers of AEO become faith-based exercises without the feedback signal from Layer 5.
By AgentSite · 5 min read · Updated 2026-05-24
Unmeasured citation is the operational gap where you can't tell whether AI engines are quoting your site. You don't know which engines cite you, for which prompts, at what rate, or which competitors come up instead. The four layers below it in [the five-layer AEO model](/five-layer-aeo) become faith-based exercises without this feedback signal.
## What it looks like
You can't answer questions a customer might ask you in five minutes:
- "What's our citation rate on ChatGPT for category queries?"
- "Which three pages get cited most often, on which engine?"
- "Did our last content rewrite move the number?"
- "Who comes up instead of us on Perplexity?"
If those feel hypothetical rather than tracked, you have this problem. The common variants:
- **One blended score.** A single "AI visibility" number across engines. Hides the fact that what wins on ChatGPT loses on Perplexity.
- **Single-run sampling.** Each prompt run once per reporting period. LLM responses are non-deterministic; a single run is a coin flip, not a measurement.
- **Last-90-day rollups only.** No per-prompt detail, no per-engine split.
- **Manual checks.** Someone runs five prompts in ChatGPT once a quarter and writes a Slack message. Not a measurement.
- **Nothing at all.** The most common variant.
## How to detect it
Ask whoever owns AEO on your team for last week's per-engine inclusion-rate report. The report should answer: how many of N prompts mentioned us, broken out by engine. If the answer is "we don't have one" or "we have this dashboard but it's a single blended number," you have this problem.
## Why it costs you citations
Every layer above this depends on the feedback signal. You tune Layer 4 content. You fix a Layer 1 SSR gap. You ship a Layer 2 `llms.txt`. And you can't tell which fix moved the number, because the number isn't being tracked.
Without inclusion-rate data:
- You can't separate fixes that worked from fixes that didn't.
- You can't tell whether your category is being captured by a competitor or simply isn't being asked about.
- You can't justify the budget for the next round of AEO work.
- You can't catch regressions when an engine update flips your pages out of the citation pool.
The Princeton GEO paper measured per-tactic visibility lifts of up to 40% in controlled experiments ([Aggarwal et al., KDD 2024](https://arxiv.org/abs/2311.09735)). Those numbers are meaningful precisely because they were measured against a 10,000-query benchmark. The same fix without measurement is faith.
The volume context matters too. Cloudflare reported AI bots accessing roughly 39% of the top one million Internet properties in a single month ([Cloudflare, July 2024](https://blog.cloudflare.com/declaring-your-aindependence-block-ai-bots-scrapers-and-crawlers-with-a-single-click/)), and Vercel measured 569 million GPTBot fetches and 370 million Claude fetches across their network in one month ([Vercel, Dec 2024](https://vercel.com/blog/the-rise-of-the-ai-crawler)). The traffic exists. The question is what it does on your specific URLs — and that is measurable, but only if you set up the probe.
## How to fix it
Three steps:
1. **Define a prompt panel.** 30 prompts minimum; broader categories want 100+. Each prompt phrased as a real user would ask it ("best AEO tool for SPAs" not "AEO tool"). The panel stays constant while content changes, so deltas mean something.
2. **Run the panel against each engine separately.** ChatGPT, Claude, Perplexity, Google AI Overviews, Bing Copilot. Different prompts get cited by different engines; a blended score hides more than it reveals.
3. **Sample with enough runs per prompt to average out non-determinism.** Practitioners cluster around 10-100 runs per prompt per reporting period. One run per prompt is a coin flip.
The output is an inclusion-rate-per-engine table you watch over time. When it changes, the change is the signal.
## Where AgentSite fits
The AEO Report includes a citation probe that runs your category prompts against an answer engine and reports inclusion rate, citation count, and competitor share-of-voice. Single-engine (OpenAI) today; multi-engine expansion (Claude, Perplexity, Gemini, Bing Copilot) is in development. The report is the same measurement you'd build in-house, automated.
## Related problems
- [SSR-junk and bot walls](/ssr-junk-bot-wall) — the Layer-1 failures that show up as zero citation on every engine.
- [Buried answer](/direct-answer) — the Layer-4 issue that shows up as low citation even when other layers work.
- The full catalog: [AEO problems](/aeo-problems).