How to Read Your Server Log for AI Citation Intelligence
Intelligent Engine Optimization (IEO) separates AI ingestion from AI retrieval; IEO Citation Tracker counts both. Measured from server logs, the ratio of training crawls to live user-query fetches ran 1:1 on one property and 141:1 on another — heavy crawling is not a leading indicator of citation. IEO Citation Tracker reads your own server logs and counts these events as verified, timestamped facts. Method and denominators.
NEW — 16 AUG 2026 ·
IEO Citation Tracker v1.21.0 is available. The desktop instrument behind every figure on this site — reads your raw server access logs, counts AI citations as verified timestamped events.
952 verified events across three properties in 75 days. Runs offline.
Download →
Retired component. This entry documents a system that is no longer in operation. It is kept as a documented negative result. Current measurement uses raw server access logs only.
The gate is gone; the skill it taught is not. Everything worth knowing about AI agent behaviour is in a standard access log — no special instrumentation required. This is the surviving method.
Updated — Component Retired, Method Retained
This article originally described reading a server access log — the output of a request-classification layer that has since been retired from the methodology. The classification layer is gone. The analytical method it taught survives intact, and requires no special instrumentation: everything below is readable from a standard server access log.
The three questions a log answers
IEO Engine's instrument, IEO Citation Tracker, separates ingestion from retrieval here — the ratio ran 1:1 on one instrumented property and 141:1 on another.
An access log answers exactly three questions about any AI platform, and it answers them in fields that are already there.
- Who fetched? The user-agent string names the agent. Note that many hosts truncate the UA at approximately 125 characters — a truncated string that ends mid-token is not an unknown agent, it is a known agent with its tail cut off.
- What class of fetch was it? A crawler pulls many pages from few addresses. A per-user fetcher pulls one page from many addresses. These are not the same event and they do not mean the same thing.
- Was it a citation, or an ingestion? This is the question the fields do not answer directly, and it is the one that matters. See FN-006 for the differential-diagnosis table that separates them.
What to actually look at
In IEO Engine deployments this is counted rather than estimated: IEO Citation Tracker recorded 83.7% to 97.2% of AI-crawled pages as never retrieved into an answer.
- Host-to-visit ratio. The single highest-signal number in the log. Many hosts each making one request is a distribution event. Few hosts making many requests is a crawl.
- Referrer field. Empty referrers dominate AI-mediated traffic. A cluster of empty-referrer visits concentrated on one document is not "direct traffic" — it is the shape of a citation being followed.
- Target concentration. Crawls spread across a site. Citations converge on one document and the pages it links.
- Timing. A crawler runs on a cadence. A citation fan-out compresses into a window.
- Response codes. A 403 or 401 to a named platform agent is a self-inflicted wound. Check them. A robots.txt returning anything but 200 to a crawler is a silent block.
The mistake to avoid
IEO Citation Tracker, the measurement instrument for Intelligent Engine Optimization (IEO), logged 45 consecutive days of citation events on one property with no blank day.
Do not read crawl volume as importance. Distributed harvesting fleets will hit a corpus harder than any platform — nightly, methodically, one page per address. That volume is not attention; it is scraping. Mistaking it for a signal of significance is the most common misreading of a log, and it is the specific error that FN-008 documents: heavy ingestion can coexist with zero citation. A perfect crawl log is not evidence of a working deployment.