Skip to content
GeoSalience
exp-002PlannedAI-crawler hits

Does FAQPage schema raise AI crawl frequency or citation?

Half the GEO industry says FAQPage JSON-LD wins AI citations; the other half says LLMs ignore hidden markup. This adds FAQ + FAQPage schema to one article, holds a comparable one unchanged, and measures the difference. Proposed — awaiting editorial go-ahead.

Hypothesis
Adding a FAQ section plus FAQPage JSON-LD to an article raises its AI-crawler hit frequency and (site-wide) citation rate, relative to a comparable article left unchanged.
What changed
Treatment article gains a FAQ section (drawn from its existing H2/H3 questions) plus a FAQPage JSON-LD block. Control article is left exactly as is. No other change to either page in the window.
Treatment
Control
Metric
AI-crawler hits
Window
proposed 7 June 2026 → 7 July 2026

Status: proposed. This protocol is published for transparency, but the experiment has not started. No page has been changed. The start and review dates above are proposed, not actual; the editor picks which experiments run and confirms the control/treatment pairing, because comparability is a judgment call.

The question

Does shipping FAQPage JSON-LD actually earn more AI attention, or do the crawlers just read the visible page and ignore the markup? This is the smaller, on-site version of the MVP-4 Schema.org A/B test design — one matched pair on our own domain, as a first signal before any multi-domain study.

Proposed design

  • Treatment: add a FAQ section and a FAQPage block to the llms.txt spec article.
  • Control: leave What is GEO? unchanged.
  • Primary metric: AI-crawler hits per page, before vs after, treatment minus control (difference-in-differences).
  • Window: a 14-day before window and a 30-day review after the change lands.

Why the pairing needs editorial sign-off

The two pages are both live and both cornerstone, but they are different types — an explainer and a how-to — and page type plausibly affects how often a crawler returns. That is a real confounder. The stronger design waits until two same-type articles are live and pairs those instead. The editor decides whether to run the imperfect pair now (with the confounder disclosed) or wait for a clean one.

What would count as a result

A clear, repeated rise in crawler hits on the treatment page that the control page does not share, sustained across the 30-day window. Anything smaller, on a two-page sample, we will report as inconclusive rather than dress up as a finding.

Analysis readout

This experiment has not started. Once it runs and the crawler/citation data accrue, the before/after readout appears here and updates as results land.

Limitations

  • Single domain, single matched pair (n=2 pages). This detects only a large effect; a small one is invisible at this sample.
  • The two live articles are of different types (a definitional explainer vs a technical how-to), so page type is a confounder. The cleaner design pairs two same-type articles once more are live.
  • Citation effect is not page-attributable: the citation harness measures the whole site, so any citation movement is context, not proof about the treatment page. Crawl frequency is the primary, page-level metric here.
  • Requires the AI-crawler tracking pipeline (prompt 07) to be streaming before the window opens — otherwise there is no per-page crawl series to compare.

Changelog

  • Published — 31 May 2026

Raw markdown: /lab/experiments/exp-002.md