# AI Index Lab

> Independent research publisher on what actually predicts AI citation. We run controlled studies comparing pages AI engines cite against pages they ignore for the same queries, and publish the methodology, data, and statistical analysis in full. Our mission is to replace assumption with evidence in the emerging field of AI search optimisation.

## About

AI Index Lab is an independent research organisation founded in 2026 by Chris Levick. We study how AI answer engines — including ChatGPT, Perplexity, Google AI Overviews, Claude, and Microsoft Copilot — select which sources to cite. Rather than relying on correlation or intuition borrowed from traditional SEO, we use controlled experimental design to isolate which page-level, domain-level, and content signals actually predict AI citation.

All findings are published with full methodology, raw effect sizes, p-values, sample-size disclosures, and limitations. We report null results explicitly.

## Research Methodology

Our studies use controlled comparison design — the same approach used in clinical trials. We don't observe AI-cited pages in isolation; we compare them against pages that rank in traditional search for the same query but weren't cited by AI. This controls for topic, competitive intent, and baseline discoverability.

Statistical methods we use:
- Welch's t-test for unequal-variance group comparison
- Cohen's d for standardised effect sizes
- Per-query within-subject paired comparisons
- Bonferroni correction for multiple comparisons
- Sample-size power analysis disclosed alongside results

## Published Research

### Report #001 — What Actually Predicts AI Citation? A Controlled Study of 892 URLs (April 2026)

Our founding study. Phase 1 collected 646 AI-cited URLs from Perplexity and Claude across 50 informational queries spanning 10 verticals. Phase 2 introduced controls by collecting Google top-10 results for each query and comparing cited vs uncited pages within the same competitive set.

**Key findings:**
- On-page signals that appeared predictive in Phase 1 (images +433%, external links +278%, word count +83%) collapsed to statistical insignificance under controls (all p > 0.2).
- Four of five statistically significant signals were **negative**: pages with author tags (d=-0.31, p=0.0002), publish dates (d=-0.23, p=0.004), schema type counts (d=-0.17, p=0.03), and Open Graph metadata (d=-0.18, p=0.03) were LESS likely to be cited.
- The data suggests AI citation operates primarily at the domain level, not the page level. Standard AEO recommendations built around on-page signal optimisation are likely misallocating effort.

Full report: https://www.aiindexlab.com/research/001-what-predicts-ai-citation

## Subscription Model (Planned)

The founding report is free. Future paid tiers will differentiate by access to raw datasets, early-access, custom queries, and API access — not by gating already-published content. We will not gate research findings themselves.

- Free: weekly intelligence email + all published reports
- Paid tiers (launching after subscriber base reaches scale): raw CSV datasets, client-ready templates, early access, quarterly deep-dives, API access

## Contact

- Website: https://www.aiindexlab.com
- Email: chris@aiindexlab.com
- Founder: Chris Levick

## Links

- [Home](https://www.aiindexlab.com/)
- [Research archive](https://www.aiindexlab.com/research)
- [Report #001](https://www.aiindexlab.com/research/001-what-predicts-ai-citation)