After grading 10,000 ChatGPT answers across 50 local-service categories, 4 page-level patterns predicted citation rate better than any backlink metric.

TL;DR

We ran 10,000 GPT-4o queries across 50 local-service categories (plumbers, dentists, HVAC, lawyers, salons, etc.) in 25 US metros and graded every cited source. Four page-level features predicted citation rate better than Domain Rating, backlinks, or word count:

A clear **definition paragraph** in the first 200 words (≈3x lift)
**FAQPage JSON-LD** with 5+ Q&A pairs (≈2.4x lift)
A **comparison table** anywhere on page (≈1.9x lift)
A linked **llms.txt** at the site root (≈1.6x lift)

Method

We sampled 200 questions per category — a mix of "what is", "best", "near me", and "how much does" intents — and asked GPT-4o to answer each one with citations. We recorded every cited URL, then crawled each URL and extracted 32 on-page features. We fit a logistic regression with citation as the dependent variable.

What didn't matter

**Backlinks.** Ahrefs DR explained <3% of citation variance.
**Word count past 1,200 words.** Long posts didn't outperform medium ones.
**Page speed beyond LCP < 4s.** Below that threshold, faster pages weren't cited more.

What mattered most

The four features above. Crucially, they compound — pages with all four were cited in 41% of relevant queries, vs 6% for pages with zero.

What this means for local businesses

Stop chasing DR. Start instrumenting your pages for extractability. The cheapest path to AI visibility in 2026: rewrite your top 10 pages to lead with a definition, ship FAQ schema, add a comparison table, and publish llms.txt.

What we built

Every page generated on Local Pages now ships with all four features by default — definition block, FAQ schema, comparison table, and a llms.txt/llms-full.txt pair per business. We'll re-run the study in 90 days and publish the deltas.

We analyzed 10,000 ChatGPT answers about local services — here's what got cited