GEO playbook: what the research actually supports

Generative engine optimization (GEO) means shaping content so AI answer engines select and use it. Here is what the evidence says, with sources.

1. Statistics: supported

The KDD 2024 GEO paper (Aggarwal et al.) reported visibility improvements of up to 40% from methods including adding statistics, and up to 37% on Perplexity. Paper summary. A replication across 3,205 real pages on ChatGPT, Claude, Perplexity and Google AI Mode found statistics still helped on every platform. Replication.

2. Citation and quote stuffing: not supported in production

The same replication found citation and quotation density did not replicate and often went the wrong direction.

3. Be an "evidence container"

A 2026 study of 602 prompts across ChatGPT, Google AI Overview/Gemini and Perplexity found high-influence pages were longer, modular, semantically aligned with the answer, and rich in definitions, numbers, comparisons and procedural steps. Q&A formatting alone didn't help. arXiv 2604.25707.

4. Technical basics

• Allow AI crawlers (GPTBot, PerplexityBot, ClaudeBot, Google-Extended) in robots.txt.
• Add Article, Organization and FAQ structured data.
• Rank in classic search; AI Overviews draw heavily on it.
• Earn independent third-party mentions.

Check a page yourself: free AI citation readiness checker. Tools to measure it: AI visibility tools compared.

FAQ

Does adding statistics help AI citations?

The original GEO paper (KDD 2024) reported gains from adding statistics, and a 3,205-page replication on production engines found statistics still helped.

Do citations and quotations help?

The original paper reported gains, but the production-platform replication found citation and quotation density did not help and often went the wrong direction.

Does Q&A formatting alone help?

A 2026 measurement study (arXiv 2604.25707) found Q&A formatting alone did not improve how much a cited page shaped the answer.