GEO Optimization Checklist
What actually drives AI citation — and what doesn't. Organized around the signals our four-study research line validated as drivers.
What changed in this checklist
Previous versions of this guide told you to build topical authority, add E-E-A-T signals like credential badges and expert-reviewed labels, and load pages with multimedia. Our research did not validate any of those as drivers of AI citation. This version is rebuilt around the signals that did predict citation. See the supporting work on the research page and the cross-study summary on what predicts AI citations.
How to use this checklist
This is a single unified checklist. There's no Free/Pro/Agency split anymore — these signals are the signals, regardless of which plan you're on.
Pillars are ordered by the strength of evidence in our research. Work top-down. Within each pillar, items are tagged High, Medium, or Low — priority reflects what mattered in the data, not what's quickest to ship. If you only do the High items, you've done the work that counts.
Priority Legend
Checklist by pillar
1. Answerability
Answer the question directly, near the top, in clear declarative sentences. This was the strongest predictor of AI citation in our research.
- Put the direct answer in the first sentence of each major sectionHigh
- Lead with the answer, not preamble like "In this article we'll explore..."High
- Write in declarative sentences: "X is Y" rather than rhetorical questionsHigh
- Include self-contained quotable snippets that stand on their own without surrounding contextHigh
- Cut hedging language ("maybe," "perhaps," "might possibly") when you can state a factMedium
- Replace marketing copy with concrete claims that a reader can verifyMedium
2. Citation Quality
Pages that cite authoritative outside sources are more likely to be cited themselves by AI assistants. Treat citations as load-bearing, not decorative.
- Link out to primary sources: .gov, .edu, peer-reviewed research, original studiesHigh
- Use inline attribution ("according to the Pew Research Center...") next to the claim it supportsHigh
- Cite the original source, not a blog summarizing the original sourceHigh
- Include a References or Sources section on long-form pagesMedium
- Quote statistics with the source and year next to the numberMedium
- Audit existing pages and replace dead or downgraded linksLow
3. Definitions
AI assistants pull definitional sentences disproportionately often. Give them clean, explicit "X is Y" statements to lift.
- Open each topic page with an explicit definition in the first paragraphHigh
- Use definitional patterns crawlers recognize: "X is...", "X refers to...", "X is defined as..."High
- Include the topic's canonical name (not just a pronoun) inside the definition sentenceHigh
- Define each technical term the first time it appears, inline or via glossary linkMedium
- Avoid circular definitions ("AI search is search powered by AI")Medium
- Mark definitional sentences with semantic HTML (<dfn>) where it fits naturallyLow
4. Readability
Plain, well-structured prose is easier for both readers and LLMs to parse and quote. Tight sentences, clear sectioning, no walls of text.
- Keep sentences short — aim for around 15 to 20 words on averageHigh
- Use one H1, then H2s for major sections, then H3s under those. Don't skip levelsHigh
- Break content into short paragraphs (roughly 50 to 100 words)Medium
- Use bullet lists and numbered lists for anything that's actually a listMedium
- Add tables for comparisons, specs, or structured dataMedium
- Cut compound-complex sentences. Two short sentences beats one long oneLow
5. AI Accessibility
If crawlers and LLMs can't reach or parse the page, nothing else matters. Treat this as the floor, not the ceiling.
- Render content server-side. If the main text only exists after JS executes, most AI crawlers will miss itHigh
- Allow major AI crawlers in robots.txt (GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot)High
- Remove noindex, nosnippet, and noarchive directives from pages you want citedHigh
- Add an llms.txt file at the site root pointing to your most quotable canonical contentMedium
- Use Article and FAQPage schema (JSON-LD) where appropriateMedium
- Reference your sitemap in robots.txtMedium
- Audit Cloudflare, WAF, and bot-mitigation rules — aggressive blocking quietly excludes AI crawlersMedium
6. Content Freshness
A mild but real positive signal. Visible recency cues help AI assistants prefer your version of an answer over a competitor's older one.
- Show a visible "Last updated" date on evergreen pagesHigh
- Keep datePublished and dateModified accurate in your schemaHigh
- Re-check statistics and claims annually — replace anything stale rather than letting it rotMedium
- Use current-year references where the fact actually changes year to yearMedium
- Add a brief changelog line on substantive updates ("Updated June 2026: refreshed data")Low
Don't bother
Tactics you'll be told to do that didn't move the needle in our research. Spend your time on the pillars above instead.
- Building "topical authority" by linking related articles into clusters. We didn't see this drive citation.
- Adding E-E-A-T badges, "expert reviewed by" labels, or author credential boxes for the sake of AI. Visible E-E-A-T signaling didn't predict whether AI assistants cited the page.
- Padding pages with images, infographics, or video for "AI visibility." Heavy multimedia was associated with worse recommendation survival, not better.
- Trying to answer every related question on a single page. Question Coverage was flat in our data — depth on the actual question beats breadth across adjacent ones.
- Chasing word-count thresholds. Length without answers doesn't buy you anything.
Read the research behind this
This checklist is downstream of our four-study research line. The research page covers methodology and findings. What predicts AI citations summarizes the cross-study takeaways in one place. The definitions reference covers the terminology used throughout.