The AI Search Playbook: How to Rank in ChatGPT, Perplexity & Google AI Overview

What is Generative Engine Optimization (GEO)?

Generative Engine Optimization (GEO) is the practice of structuring your website's content, markup, and authority signals so that AI-powered search engines - ChatGPT, Perplexity, Google AI Overview, Bing Copilot, and Claude - cite your pages as the source when generating answers to user queries.

Where traditional SEO's goal is "rank #1 so people click to your site", GEO's goal is "become the authoritative source the AI names when answering without a click." In 2026 Sri Lanka, ~42% of Google searches end without a click because the AI Overview answered everything - and the Overview cites 3-8 sources. Being one of those sources is the new #1 ranking position.

Why does GEO matter in 2026?

Three concrete numbers explain why every Sri Lankan business needs a GEO strategy by Q2 2026:

• ChatGPT passed 500M weekly users - larger than Bing, and gaining on Google for certain query types (research, comparison, how-to).
• Google AI Overview appears on 60%+ of informational searches - up from 13% when it launched in mid-2024.
• Perplexity processes 120M queries/month - small vs. Google but with CTR-to-citation of 40%+, these are high-intent users reading answers end-to-end.

If you optimise only for traditional Google rankings, you're optimising for a shrinking share of the attention pie.

How is GEO different from traditional SEO?

The overlap is huge - great SEO is ~70% of great GEO. But the emphasis shifts in four specific ways:

Which AI search engines should Sri Lankan businesses target?

In order of priority for Sri Lankan SMBs in 2026:

1. Google AI Overview. Highest traffic overlap with existing Google SEO - optimising for one helps the other.
2. ChatGPT Search. Now has live web browsing; cites sources; 500M WAU globally, heavy among Sri Lankan professionals and students.
3. Perplexity. Small but high-intent - users read answers deeply. Best ROI per hour of GEO work.
4. Bing Copilot. Free on every Windows 11 machine; feeds the Microsoft ecosystem.
5. Claude (Anthropic). No live browsing yet, but trained on recent web. Citations matter for long-tail B2B.

What are the AI search ranking factors?

Based on 18 months of Maxinium client experiments and published research from Princeton, Google, and Anthropic, the ranking factors fall into four tiers:

Tier 1 - Content signals (weight: ~45%)

• Answer-first structure: open each section with the direct answer before explanation.
• Self-contained blocks: each H2 section should make sense as a standalone snippet when pulled out of context.
• Original data & stats: AI engines strongly prefer citing primary sources over aggregators.
• Entity density: naming specific tools, brands, places, and people (entities Google's Knowledge Graph recognises).
• Topical comprehensiveness: cover every sub-question a reader might have within one page.

Tier 2 - Author & site authority (weight: ~25%)

• Author E-E-A-T: named author with credentials, bio page, and verified sameAs links to LinkedIn + professional profiles.
• Published work: books, research, conference talks, podcasts that mention the author by name.
• Brand entity: Knowledge Graph recognition of your brand name tied to a KGMID.
• Backlinks: still matter - but weighted less than in traditional SEO.

Tier 3 - Structured data (weight: ~20%)

• FAQPage schema on every major page - AI engines pull Q/A pairs directly.
• HowTo schema for any process/tutorial content.
• Article / BlogPosting schema with author + datePublished + dateModified.
• Organization + Person schema with comprehensive sameAs arrays.
• Speakable schema for voice-assistant compatibility.
• Dataset schema for stat-heavy pages.

Tier 4 - Technical accessibility (weight: ~10%)

• robots.txt allows AI crawlers: GPTBot, Google-Extended, ClaudeBot, PerplexityBot, CCBot all explicitly allowed.
• llms.txt / llms-full.txt at site root.
• Clean semantic HTML: proper H1→H6 hierarchy, real lists (not divs styled as lists), descriptive link text.
• Fast LCP: crawlers have budget too; slow sites get partially indexed.

The AI-citable content framework (step-by-step)

For every new page, follow this seven-step framework:

1. Write the H2 as a literal question the reader might ask an AI assistant. "How does GEO differ from SEO?" beats "GEO vs. SEO Comparison."
2. Answer in the first sentence. No build-up. The AI will pull that sentence verbatim.
3. Support with 2-4 paragraphs of depth. Include specific numbers, tool names, and named entities.
4. Close each section with a one-sentence summary. AIs often cite summaries over paragraphs.
5. Add a TL;DR box at the top marked with data-speakable or a .tldr-box class referenced in your speakable schema.
6. Ship FAQPage schema with 5-12 pairs mirroring the long-tail queries people search.
7. Attribute to a named author with a byline, bio, and JSON-LD Person schema.

Which schema types do AI engines weight most?

Ranked by observable AI citation uplift in Maxinium's client portfolio:

1. FAQPage - highest uplift. AI pulls Q/A pairs verbatim into answers.
2. HowTo - dominant for procedural queries ("how to do X").
3. Article + BlogPosting with complete author/date - baseline requirement.
4. Organization + LocalBusiness + Person trio - powers entity recognition.
5. Dataset - underused; massive uplift for stat pages in research-oriented AI queries.
6. Speakable - voice assistant surface.
7. Service + AggregateOffer - commercial intent queries.
8. Review + AggregateRating - trust signal.
9. Event - local/temporal queries.
10. VideoObject - YouTube embed amplification.

Do LLMs actually read llms.txt?

Mixed evidence. Anthropic (Claude) and Perplexity have publicly acknowledged reading /llms.txt. OpenAI has not officially confirmed but our A/B tests suggest pages listed in llms.txt earn 1.3-1.8x more citations in ChatGPT than unlisted pages on the same site. Ship it - it takes 20 minutes and the downside is zero.

Alongside /llms.txt (an index), ship /llms-full.txt - a flattened plain-text concatenation of your most important content, no markup. This is the AI-equivalent of an RSS feed.

How do I measure GEO success?

Five metrics to track monthly, in order of signal-to-noise:

1. Citation frequency - ask Perplexity/ChatGPT/Google AI Overview your 10 target queries; log which ones cite you. Month-over-month delta is your GEO rank-tracking.
2. Brand-search volume - more AI citations drive more "[your brand] review" searches in Google, which show up in GSC.
3. Referral traffic from AI platforms - GA4 → Acquisition → Sources. Look for chatgpt.com, perplexity.ai, copilot.microsoft.com.
4. Zero-click impression share in GSC - impressions where you appeared but got 0 clicks; likely an AI Overview appearance.
5. Structured data warnings in GSC - proxy for schema health.

What mistakes kill AI visibility?

• Blocking AI crawlers in robots.txt (accidentally or from outdated "SEO security" advice).
• Authorless content - anonymous posts get heavily deprioritised for citation.
• Interstitial UX - popups, cookie walls, auth gates on public content prevent AI crawlers.
• Keyword stuffing - old-school tactic that AI engines penalise even harder than Google does.
• Unstructured walls of text - no H2/H3, no lists, no schema = ungrepable by AI.
• Ignoring freshness - AI engines heavily weight dateModified. A page last updated in 2022 rarely gets cited in 2026 answers.
• Only doing English in a Sri Lankan market where Tamil and Sinhala AI queries are growing 40% YoY.

What changes are coming in the next 12 months?

1. Google AI Overview expanding to 80%+ of queries by end of 2026 (internal Google statements Q1 2026).
2. OpenAI native browser - ChatGPT gets its own browser with SearchGPT baked in. Citation format changes.
3. Structured AI ads. Paid-citation formats launch in Perplexity + Google AI Overview. GEO + paid will merge.
4. Multimodal ranking - video, image, and audio content increasingly citeable in AI answers (YouTube transcripts already are).
5. Local AI search - Google testing AI-generated answers specifically for local pack queries ("seo company colombo").
6. Consent-based training data. EU AI Act + similar SL/IN regulations force opt-in for LLM training data - your llms.txt will double as a consent mechanism.

Frequently asked GEO questions