First published: 12 May 2026 · Last updated: 12 May 2026
Entity optimisation
Wikidata, sameAs schema, knowledge graph alignment.
Answer-first structure
40 to 60 word direct answer opens every H2.
Information gain
Original data, frameworks, named quotes.
Author E-E-A-T
Bylines, credentials, sameAs, author schema.
Topical cluster depth
15+ articles per pillar, internal linking architecture.
AI crawler access
robots.txt allows GPTBot, ClaudeBot, PerplexityBot.
Brand mentions
Coverage in trusted publications, podcasts, datasets.
Prompt baselining
Quarterly multi-engine query set, citation tracking.
Advanced schema
FAQPage, HowTo, Speakable, Article + ItemList stacks.
Tactic 1: Entity Optimisation via Wikidata and Schema
AI engines do not retrieve URLs the way Google's classical SERPs do. They retrieve entities, then attach passages and citations to those entities. If your brand, your founders, your products and your services are not first-class entities in the engines' knowledge graphs, you are competing for citation space with a handicap. The minimum viable entity setup:- A Wikidata item for the brand, with `instance of`, `industry`, `headquarters`, `founded`, `founder`, `official website`, and `described at URL` properties populated.
- An Organization schema block on your homepage with matching `name`, `url`, `logo`, `sameAs` (linking to Wikidata, LinkedIn, Crunchbase, X), `founder`, and `address`.
- Person schema for each subject-matter expert who appears as a byline, with `sameAs` linking to LinkedIn, Wikipedia (where eligible), ORCID (for academics), and Wikidata.
- For SaaS or product brands, SoftwareApplication or Product schema with `applicationCategory`, `operatingSystem`, and aggregate ratings where genuine.
Knowledge graph layer
- Wikidata item (Q-number)
- Wikipedia article (where eligible)
- Google Knowledge Panel
Schema layer (your site)
- Organization with sameAs array
- Person schema for authors
- Product or Service schema
Cross-reference layer
- LinkedIn company + author profiles
- Crunchbase, AngelList, G2
- Industry directories with bidirectional links
Citation layer
- Press coverage with brand name
- Podcast appearances
- Conference speaker listings
Tactic 2: Answer-First Content Structure
AI engines do not read your full article when they synthesise an answer. They lift the opening passage of the section that best matches the query intent. If your H2 opens with throat-clearing, context-setting prose before getting to the answer, you forfeit the citation to whoever was more direct. The discipline is mechanical: every H2 in a GEO-optimised article opens with a 40 to 60 word direct answer to the implied question of that H2, then expands. The opening passage must stand alone as a quotable definition or finding. Anti-pattern: > "When we look at how AI engines have evolved over the past two years, it becomes clear that traditional SEO approaches need rethinking. In this section, we will explore how schema markup factors into modern retrieval pipelines..." Pattern that wins citations: > "Schema markup contributes to AI engine retrieval in three specific ways: it disambiguates entities, it surfaces structured passages for direct extraction, and it signals page type so the engine can match query intent. Of the three, entity disambiguation is the highest-leverage..." The second version is liftable. The first is not. Across 80 client pages we audited in Q1 2026, replacing throat-clearing intros with answer-first openings raised citation rates an average of 22 percent within 60 days, controlling for other factors.Tactic 3: Information Gain (Original Data, Frameworks, Named Quotes)
Google's 2020 patent on Information Gain scored documents on how much new information they added beyond what the user had already seen on prior search results. AI engines apply a similar logic at the synthesis layer: regurgitated content, even if well-written, is poor citation material because it is redundant with the rest of the candidate set. Three reliable ways to manufacture information gain:- Original data, even small. "We audited 50 Singapore SaaS sites in Q1 2026 and found 38 had no Article schema." That single sentence is more citable than 500 words of generic schema explainer.
- Named frameworks. A memorable, named framework (like the 4-engine baseline test from our multi-engine playbook) gives engines a unit of content to attribute. Frameworks travel.
- Quotes from named experts. A single sentence quote from a named expert with a verifiable role lifts citations more than three paragraphs of analysis. Pull from interviews, podcasts, or direct outreach.
Tactic 4: Author E-E-A-T Signals
E-E-A-T is the Quality Rater framework Google uses to score human evaluation of pages, and we cover the full mechanism in our 2026 E-E-A-T deep-dive. For GEO purposes, the relevant subset is author-level signals. AI engines disproportionately cite content with strong, verifiable author attribution. The retrieval pipeline runs a quick trust scoring pass on candidate pages, and unauthored content (or content with only a brand byline) is downweighted relative to expert-authored content on the same query. The author E-E-A-T checklist:- Visible byline on every article, linking to a full author page.
- Author page with credentials, professional history, sameAs links to LinkedIn (mandatory), Wikipedia (where eligible), ORCID, X, GitHub, or other verifiable platforms.
- Person schema on the author page with `jobTitle`, `worksFor`, `alumniOf`, `sameAs`, `knowsAbout`.
- Article schema on each post with `author` referencing the author entity.
- First-person experience signals in the prose itself: "we ran this audit", "in our 12 client engagements", "I observed".
Tactic 5: Topical Cluster Depth
A single optimised page rarely gets cited consistently. A pillar of 12 to 20 interlinked pages on the same topic gets cited routinely. AI engines weight domain-level topical authority heavily because it reduces the risk of citing a one-off article from an otherwise irrelevant site."What is X" definitional
"How X works" mechanism
"X vs Y" comparison
"How to do X step by step"
"X tools compared"
"X frameworks"
Niche use case
Industry vertical
Geographic angle
The architecture rules:
- Every cluster post links to the pillar page using the primary keyword as anchor (1 link).
- Every cluster post links to 2 to 3 sibling cluster posts using descriptive anchors.
- The pillar page lists every cluster post in a "Related reading" or sub-navigation section.
- Schema-wise, the pillar uses `mainEntityOfPage` Article and lists cluster posts as `hasPart`. Cluster posts reference the pillar as `isPartOf`.
We have written more on the strategic architecture of pillar clusters as part of our content strategy service. The point for GEO: depth compounds. A site with three 20-page clusters out-cites a site with sixty unrelated articles.
Tactic 6: AI Crawler Accessibility
This is the lowest-effort, highest-yield tactic on this list. Most sites we audit have at least one AI crawler accidentally blocked, usually because a default Nginx configuration, a CDN security rule, or an over-eager robots.txt edit cut off PerplexityBot, GPTBot, or ClaudeBot. We covered the full crawler matrix in our AI crawlers reference, but the GEO-relevant config is:
```
User-agent: GPTBot
Allow: /
User-agent: ChatGPT-User
Allow: /
User-agent: PerplexityBot
Allow: /
User-agent: ClaudeBot
Allow: /
User-agent: Claude-Web
Allow: /
User-agent: CCBot
Allow: /
```
Audit this on every site quarterly. A blocked crawler is the difference between zero citations and full citation eligibility on that engine. Cloudflare's "Block AI Bots" toggle, introduced in mid-2024, is now a default-on setting for many new accounts and frequently the culprit when we diagnose sudden Perplexity drop-offs.
Tactic 7: Brand Mention Building in Trusted Publications
AI engines weight authority signals from trusted publications heavily, and this is most pronounced in Claude (which favours Reuters, BBC, established trade press) and Google AI Overviews (which inherits Google's E-E-A-T scoring).
The mechanism is simple: when an AI engine evaluates whether to cite your page on a query, it cross-references whether your brand is mentioned in trusted external sources for the same topic. A page that says "we are a leader in X" with no third-party confirmation is downweighted relative to a page from a brand that has been independently named in trade press as a leader in X.
The brand mention building stack, by leverage:
- Tier 1: Reuters, Bloomberg, mainstream business press, government publications. Single mention here moves the needle for months.
- Tier 2: Established trade publications (Search Engine Land, MarTech, sector-specific journals). Routine mentions compound.
- Tier 3: Industry podcasts with transcripts, conference speaker listings, named research datasets. Slow burn but cumulative.
- Tier 4: G2, Capterra, Crunchbase profiles with active customer review velocity. Foundational.
A practical Singapore-specific note: local trade press (The Straits Times Business, Tech in Asia, e27) is undervalued by most agencies but disproportionately well-indexed by Perplexity and Google AI Overviews on SG-specific queries. Authority-building work at this level produces both classical link equity and GEO citation lift from the same effort.
Tactic 8: Multi-Engine Prompt Testing Baseline
You cannot optimise what you do not measure. The discipline that separates GEO programmes that compound from GEO programmes that drift is a stable, quarterly multi-engine prompt test.
Define a stable query set
Pick 15 to 25 buyer-intent queries spanning category-defining ("what is X"), comparison ("X vs Y"), tactical ("how to do X"), and brand-adjacent ("best X for [vertical]") prompts. This set does not change quarter to quarter.
Run in 4 engines
ChatGPT Search, Perplexity, Google AI Overviews, Claude with web search. Same wording, fresh sessions, no chat history bias. Capture screenshots and full source lists per query.
Score citation status
For each of the 60 to 100 cells (queries x engines), record: cited (full source), mentioned (named in answer text without citation), paraphrased (concept attributed loosely), or absent. Track competitor citations in the same cells.
Compute Share of Model
Share of Model (SoM) is the GEO equivalent of share of voice: your brand citations divided by total brand citations across the query set. Track per engine and aggregate.
Re-run quarterly, course-correct
The quarterly delta tells you which tactics moved which engines. Add 2 to 3 new queries per quarter for emerging buyer language, but keep the original set stable for trend-line continuity.
For agencies and in-house teams under SGD 5,000 per month tooling budget, manual prompt testing on a stable query set still produces the highest signal-to-noise ratio. Above that budget, Profound (enterprise multi-engine), Athena HQ (AI Overviews focused), and Otterly (lightweight) are the maturing tools as of Q2 2026.
Tactic 9: Schema Markup Beyond the Basics
Most sites stop at Article and Organization schema. AI engines reward sites that go further, particularly with schema types that mark up specific extractable content units.
The advanced schema stack for GEO:
- FAQPage: Already a strong AI Overviews signal. Mark up genuine FAQ sections with at least three Q+A pairs. Each Q+A is independently citable.
- HowTo: For step-by-step content, HowTo schema lets engines extract individual steps as ordered citations. Particularly valuable for tutorial content.
- Speakable (BETA): Marks passages suited for voice assistant playback. Currently more relevant for voice-led AI engines, but expected to expand.
- Article + ItemList stacking: For "Top N" listicle content (which dominates AI citations per recent analyses), stacking Article schema with ItemList containing each entry as a ListItem produces structured retrieval gains.
- Product + AggregateRating + Review: For ecommerce, the Product/Review/Rating triplet drives citation in shopping-intent queries on ChatGPT and Perplexity.
- Person on every author bio: Closes the E-E-A-T loop from Tactic 4.
For a deeper treatment of schema strategy, our schema markup types reference covers the full taxonomy. The GEO-relevant insight is that triple-stacking schema (Article + ItemList + FAQPage) on a single page is now standard practice for citation-eligible content, not the over-optimisation it would have been treated as five years ago.
Worked example for an ecommerce category page: Stack Product schema for each item, ItemList wrapping the full collection in displayed order, FAQPage for the genuine buyer-question section at the bottom, and BreadcrumbList for navigation context. Run the result through Google's Rich Results Test and Schema.org validator before deploying. The same page is now eligible for Product cards, ItemList expansion, FAQ rich snippets, and AI engine extraction of any of the marked-up units. We have measured citation lift on Perplexity shopping-intent queries of 18 to 24 percent on properly stacked category pages compared to single-Article schema versions, holding other variables constant.
A final principle on schema for GEO: validity beats ambition. A clean, validated FAQPage and Article stack outperforms a sprawling, partially broken schema graph every time. Run the Rich Results Test on every page that targets AI citations, fix every warning, and treat schema as production-grade infrastructure rather than a content afterthought. Keyword research at the topical-cluster level identifies which pages deserve the schema investment first.
Frequently Asked Questions
Which GEO tactic should I deploy first?
Audit your robots.txt and CDN security rules for AI crawler access (Tactic 6). It takes one sitting and is the difference between zero citations and full eligibility on Perplexity and Claude. Then run a baseline prompt test (Tactic 8) so you can prove later lifts. Only then start the content and schema work, because without a baseline you cannot attribute future gains to specific tactics.
How long until GEO tactics show measurable lift?
Tactic 6 (crawler access) shows lift within 7 to 14 days as engines re-crawl. Tactic 2 (answer-first structure) and Tactic 9 (schema) typically show lift within 30 to 60 days on the rewritten pages. Tactic 1 (entity work) and Tactic 5 (topical depth) compound over 90 to 180 days. Tactic 7 (brand mentions) is the slowest, often six to twelve months for measurable Share of Model gains.
Is GEO replacing SEO?
No. GEO and SEO are complementary, with roughly 70 percent signal overlap. Strong technical SEO (crawl, index, schema, page speed, internal linking) is the foundation that GEO compounds on. The 30 percent that differs (entity optimisation depth, advanced schema stacking, multi-engine prompt baselining, AI crawler access) is the GEO-specific layer. Sites that try to GEO-optimise without a sound SEO base see flat results.
What is Share of Model and how do I track it?
Share of Model (SoM) is the GEO equivalent of share of voice: the percentage of AI engine citations on a defined query set that name your brand, divided by total citations across all brands in the same set. Track manually with a quarterly prompt test (Tactic 8) for query sets under 25 prompts. Above that volume, Profound, Athena HQ or Otterly automate the measurement.
Do I need to publish more content for GEO to work?
Volume alone does not move citations. Depth does. Three 20-page topical clusters out-perform sixty unrelated articles. The Princeton research is clear that source-level signals (citing sources, statistics, quotations) and structural signals (answer-first H2, schema) drive lift more than raw publishing velocity. Build cluster depth on the topics where you have genuine expertise, not breadth across topics where you do not.
How does GEO interact with E-E-A-T?
E-E-A-T is the underlying trust framework that AI engines apply when scoring candidate sources. Tactic 4 (author E-E-A-T) directly addresses the Expertise and Experience dimensions. Tactic 7 (brand mentions) addresses Authoritativeness. Tactic 1 (entity optimisation) addresses the disambiguation needed for Trust scoring to even apply. We cover the full E-E-A-T mechanism in our 2026 E-E-A-T deep-dive.