generative engine optimization (GEO): the complete guide, and how to grade every change.

Before 2023, ranking on page one of Google was the primary goal of content marketing. That model worked because Google sent traffic to ranked pages through blue links. AI search breaks this loop.

When a user asks ChatGPT "what is the best way to optimize for AI search engines", the model doesn't return ten links. It writes a synthesised answer and, if citations are on, selects two to four sources to attribute. GEO is the work of making your pages those sources.

The term was coined in the academic paper “GEO: Generative Engine Optimization” (Aggarwal et al., 2023) and has since become the standard industry label for the discipline.

• 01

  Citable definitions

  Lead every key concept with a crisp, one-sentence definition. AI models lift the first definition they can quote verbatim. If your page defines a term better than anyone else, it becomes the default citation.

  • Start h2/h3 sections with a definitional sentence
  • Use explicit "X is Y" or "X means Y" phrasing
  • Keep definitions under 40 words so they fit cleanly into AI answers
• 02

  Semantic HTML5 structure

  AI crawlers parse document structure the way screen readers do. A single <h1>, a logical <h2>/<h3> hierarchy inside an <article> element and consistent nesting help models extract clean outlines.

  • One <h1> per page: your primary keyword phrase
  • Wrap body content in <article>, with itemScope/itemType if possible
  • Avoid heading skips (h1 → h3): the hierarchy must be linear
• 03

  Author authority signals

  AI models weight sources by author credibility, not just domain authority. Name real authors, link author profiles and include credentials or track records. Anonymous "team" posts lose to bylined expert pieces.

  • Add author schema (Person) with name, jobTitle and sameAs LinkedIn
  • Link author bio pages from every article
  • Include firsthand data, original research or case studies
• 04

  Entity clarity

  Models build knowledge graphs. Your brand, product and key concepts should be named consistently and unambiguously on every page. Avoid pronouns where proper nouns would be clearer.

  • Use your full brand name, not just "we" or "the platform"
  • Define acronyms on first use, even if obvious to people
  • Add Organization schema with sameAs LinkedIn and X
• 05

  Structured data (schema.org)

  JSON-LD is the cleanest signal you can send to both Google and AI crawlers. Article, BreadcrumbList and Organization schemas add machine-readable metadata that survives parsing and chunking.

  • Article schema: headline, author, publisher, datePublished, dateModified
  • BreadcrumbList for site hierarchy
  • Google limited FAQ rich results in 2023: prefer Article with Q&A prose
• 06

  Canonical + indexability

  A cited source that can't be crawled loses its authority signal. Canonical tags prevent duplicate-content dilution; correct robots.txt and noindex usage keep your best pages indexable.

  • Every page: <link rel="canonical" href="https://yoursite.com/this-page" />
  • No noindex on high-value GEO targets
  • Open Graph and Twitter Card tags for social signals

1. 01

   Allow the AI crawlers you want cited by

   Check robots.txt for the search and answer agents: OAI-SearchBot and ChatGPT-User (ChatGPT), Claude-SearchBot and Claude-User (Claude), PerplexityBot and Perplexity-User (Perplexity). GPTBot, ClaudeBot and Google-Extended govern model training, a separate decision.
2. 02

   Check your CDN and firewall too

   A bot-protection rule can block an AI crawler that robots.txt allows. Look for 403s to those user agents in your server or CDN logs.
3. 03

   Put the answer in the raw HTML

   Fetch the page with curl, no JavaScript, and search for your key sentence. Many AI fetchers do not run scripts, so text rendered only on the client may never be seen.
4. 04

   One H1, question-shaped H2s

   Use the questions buyers actually ask as H2s and answer each in the first 40 to 80 words below it.
5. 05

   Structured data that matches the page

   Organization, Article with datePublished and dateModified, BreadcrumbList, and FAQPage only where the questions are visible on the page.
6. 06

   A self-referencing canonical and a current sitemap

   Every indexable page canonicalises to itself and is listed in sitemap.xml with an honest lastmod.
7. 07

   Visible dates and authors

   A "last updated" line and a named author help engines judge freshness and who stands behind a claim.
8. 08

   An llms.txt, as an extra

   A plain-text map of your most useful pages at /llms.txt. Some agents read it; Google says AI features need no special files, so treat it as a convenience, not a ranking factor.
9. 09

   Facts with sources

   Numbers, prices and limits stated plainly, each with where it comes from. Engines quote passages that stand on their own.
10. 10

    Measure citations, then change one thing

    Run the same buyer questions through ChatGPT, Claude, Gemini and Perplexity on a schedule, ship one change, and check again.

Google’s own guidance: AI Overviews and AI Mode use the same eligibility as Search, with no extra technical requirements beyond being indexable with a snippet. See AI features and your website on Google Search Central.

• Perplexity AI

  Real-time web search. Selects top results for the query, chunks the pages and cites the most directly relevant passages. Quick to pick up new content: a page indexed today can be cited within hours.

• ChatGPT (search)

  Retrieves top pages from a web index, parses them and selects excerpts to quote. Prioritises structured, skimmable content: clean HTML with clear headings outperforms dense prose.

• Google AI Overviews

  Draws from Google's existing search index, so ranking position and AI Overview citation correlate strongly. Getting onto page one for a query is still the primary path to being cited in Google's AI answer.

The common thread: all three favour content that is well-structured, explicitly authoritative and easy to chunk into quotable units. The six levers map directly onto those preferences.

1. 01

   Month 1 — Foundation & audit

   Week 1–2

   Baseline audit

   • Export Search Console: top 50 queries by impressions, filtered to position 4–20 (the best GEO opportunity zone)
   • Run an AI citation check: search your brand and top topics in ChatGPT, Perplexity and Gemini, and document which competitors get cited
   • Audit schema: validate Article, BreadcrumbList and Organization on your top 10 pages with the schema.org validator
   • Check canonical tags: every public page should have one; missing canonicals dilute authority

   Week 3–4

   Technical fixes

   • Add Article JSON-LD to all blog posts and guides (headline, author, publisher, datePublished)
   • Add BreadcrumbList to every page but the homepage
   • Fix heading hierarchy issues (h1 → h3 skips)
   • Put author name, jobTitle and LinkedIn URL in every author schema
2. 02

   Month 2 — Content upgrade

   Week 5–6

   Definition-first rewrites

   • Pick the 5 most-searched terms that LLMs answer (from your citation audit)
   • Rewrite the opening paragraph of each page to lead with a citable, one-sentence definition
   • Add an explicit step-by-step section to any page targeting "how to" queries
   • Replace generic "team" authorship with a named person, bio link and credentials

   Week 7–8

   Content gap fill

   • Build 2–3 pillar definition pages for topics you rank for but don't own in AI answers
   • Find keywords at position 5–15 with high impressions (ask Cogny over your Search Console data)
   • Structure each page: definition → why it matters → how to implement → common mistakes → related topics
   • Link the pillar pages to each other to form a semantic cluster
3. 03

   Month 3 — Authority & measurement

   Week 9–10

   Authority building

   • Publish one original data piece (survey, benchmark or analysis): AI models cite primary sources heavily
   • Add firsthand case studies: named client, real numbers, before/after metrics
   • Get at least 3 external publications to link to your pillar pages
   • Submit an updated sitemap and request indexing of rewritten pages with URL Inspection

   Week 11–12

   Measurement & iteration

   • Re-run the AI citation check across ChatGPT, Perplexity and Gemini and measure the delta against the Week 1 baseline
   • Compare CTR and average position on rewritten pages against the 30 days before
   • Set up ongoing monitoring: Cogny's GEO Conversion Report tracks AI-referral traffic in GA4 or BigQuery
   • Pick the next 5 topics for Month 4 from the remaining citation gaps

• GEO & Search Visibility AnalysisPulls Search Console data, finds pages in the citation opportunity zone (positions 5–15, high impressions) and proposes structured content fixes.
• GEO Conversion ReportQueries GA4 or BigQuery for real AI-engine referral sessions (perplexity.ai, chatgpt.com and others), so you measure citation ROI, not just impressions.
• AI Search Optimization AuditChecks schema markup, canonicals, heading hierarchy and author signals, then outputs a fix list sorted by citation impact.