# GEOhandbook — full text > GEOhandbook is a practical, source-cited guide to Generative Engine Optimization (GEO) — how to earn citations in ChatGPT, Perplexity, Google AI Overviews, Claude, and Copilot. For marketers, webmasters, and content teams. This file contains the complete prose of every page on geohandbook.io, concatenated in reading order: core guides first, then articles newest first, then site information. Each page begins with an H1, its canonical URL, and where applicable its publication date. Inline links are preserved because every non-obvious factual claim on this site is cited to a primary source; follow them to verify any statement. Index version of this file: https://geohandbook.io/llms.txt Generated from source on build. Pages: 32. ------------------------------------------------------------------------ # The GEO Guide URL: https://geohandbook.io/guide/ Summary: A definitive guide to Generative Engine Optimization (GEO) — what it is, why it matters in 2026, how it differs from SEO, the eight pillars, and how citation selection works inside ChatGPT, Perplexity, and Google AI Overviews. ## What is Generative Engine Optimization? Generative Engine Optimization (GEO) is the practice of structuring digital content and managing online presence so that AI-powered answer engines — ChatGPT, Google AI Overviews, Perplexity, Claude and Microsoft Copilot — retrieve, cite, recommend and quote your content when responding to users. Where traditional SEO fights for a ranked link on a search results page, GEO fights to be the trusted source an assistant paraphrases in its answer ([Wikipedia](https://en.wikipedia.org/wiki/Generative_engine_optimization), [Search Engine Land](https://searchengineland.com/what-is-generative-engine-optimization-geo-444418)). The term was coined in a Princeton study and has since collected several near-synonyms — Answer Engine Optimization (AEO), Large Language Model Optimization (LLMO), AI Optimization (AIO), AI SEO — with no consensus definition distinguishing them ([Wikipedia](https://en.wikipedia.org/wiki/Generative_engine_optimization)). We use "GEO" throughout this handbook because it is the term Search Engine Land, Coursera, Seobility and most 2026 industry coverage have converged on. ## Why GEO matters in 2026 The center of gravity in search is shifting. According to Search Engine Land's 2026 guide, Google's AI Overviews now reach more than **2 billion monthly users** and ChatGPT serves **800 million users each week**. Gartner projected traditional search volume will drop **25% in 2026** as users move to answer engines, with AI search on track to reach 50% of query share by 2028 ([Search Engine Land](https://searchengineland.com/mastering-generative-engine-optimization-in-2026-full-guide-469142), [Onely](https://www.onely.com/blog/geo-vs-seo/)). The click economics have already collapsed. Onely's analysis reports the overall zero-click rate rose from 57–60% in 2023 to **68–72% in 2026**, and on queries where AI Overviews appear the click-through rate to any website falls to **17%**. Position-one CTR has dropped from 7.6% to 1.6% — a 79% decline. Some publishers have absorbed brutal traffic hits: HubSpot 70–80%, Business Insider 55%, Chegg 49% ([Onely](https://www.onely.com/blog/geo-vs-seo/)). The upside for those who adapt is real. Onely's data shows GEO-driven traffic converting at **27%** versus **2.1%** for traditional SEO — a 12.9× improvement — and Webflow reports ChatGPT-referred visitors converting at 24% versus 4% from Google ([Onely](https://www.onely.com/blog/geo-vs-seo/)).

**Bottom line.** GEO is not optional. If you sell to anyone who uses ChatGPT, Perplexity or Google, your content is already being read by AI systems that decide whether to quote you or your competitor. The only question is whether you optimize for that or not.

## How GEO differs from SEO Google's own 2026 documentation states that "optimizing for generative AI search is optimizing for the search experience, and thus still SEO" ([Google Search Central](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide), [Wikipedia](https://en.wikipedia.org/wiki/Generative_engine_optimization)). That framing is technically defensible — but it hides how much of the day-to-day work is different. | Dimension | Traditional SEO | GEO | | --- | --- | --- | | Primary goal | Rank in the SERP | Be cited in an AI answer | | Success metric | Rankings, sessions, CTR | Citation frequency, attribution rate, share of voice | | Content unit | Full page | Extractable 60–100 word chunk | | Technical requirement | Crawlability, Core Web Vitals | Same + JS-free rendering, richer schema | | Time to signal | 3–6 months | Initial impact within 30 days | | Owned vs earned | ~50/50 mix | **89% of citations come from earned sources**, 23% from owned sites | Source: [Onely, GEO vs SEO 2026](https://www.onely.com/blog/geo-vs-seo/); [Progress Sitefinity](https://www.progress.com/blogs/seo-and-geo-guide). Two implications matter. First, the same page can rank #1 and never get cited — because your paragraph structure buries the answer. Second, off-site mentions (Reddit threads, industry PR, expert quotes on third-party sites) matter far more than for classical SEO, because retrieval-augmented models weight earned media heavily. ## How citation selection actually works When an answer engine responds to a query it runs four steps: query understanding → retrieval → synthesis → optional citation. During retrieval it pulls candidate chunks from many sources (its own index, live web fetches, RAG systems); during synthesis it composes an answer and picks two to seven domains to attribute ([Search Engine Land](https://searchengineland.com/mastering-generative-engine-optimization-in-2026-full-guide-469142)). Analyses of live AI Overviews suggest three consistent signals: 1. **Extractability.** The cited snippet almost always appears near the top of its source page ([CXL](https://cxl.com/blog/google-ai-overview-citation-sources/)). 2. **Authority.** E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) matters more for AI citations than for classical rankings ([Stackmatix](https://www.stackmatix.com/blog/ai-overview-citation-analysis)). 3. **Overlap with organic.** 76–86% of AI-cited sources already rank in the top 10 of traditional search, and 52% of Google AI Overview sources come from the top 10 SERP ([Onely](https://www.onely.com/blog/geo-vs-seo/)). The takeaway is uncomfortable but honest: classical SEO is the foundation of GEO. You need to rank *and* to be extractable. ## The eight pillars The rest of this handbook goes deep on each pillar. Here is the framework, in one page. 1. **Answer-first content.** Direct 2–3 sentence answer in paragraph one of every page and every H2 section. 2. **Extractable chunks.** 60–100 word paragraphs, one claim per paragraph, clean H2/H3 hierarchy with IDs. 3. **Structured data.** Valid JSON-LD for Article, FAQPage, HowTo, Organization, Person. Validate in CI. 4. **Entity clarity.** Consistent names, `sameAs` links, disambiguated products and authors. Treat entities as primary keys. 5. **Earned authority.** PR, expert quotes on third-party sites, Reddit discussion, industry roundups. 89% of LLM citations come from earned media ([Onely](https://www.onely.com/blog/geo-vs-seo/)). 6. **Server-rendered HTML.** Ship core content in the initial response. Static-site generators (like the Eleventy site you are reading) win here. 7. **AI-friendly discovery.** `llms.txt` manifest, IndexNow pings, sitemap, and a `robots.txt` that welcomes GPTBot, PerplexityBot, ClaudeBot and Google-Extended. 8. **Citation measurement.** Track share of voice in AI answers, not just rankings. Use Google Search Console's AI performance reports, Bing Webmaster Tools' AI Performance report, and prompt-testing across the major engines ([Wikipedia](https://en.wikipedia.org/wiki/Generative_engine_optimization)). ## Where to go next - **[The GEO Playbook](/playbook/):** the step-by-step implementation plan built around the eight pillars. - **[The GEO Checklist](/checklist/):** a copy-and-run technical checklist for developers and webmasters. - **[Articles](/articles/):** deep-dives on individual tactics. - **[Glossary](/glossary/):** plain-English definitions of GEO, AEO, LLMO, AIO, RAG, entities, and llms.txt. ## Sources - Search Engine Land, [Mastering GEO in 2026 — full guide](https://searchengineland.com/mastering-generative-engine-optimization-in-2026-full-guide-469142). - Onely, [GEO vs SEO: What's the difference in 2026?](https://www.onely.com/blog/geo-vs-seo/). - Wikipedia, [Generative engine optimization](https://en.wikipedia.org/wiki/Generative_engine_optimization). - Google Search Central, [Optimizing for generative AI features on Google Search](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide). - Progress Sitefinity, [SEO and GEO — a practical guide for 2026](https://www.progress.com/blogs/seo-and-geo-guide). - CXL, [Where Google AI Overviews cite from](https://cxl.com/blog/google-ai-overview-citation-sources/). - Stackmatix, [AI Overview citation analysis](https://www.stackmatix.com/blog/ai-overview-citation-analysis). ------------------------------------------------------------------------ # The GEO Playbook URL: https://geohandbook.io/playbook/ Summary: Step-by-step GEO implementation playbook for marketers and webmasters — audit, restructure, add schema, earn authority, and measure citations. Most GEO advice reads like a laundry list. This playbook sequences the work into five phases, each with a concrete deliverable, so you can hand it to a marketing team, a developer and an editor and get moving. ## Phase 1 — Audit (week 1) Before you change anything, benchmark. You cannot claim GEO wins later without a baseline citation count. **Deliverables** - **Prompt panel.** Draft 20–40 real prompts a customer would type into ChatGPT and Perplexity. Include: brand queries ("what is X?"), category queries ("best CRM for real estate agents"), comparison queries ("X vs Y") and problem queries ("how do I…"). - **Baseline citation audit.** Run every prompt in ChatGPT, Perplexity, Google AI Overviews, Claude and Copilot. For each response record: did your domain appear? which URL? which competitors got cited? - **Content-extractability scan.** Take your ten most important pages and check: is the direct answer in the first paragraph? Are subsections 60–100 words? Do H2/H3 headings use plain question phrasing? - **Technical baseline.** Confirm the site renders core content server-side (view source, disable JavaScript), that `robots.txt` does not block AI crawlers, and that JSON-LD validates. Onely's data suggests measurable citation impact is possible **within 30 days** of implementation, so a real baseline is worth the week ([Onely](https://www.onely.com/blog/geo-vs-seo/)). ## Phase 2 — Restructure content (weeks 2–4) The single highest-leverage change is rewriting your top 10–20 pages for extraction. **The rewrite pattern** For every page and every H2 section: 1. Open with a **2–3 sentence direct answer** to the section's question. 2. Follow with a **60–100 word supporting paragraph** — one clear claim, cited to a primary source with an inline link. 3. Add examples, data or nuance in subsequent paragraphs. 4. End the section with a one-line takeaway. This is the "independently extractable unit" pattern documented by Shadow, Contently and Moxie Digital: short fact-dense chunks retrieve more reliably than long narrative passages ([Shadow](https://www.shadow.inc/resources/how-to-optimize-content-for-perplexity-and-chatgpt), [Contently](https://contently.com/2026/02/20/how-to-optimize-content-for-perplexity-ai/), [Moxie Digital](https://www.moxie-digital.com/blog/optimize-content-for-chatgpt-perplexity)). **Ancillary content moves** - Add an FAQ section to every high-intent page. FAQPage schema is one of the most reliably cited formats. - Publish comparison pages ("X vs Y", "best 5 tools for Z") — AI answers love them. - Update the year in headlines and add a visible "Last updated" date. LLMs prefer recent, sourced material ([Contently](https://contently.com/2026/02/20/how-to-optimize-content-for-perplexity-ai/)). ## Phase 3 — Technical GEO (weeks 3–5) Runs in parallel with the content rewrite. **Schema.** Add valid JSON-LD to every page: `Article` on posts, `FAQPage` on FAQs, `HowTo` on procedural content, `Organization` on the site root, `Person` on author bios. Validate with Google's [Rich Results Test](https://search.google.com/test/rich-results) and Schema.org's [validator](https://validator.schema.org/). Fail your CI build if schema is invalid ([Strapi](https://strapi.io/blog/generative-engine-optimization-geo-guide)). **Rendering.** If you use a heavy client-side framework, switch to server-side rendering or a static-site generator for content pages. Most AI crawlers don't execute JavaScript reliably. **Discovery.** Ship four files: - `robots.txt` that explicitly allows `GPTBot`, `PerplexityBot`, `OAI-SearchBot`, `Google-Extended`, `ClaudeBot` and `CCBot`. - `sitemap.xml` regenerated on every build. - `llms.txt` at your domain root — a Markdown manifest pointing AI systems at your canonical pages. Google has said `llms.txt` alone has no measurable effect ([Contentful](https://www.contentful.com/blog/llms-txt-search-visibility/)), but it costs nothing to ship and other engines are starting to consume it. - IndexNow pings on publish and update ([Bing IndexNow](https://www.bing.com/indexnow)). **Entity graph.** Add `sameAs` links from your `Organization` schema to your Wikipedia entry (if you have one), Crunchbase, LinkedIn, and Wikidata. Do the same on `Person` schema for authors. Consistent entity references are one of the signals AI systems use to disambiguate brands ([eSEOspace](https://eseospace.com/blog/key-geo-ranking-factors-explained/)). ## Phase 4 — Earn off-site authority (ongoing from week 3) This is the phase most in-house teams underweight — and it is the phase that moves the needle most. Onely's analysis found **89% of LLM citations come from earned sources** (news, PR, Reddit, forums, expert quotes on third-party sites) versus **23% from owned websites** ([Onely](https://www.onely.com/blog/geo-vs-seo/)). If you only optimize your own site, you are competing for 23% of the pie. **High-leverage earned tactics** - **Digital PR.** Pitch data-driven angles to journalists in your category. A single link from a mainstream outlet often outweighs 20 owned pages. - **Reddit and community presence.** LLMs draw heavily on Reddit for opinion and comparison queries. Answer questions in the subreddits your customers use, honestly and without spam. - **Expert quotes.** Get your subject-matter experts quoted in industry roundups, "expert says" articles and podcast transcripts. - **HARO / Qwoted / Featured.** Answer journalist queries daily. - **Wikipedia entity.** If your organization is genuinely notable, ensure the Wikipedia entry is accurate and well-sourced. ## Phase 5 — Measure and iterate (week 5+) Rankings alone lie now. Measure GEO with GEO metrics. **Core metrics** | Metric | How to measure | | --- | --- | | AI citation rate | % of your prompt panel where your domain is cited. Rerun monthly. | | Share of voice | Your citations ÷ total citations across all responses for a prompt category. | | Referral traffic | Direct traffic from `chat.openai.com`, `perplexity.ai`, `gemini.google.com` in GA4. | | AI Performance report | Bing Webmaster Tools' AI Performance report; Google Search Console's Search Generative AI report ([Wikipedia](https://en.wikipedia.org/wiki/Generative_engine_optimization)). | | Schema-validation pass rate | % of pages passing JSON-LD validation in CI. | **Toolkit worth trying** - [Google Search Console](https://search.google.com/search-console/about) — includes AI Overview appearance data as it rolls out. - [Bing Webmaster Tools AI Performance report](https://www.bing.com/webmasters). - [Ahrefs Brand Radar](https://ahrefs.com/brand-radar) and similar AI-mention monitoring products. - Manual prompt testing on a monthly cadence. Do not skip this — automated tools miss nuance.

**A realistic timeline.** Onely reports "initial impact within 30 days" for GEO work, versus 3–6 months for SEO. That does not mean citation dominance in a month — it means you should see your first new citations by day 30 if the work is done well ([Onely](https://www.onely.com/blog/geo-vs-seo/)).

## Ninety-day plan on one page - **Weeks 1:** Audit prompts, baseline citations, technical scan. - **Weeks 2–4:** Rewrite top 10 pages using the answer-first + 60–100 word chunk pattern. Add FAQ blocks. - **Weeks 3–5:** Ship JSON-LD, llms.txt, IndexNow, `robots.txt` updates, entity `sameAs` links. - **Weeks 3–12:** Digital PR sprint. 8 pitches per week. Community answers three times a week. - **Week 5 onward:** Monthly prompt-panel rerun. Track citation share by category. ## Sources - Onely, [GEO vs SEO — Onely, 2026](https://www.onely.com/blog/geo-vs-seo/). - Search Engine Land, [Mastering GEO in 2026](https://searchengineland.com/mastering-generative-engine-optimization-in-2026-full-guide-469142). - Shadow, [How to optimize content for Perplexity and ChatGPT](https://www.shadow.inc/resources/how-to-optimize-content-for-perplexity-and-chatgpt). - Contently, [Content optimization for Perplexity AI 2026](https://contently.com/2026/02/20/how-to-optimize-content-for-perplexity-ai/). - Strapi, [Generative engine optimization — complete guide](https://strapi.io/blog/generative-engine-optimization-geo-guide). - eSEOspace, [Key GEO ranking factors](https://eseospace.com/blog/key-geo-ranking-factors-explained/). - Contentful, [Do llms.txt files improve AI search visibility?](https://www.contentful.com/blog/llms-txt-search-visibility/). ------------------------------------------------------------------------ # The GEO Checklist URL: https://geohandbook.io/checklist/ Summary: A practical, source-cited GEO checklist covering content, structured data, rendering, discovery, entities, off-site authority, and measurement. Below is the full GEO checklist — the technical items are drawn from Strapi's implementation guide and Onely's 2026 breakdown; the editorial items are informed by Shadow, Contently and Moxie Digital's citation-analysis work. ## Content & markup - [ ] Every page opens with a **2–3 sentence direct answer** in paragraph one. - [ ] Every H2 section opens with a 2–3 sentence direct answer. - [ ] Body paragraphs are **60–100 words**, one clear claim each. - [ ] Headings use **plain question phrasing** ("What is …?", "How do I …?"). - [ ] `

` exactly once per page; `h2` → `h3` never skips levels. - [ ] Subsection headings have explicit `id` attributes for deep-linking. - [ ] Uses semantic HTML5: `
`, `
`, `
`, `
`, `
`. - [ ] Author byline visible, linked to a bio page with `Person` schema. - [ ] "Last updated" date visible on every content page. - [ ] FAQ block near the end of every high-intent page. - [ ] Data-backed statistics with inline links to primary sources. - [ ] Comparison content ("X vs Y", "best 5 …") where relevant. - [ ] Descriptive internal link anchors — never "click here" or "read more". Source: [Shadow](https://www.shadow.inc/resources/how-to-optimize-content-for-perplexity-and-chatgpt), [Moxie Digital](https://www.moxie-digital.com/blog/optimize-content-for-chatgpt-perplexity), [Contently](https://contently.com/2026/02/20/how-to-optimize-content-for-perplexity-ai/). ## Structured data (JSON-LD) - [ ] `Article` schema on posts with `headline`, `author`, `datePublished`, `dateModified`, `image`, `mainEntityOfPage`. - [ ] `FAQPage` schema on FAQ sections. - [ ] `HowTo` schema on procedural content. - [ ] `Organization` schema on the homepage with `name`, `url`, `logo`, `sameAs`. - [ ] `Person` schema on author pages with `sameAs` (LinkedIn, Twitter/X, personal site). - [ ] `BreadcrumbList` on non-root pages. - [ ] JSON-LD validates on [Rich Results Test](https://search.google.com/test/rich-results) and [Schema.org validator](https://validator.schema.org/). - [ ] Schema validation runs in CI and fails the build on error. Source: [Strapi](https://strapi.io/blog/generative-engine-optimization-geo-guide), [SEOptimer](https://www.seoptimer.com/blog/schema-markup-for-ai-search/). ## Rendering & performance - [ ] Core content is present in the initial HTML response (view-source test). - [ ] Site renders correctly with JavaScript disabled. - [ ] Uses static-site generation, SSR or ISR — not client-side-only rendering. - [ ] LCP under 2.5s on mobile. - [ ] CLS under 0.1. - [ ] Images use `srcset` and modern formats (WebP/AVIF) with descriptive `alt` text. - [ ] Non-critical JavaScript is deferred. - [ ] HTTPS enforced sitewide, HSTS enabled. ## Discovery for AI crawlers - [ ] `robots.txt` explicitly `Allow`s `GPTBot`, `PerplexityBot`, `OAI-SearchBot`, `Google-Extended`, `ClaudeBot`, `CCBot`. - [ ] `sitemap.xml` regenerated on every build and linked in `robots.txt`. - [ ] `llms.txt` at the domain root with a Markdown map of canonical pages. - [ ] IndexNow ping fired on publish and content update ([bing.com/indexnow](https://www.bing.com/indexnow)). - [ ] Server logs show `GPTBot` and `PerplexityBot` requests (proof of crawl). - [ ] Google Search Console and Bing Webmaster Tools are verified for the domain. Note: Google has stated `llms.txt` alone doesn't currently affect Google Search ranking ([Contentful](https://www.contentful.com/blog/llms-txt-search-visibility/)); ship it anyway as low-cost future insurance and for engines that do consume it. ## Entities & authority signals - [ ] Every mention of your brand uses the same canonical name across pages. - [ ] `Organization.sameAs` links to Wikipedia (if you have an entry), Wikidata, Crunchbase, LinkedIn, GitHub. - [ ] Author bios link to LinkedIn, Twitter/X, personal site, and any staff page. - [ ] Consistent NAP (Name, Address, Phone) if you have a physical presence. - [ ] `Organization.description` matches the description on your About page. Source: [eSEOspace](https://eseospace.com/blog/key-geo-ranking-factors-explained/), [Visalytica](https://www.visalytica.com/blog/generative-engine-ranking). ## Off-site earned authority Reminder: 89% of LLM citations come from earned sources ([Onely](https://www.onely.com/blog/geo-vs-seo/)). This section is not optional. - [ ] Active digital-PR pipeline (target: 1–2 earned placements per month). - [ ] Journalist-query response cadence (HARO, Qwoted, Featured, Terkel). - [ ] Presence in category-relevant subreddits — helpful answers, not spam. - [ ] Expert quotes contributed to industry roundups. - [ ] Podcast appearances by named subject-matter experts. - [ ] Product/service listed in reputable third-party comparison sites. - [ ] Wikipedia entry (if genuinely notable) is accurate and well-sourced. ## Measurement - [ ] Monthly prompt-panel rerun across ChatGPT, Perplexity, Google AI Overviews, Claude, Copilot. - [ ] Citation-rate tracked by category (brand, category, comparison, problem). - [ ] Referral traffic segmented for `chat.openai.com`, `perplexity.ai`, `gemini.google.com`, `copilot.microsoft.com` in GA4. - [ ] Bing Webmaster Tools' AI Performance report reviewed monthly ([Wikipedia](https://en.wikipedia.org/wiki/Generative_engine_optimization)). - [ ] Google Search Console Search Generative AI performance report reviewed monthly. - [ ] Alerts on schema-validation failure and citation drops. ## Governance - [ ] `SECURITY.md`, `humans.txt`, and a public change-log for the site. - [ ] Content ownership defined per URL (author + editor). - [ ] Editorial disclosure page for advertising, affiliates and sponsored content. - [ ] Accessibility statement (WCAG 2.2 AA target). ## Sources - Onely, [GEO vs SEO 2026](https://www.onely.com/blog/geo-vs-seo/). - Strapi, [Generative Engine Optimization — complete guide](https://strapi.io/blog/generative-engine-optimization-geo-guide). - Search Engine Land, [Mastering GEO in 2026](https://searchengineland.com/mastering-generative-engine-optimization-in-2026-full-guide-469142). - SEOptimer, [Schema markup for AI search](https://www.seoptimer.com/blog/schema-markup-for-ai-search/). - eSEOspace, [Key GEO ranking factors](https://eseospace.com/blog/key-geo-ranking-factors-explained/). - Contentful, [Do llms.txt files improve AI search visibility?](https://www.contentful.com/blog/llms-txt-search-visibility/). ------------------------------------------------------------------------ # The GEO Glossary URL: https://geohandbook.io/glossary/ Summary: A plain-English glossary of GEO, Generative Engine Optimization, AEO, LLMO, AIO, RAG, entities, embeddings, llms.txt, IndexNow, and citation rate. ## GEO — Generative Engine Optimization The practice of structuring content and digital presence so AI answer engines cite, quote and recommend your brand. The term was coined in a Princeton study and is used by Search Engine Land, Coursera, Seobility and most 2026 industry coverage ([Wikipedia](https://en.wikipedia.org/wiki/Generative_engine_optimization)). ## AEO — Answer Engine Optimization A near-synonym for GEO. Some practitioners use AEO specifically for optimization aimed at answer-style responses (Google featured snippets, People Also Ask, AI Overviews) and reserve GEO for the broader generative-model surface ([Progress Sitefinity](https://www.progress.com/blogs/seo-and-geo-guide)). No academic consensus separates them ([Wikipedia](https://en.wikipedia.org/wiki/Generative_engine_optimization)). ## LLMO — Large Language Model Optimization Another synonym. Emphasizes optimization for the LLMs themselves (ChatGPT, Claude, Gemini) rather than the search-adjacent surface. Nikhil Lai of Forrester Research argued in 2025 that AEO, LLMO and related terms are "significantly, but not fundamentally, different from SEO" ([Wikipedia](https://en.wikipedia.org/wiki/Generative_engine_optimization)). ## AIO — AI Optimization The broadest catch-all. Some vendors use "AIO" to cover any optimization of digital presence for AI systems, including chatbots that aren't strictly answer engines. ## AI Overviews Google's AI-generated summaries that appear above the traditional blue links. Reach more than 2 billion monthly users and are triggered on 18% of commercial queries and 14% of transactional queries as of late 2025 ([Search Engine Land](https://searchengineland.com/mastering-generative-engine-optimization-in-2026-full-guide-469142), [Onely](https://www.onely.com/blog/geo-vs-seo/)). ## RAG — Retrieval-Augmented Generation The architecture behind most modern answer engines. When a user asks a question, the system retrieves candidate documents from an index or the live web, then a language model generates an answer grounded in those documents and optionally cites them. GEO is largely about making your content retrieve well and generate well ([Search Engine Land](https://searchengineland.com/mastering-generative-engine-optimization-in-2026-full-guide-469142)). ## Embeddings Numeric vector representations of text that let retrieval systems compare "meaning" rather than exact keywords. Answer engines embed both user queries and candidate content chunks, and pull the closest matches. Short, self-contained paragraphs embed more cleanly than sprawling ones — which is why "chunk hygiene" matters for GEO. ## Chunks The unit of retrieval. Answer engines typically split pages into 300–800 character chunks and index each chunk separately. Design your paragraphs to survive being lifted out of their surrounding context. ## Entities People, places, organizations, products and concepts that AI systems track as first-class objects. Using consistent names and `sameAs` links across pages helps engines disambiguate your brand and connect it to other authoritative sources ([eSEOspace](https://eseospace.com/blog/key-geo-ranking-factors-explained/)). ## E-E-A-T Experience, Expertise, Authoritativeness, Trustworthiness. Originally a Google search-quality framework; AI citation analyses consistently find E-E-A-T signals matter more for AI citations than for classical rankings ([Stackmatix](https://www.stackmatix.com/blog/ai-overview-citation-analysis)). ## llms.txt A proposed plain-text or Markdown file at your domain root, analogous to `robots.txt`, that offers AI systems a curated map of your important pages. Google has stated `llms.txt` alone doesn't currently affect Search ranking ([Contentful](https://www.contentful.com/blog/llms-txt-search-visibility/)); other engines are experimenting with it. ## IndexNow A push-notification protocol supported by Bing, Yandex, Seznam and others that lets you notify search engines the moment content is created or updated ([bing.com/indexnow](https://www.bing.com/indexnow)). Useful for GEO because recency is a citation signal. ## Schema.org The shared vocabulary for structured data used by Google, Microsoft, Yandex and increasingly by AI systems. Deployed as JSON-LD in a ` ``` ## Validate in CI If schema is worth adding, it is worth failing your build when it breaks. Strapi's guide walks through a GitHub Actions setup that extracts JSON-LD from rendered HTML, asserts required properties exist, and fails the build on error — with a stated pipeline overhead of under 30 seconds ([Strapi](https://strapi.io/blog/generative-engine-optimization-geo-guide)). ## Common mistakes - Schema that doesn't match the visible page. Google will demote or ignore it; AI systems will notice too. - `FAQPage` with fake questions. Only mark up genuine FAQ content. - `Person` without `sameAs`. A bare author name is a low-signal entity. Link it out. - Multiple `@id` values for the same entity across pages. Pick one canonical ID and reuse. ## Checklist - [ ] `Article` on posts with `author`, `datePublished`, `dateModified`. - [ ] `FAQPage` on FAQ sections. - [ ] `Organization` on the homepage with `sameAs`. - [ ] `Person` on author bios with `sameAs`. - [ ] JSON-LD validates on the [Rich Results Test](https://search.google.com/test/rich-results) and [Schema.org validator](https://validator.schema.org/). - [ ] Validation runs in CI on every pull request. ## Sources - [Google Search Central — Optimizing for AI features](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) - [Blck Alpaca — Schema markup for AI search engines](https://blckalpaca.at/en/knowledge-base/seo-geo/geo-generative-engine-optimization/schema-markup-for-ai-search-engines-confirmed-and-measurable) - [Whitehat SEO — Schema markup for AI search](https://whitehat-seo.co.uk/blog/schema-markup-ai-search) - [Strapi — Generative Engine Optimization guide](https://strapi.io/blog/generative-engine-optimization-geo-guide) - [SEOptimer — Schema markup for AI search](https://www.seoptimer.com/blog/schema-markup-for-ai-search/) - [Stackmatix — AI Overview citation analysis](https://www.stackmatix.com/blog/ai-overview-citation-analysis) ------------------------------------------------------------------------ # Entity-first SEO: how to make your brand exist in machine memory URL: https://geohandbook.io/articles/entity-first-seo/ Published: 2026-09-09 Summary: LLMs organize knowledge as graphs of entities. If your brand isn't a clean node in that graph, you compete with better-defined sources like Wikipedia and Wikidata — and lose. The one-line version: Modern AI answer engines don't search keywords — they resolve entities. If a page doesn't clearly define what a brand, product, or term is, LLMs default to better-defined sources like Wikidata, Wikipedia, or the top academic paper ([DataDab](https://www.datadab.com/blog/entity-first-seo-building-a-public-knowledge-graph-that-llms-can-actually-use/)). Entity-first SEO — making a brand exist as a well-formed node in the machine-readable knowledge graph — is now the foundation everything else sits on. ## What "entity" actually means An entity is anything with a definition, relationships, and a place in a knowledge universe: - A person (an executive, a domain expert) - A product (a specific SKU or SaaS module) - A concept (a methodology, a framework) - A process (a workflow, a technique) - A feature (a capability, an integration) - A company (with its industry, size, founding date, subsidiaries) - A pricing plan (with its inclusions and constraints) Anything that can be named, described, and linked to other named things is an entity. LLMs internally organize knowledge as a graph of these entities and their properties — a data structure keywords alone can't populate ([DataDab](https://www.datadab.com/blog/entity-first-seo-building-a-public-knowledge-graph-that-llms-can-actually-use/)). ## Why keywords aren't enough anymore Traditional SEO taught pages to rank for query strings. A page titled "Best CRM for small teams" tried to match a search phrase. But when Perplexity decomposes that query into sub-questions and reads candidate pages in full ([Fokal](https://www.fokal.com/ai-seo/how-perplexity-chooses-sources/)), it's not matching strings — it's asking: - What entity is being discussed on this page? - What category does that entity belong to? - What properties does it have that answer the sub-question? - Does the entity's own metadata agree with what this page says about it? If the page is a wall of prose that doesn't explicitly name the entity and its category, the model may pattern-match to a better-defined external source and cite that source's characterization instead of yours. ## The three-step entity-first process DataDab summarizes it as Define → Structure → Get cited. ### 1. Define Write an explicit, one-paragraph canonical definition for every entity that matters to your brand: - Your company (what it is, what category it belongs to, what makes it distinct) - Each product or module (the same, plus its relationship to the parent brand) - Each pricing tier (with what it includes and who it's for) - Each proprietary concept, framework or methodology you use These definitions should be identical everywhere they appear — homepage, about page, product pages, LinkedIn profile, Crunchbase, Wikipedia (if applicable), any bylined article. Inconsistent definitions split your entity in the model's internal representation. ### 2. Structure Make the definition machine-readable in three places: On your own site, in JSON-LD schema.org markup. For the company itself: ```json { "@context": "https://schema.org", "@type": "Organization", "name": "Your Brand", "url": "https://example.com", "logo": "https://example.com/logo.png", "description": "Your canonical one-paragraph definition.", "foundingDate": "2018", "sameAs": [ "https://en.wikipedia.org/wiki/Your_Brand", "https://www.wikidata.org/wiki/Q123456", "https://www.crunchbase.com/organization/your-brand", "https://www.linkedin.com/company/your-brand", "https://twitter.com/yourbrand", "https://github.com/yourbrand" ] } ``` The `sameAs` array is critical. It tells any crawler that these URLs all refer to the same entity, so the model can consolidate them into a single graph node. In Wikidata. Create or enrich your Wikidata item. The properties that matter most: - `instance of` (Q123 — the type of entity) - `industry` - `country` - `official website` - `founded` - `subsidiary` / `parent organization` - `logo image` Wikidata items are one of the most heavily-weighted sources in every major LLM's training pipeline, and they're free and fast to create if you meet notability. In your own knowledge base / documentation. For product entities, write pages that follow a predictable pattern: name, one-line summary, category, relationship to parent product, features, pricing, canonical URL. Make them linkable and cite-worthy on their own. ### 3. Get cited Get third-party sources to describe your entities using the same language you've canonicalized. Editorial coverage that says "Your Brand is a [category] built for [audience]" reinforces the entity definition every time it's crawled. Interviews with named team members reinforce their entity records. Guest articles by domain experts on your team reinforce theirs. ## The fallback problem If your brand doesn't exist as a clean entity, LLMs will fill the gap. According to DataDab, they default to: - Wikidata (if you have an item, however sparse) - Wikipedia (if you have an article) - The top academic paper on the topic - The most-linked review site in the category None of those describe your product on your terms. The result is a brand caricature — the model knows your name but not what makes you different. ## Practical checklist - [ ] Homepage has a schema.org `Organization` JSON-LD block with a `sameAs` array of at least 4 authoritative profile URLs. - [ ] Every named team member with a bio page has a `Person` JSON-LD block with a `worksFor` reference to the organization. - [ ] Every product / module has a `Product` or `Service` JSON-LD block, linked to the parent org. - [ ] Wikidata item exists and is filled out — at minimum, `instance of`, `industry`, `country`, `official website`, `founded`. - [ ] Wikipedia article exists if you meet notability. If not, at least the Wikidata item. - [ ] Canonical one-paragraph definition of the brand appears identically on: homepage, about page, LinkedIn company page, Crunchbase, Wikipedia (if applicable), and every guest-article byline bio. - [ ] Pricing and feature pages have their own schema and are linked from the main Organization entity. ## The strategic point Entity-first SEO isn't a replacement for content or link earning — it's the substrate those things sit on. Without it, your best content earns citations that reinforce a fuzzy or wrong entity in the model's memory. With it, every citation reinforces the version of your brand you want the AI to know. Related reading: [Schema for AI search](/articles/schema-for-ai-search/) covers JSON-LD patterns in more depth. [Brand mentions in training data](/articles/brand-in-training-data/) explains why entity consistency across sources matters over months and years. ------------------------------------------------------------------------ # Measuring GEO in 2026: what actually matters when 67% of AI-driven traffic is untracked URL: https://geohandbook.io/articles/measuring-what-matters/ Published: 2026-09-07 Summary: AI bots are 51%+ of global internet traffic. Two-thirds of AI-driven sessions never reach Google Analytics. Here's a measurement framework that survives those blind spots. The one-line version: Standard analytics undercounts AI-driven traffic by up to 67%, and citation share drifts by 40–59% month over month depending on the platform ([MaximusLabs, 2026](https://www.maximuslabs.ai/ai-search-101/geo/measurement)). The only defensible GEO measurement approach in 2026 is statistical sampling of citation share paired with revenue attribution downstream, not raw traffic counts. ## The measurement crisis, in numbers | Metric | Figure | Source | |---|---:|---| | AI bot share of global internet traffic (2026) | >51% | MaximusLabs | | B2B searches ending without a website click | >60% | MaximusLabs | | AI-driven traffic that goes untracked by conventional analytics | Up to 67% | MaximusLabs | | Monthly citation drift on Perplexity | 40.5% | MaximusLabs | | Monthly citation drift on Google AI Overviews | 59.3% | MaximusLabs | | AI-referred session growth (Jan–May 2025) | +527% | MaximusLabs | | Cross-platform citation overlap (ChatGPT ∩ Perplexity) | 11% of domains | MaximusLabs | | LLM-referred user conversion rate vs. standard organic | 11× | MaximusLabs | | Brand search volume correlation with LLM citation inclusion | r = 0.334 | MaximusLabs | Two implications sit inside those numbers: 1. Traffic count as the primary GEO KPI is broken. Two out of every three AI-driven visits don't identify themselves as such. GA4's default channel grouping puts them in Direct or Organic. 2. Citation share is high-variance. A monthly drift of 40–59% means point-in-time citation snapshots are unreliable. You need multiple samples per period to reach a stable signal. ## What to measure — the four-metric framework ### 1. Citation share (the primary metric) Citation share is: out of a representative sample of category prompts run through each AI engine, how often does your domain appear as a cited source, and in what position? How to sample: - Assemble a fixed list of 30–100 prompts your buyers plausibly ask, ranging from top-of-funnel category queries to bottom-of-funnel branded/comparison queries. Freeze this list — the same prompts each period. - Run each prompt through ChatGPT, Perplexity, Google AI Overviews and Gemini in fresh sessions with no memory. Log the visible citations and their positions. - Repeat monthly. Because Perplexity drifts 40.5% and AI Overviews drift 59.3% month-over-month, a single run isn't a reliable measurement — run the sample 3–5 times per period and average. How to score: - Citation rate: fraction of prompts where your domain appears at least once. - Weighted citation share: average position of your citations (higher = better; use 1/rank as a scoring function). - Share of voice vs. named competitors: your citation rate ÷ (your citations + top 5 competitors' combined). ### 2. Referral traffic from LLM engines, correctly attributed GA4's default referral tracking misses most LLM traffic because ChatGPT, Perplexity and Copilot links often arrive as Direct. Manually classify: - Perplexity referrals — `referrer contains perplexity.ai` - ChatGPT referrals — `referrer contains chat.openai.com` or `chatgpt.com` - Copilot referrals — `referrer contains copilot.microsoft.com` or `bing.com/chat` - Direct with UTM `utm_source=ai_engine` — for any campaign where you seed citations manually Create a custom channel group so these bucket cleanly. Track sessions, conversion rate, and revenue per session. Given the 11× conversion multiplier reported by [MaximusLabs](https://www.maximuslabs.ai/ai-search-101/geo/measurement), even small AI-referred traffic volumes may dominate revenue attribution. ### 3. Brand search volume as a proxy signal MaximusLabs reports a correlation coefficient of r = 0.334 between brand search volume and LLM citation inclusion. It's not causal in either direction — brand mentions grow both — but as a proxy signal for entity strength, brand-search-volume trend lines are one of the most reliable long-run measurements available. Track monthly branded impressions in Google Search Console alongside citation share. ### 4. Revenue attribution downstream Because 67% of AI-driven traffic goes untracked, session counts systematically undersell impact. The credible fallback is pipeline lift — do the deals you close self-report having heard about you through an AI tool? - Add "Where did you hear about us?" to your demo-request forms with ChatGPT, Perplexity, Gemini and Copilot as named options. - On closed deals in the CRM, tag the primary discovery source. Compare AI-source-tagged deals against your traffic-based attribution to see the gap. - Where the CRM shows 20% AI-attributed pipeline and GA4 shows 4% AI-attributed sessions, the 16-point gap is roughly the invisibility discount you're paying. ## What to stop measuring - AI Overview impression counts as a standalone metric. They correlate poorly with citation, click and conversion outcomes. - Total traffic as the north-star KPI. It's collapsing across nearly all publisher classes ([22% for large, 47% for medium, 60% for small](/articles/google-may-2026-ai-update/)) and traffic loss can happen alongside citation gain. - Any point-in-time citation snapshot from a single sample. With 40–59% monthly drift, one sample means nothing. ## The measurement stack — a practical setup | Layer | Tool option A (paid) | Tool option B (DIY) | |---|---|---| | Prompt sampling | Peec.ai, LLMrefs, Frase | Manual monthly runs logged in a Google Sheet | | Citation extraction | Peec.ai, Ahrefs Brand Radar | Playwright script that runs each prompt and parses the citation footer | | Referral classification | GA4 custom channel group | Same, plus a monthly export cross-check | | Brand mention tracking | Brand24, Mention | Google Alerts + Reddit RSS + F5Bot | | Deal-level attribution | HubSpot / Salesforce CRM fields | The same | Full comparisons of paid GEO tools are covered in [Frase's 2026 tool roundup](https://www.frase.io/blog/best-geo-tools-2026), [Licheo's tested-tool list](https://www.licheo.com/blog/best-geo-tools-ai-visibility-tracking-2026/), and [aiclicks' 17-tool review](https://aiclicks.io/blog/best-generative-engine-optimization-tools). ## The one thing that fixes the measurement problem Cross-platform citation overlap is only 11%. A brand cited on Perplexity is not the same brand cited on ChatGPT. This is the single most important operational fact in GEO measurement, and it means you cannot treat "AI citations" as one thing — you need per-engine measurement, per-engine content strategy, and per-engine tracking. Once the four-metric framework is in place — citation share, correctly-attributed referral traffic, brand search volume, and CRM-tagged pipeline — the fact that GA4 misses 67% of the traffic stops being fatal. You're measuring the outcome (revenue) and the cause (citation share) directly, and letting the noisy middle metrics stay noisy. Related reading: [Measuring GEO](/articles/measuring-geo/) is the shorter primer. [The GEO playbook](/playbook/) puts measurement into the 90-day cadence where it belongs. ------------------------------------------------------------------------ # Is llms.txt worth the effort in 2026? URL: https://geohandbook.io/articles/llms-txt-worth-it/ Published: 2026-09-05 Summary: A pragmatic 2026 take on llms.txt — what it is, what the evidence says, and why the file is still worth shipping even after Google's dismissal. Short answer: ship it anyway. Google confirmed in 2026 that `llms.txt` alone has no effect on Google Search — including its AI features ([PBN Ltd](https://pbn.ltd/google-confirms-llms-txt-does-nothing-and-what-that-says-about-2026s-seo-myths/)). Independent studies broadly agree there is no measurable citation lift from `llms.txt` alone ([AI Growth Agent](https://aigrowthagent.co/articles/ai-search-optimization-llms-txt/)). At the same time, several engines and AI agent frameworks are actively experimenting with reading it, and the cost of shipping is essentially zero. That combination — probable-zero downside, plausible-nonzero upside — makes it worth an hour of a webmaster's time. ## What llms.txt is `llms.txt` is a proposed plain-text or Markdown file at your domain root — analogous to `robots.txt` — that gives AI systems a curated map of your important pages ([Contentful](https://www.contentful.com/blog/llms-txt-search-visibility/), [Yotpo](https://www.yotpo.com/blog/what-is-llms-txt/)). Unlike `robots.txt`, which restricts access, `llms.txt` is a hint file: "here are our canonical documents, in reading order, with descriptions." The format is Markdown. A minimal file: ```md # Example Inc. > Independent editorial site on generative engine optimization. ## Core pages - [Guide](https://example.com/guide/): the definitive primer. - [Playbook](https://example.com/playbook/): implementation plan. ``` ## What the evidence actually says - Google. Says `llms.txt` does not affect Search or its AI features. Google's guidance is that indexable HTML is what matters ([Google Search Central](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide), [PBN Ltd](https://pbn.ltd/google-confirms-llms-txt-does-nothing-and-what-that-says-about-2026s-seo-myths/)). - Independent studies. No measurable citation lift from `llms.txt` used alone ([AI Growth Agent](https://aigrowthagent.co/articles/ai-search-optimization-llms-txt/)). - Contentful review. Concludes it may become useful over time but currently produces no consistent measurable impact ([Contentful](https://www.contentful.com/blog/llms-txt-search-visibility/)). - Yotpo, Brainz Digital. Advocate for it as low-cost future insurance and as documentation of your own content architecture ([Yotpo](https://www.yotpo.com/blog/what-is-llms-txt/), [Brainz Digital](https://www.brainz.digital/blog/llmtxt-optimisation-guide/)). ## The pragmatic take `llms.txt` is not going to move your citation share this month. But: 1. It costs nothing to ship if you already have a static site. 2. Newer AI-agent frameworks (LangChain-style tool-calling agents, browsing agents) increasingly do read it. 3. It doubles as a forcing function to write clean page descriptions. Ship it. Do not ship it instead of schema, `robots.txt` tuning, and content restructuring — those are the interventions with actual evidence behind them. ## What to put in yours - One-paragraph description of your site. - Canonical URLs of your five to twenty most important pages. - One-line description per page. - Optional: a `## Optional` section for supplementary pages. ## Common mistakes - Publishing the full raw content of every page. That is what your sitemap is for. `llms.txt` is a curated map. - Skipping descriptions. The one-line description is the point. - Never updating it. Regenerate on every content change, ideally at build time. ## Sources - [Contentful — Do llms.txt files improve AI search visibility?](https://www.contentful.com/blog/llms-txt-search-visibility/) - [Brainz Digital — How to optimise your llms.txt](https://www.brainz.digital/blog/llmtxt-optimisation-guide/) - [Yotpo — What is llms.txt?](https://www.yotpo.com/blog/what-is-llms-txt/) - [AI Growth Agent — llms.txt: boost or waste?](https://aigrowthagent.co/articles/ai-search-optimization-llms-txt/) - [PBN Ltd — Google confirms llms.txt does nothing](https://pbn.ltd/google-confirms-llms-txt-does-nothing-and-what-that-says-about-2026s-seo-myths/) ------------------------------------------------------------------------ # The 89% rule — why earned media matters more than your website for GEO URL: https://geohandbook.io/articles/earned-media-89-percent/ Published: 2026-08-28 Summary: 89% of LLM citations come from earned sources, not owned websites. Here's what that means for how marketers and webmasters should allocate GEO effort. The single most useful number in the 2026 GEO literature: 89% of LLM citations come from earned sources, and 23% come from owned websites ([Onely](https://www.onely.com/blog/geo-vs-seo/)). Those two numbers exceed 100% because a single response often cites multiple sources — but the ratio is the point. If you only optimize your own site, you are fighting for less than a quarter of the pie. ## What "earned" means here Onely's categorization, echoed by Firebrand Communications' 2026 best-practices roundup, treats "earned" as any source you don't own outright — journalist coverage, industry PR, Reddit threads, forums, expert quotes on third-party sites, podcast transcripts, review sites, and Wikipedia ([Firebrand](https://www.firebrand.marketing/2025/12/geo-best-practices-2026/)). "Owned" is your website, your app, your blog. The reason earned dominates is retrieval-augmented generation. When Perplexity or ChatGPT answers "what is the best CRM for small law firms?", it doesn't just fetch vendor sites — it fetches the discussion around those vendors, because that is where independent opinion, edge cases and comparison live. ## What this reshuffles The classic content-marketing playbook (own the topic on your own site) still matters — 23% of citations is not nothing — but it stops being the whole plan. The 2026 plan looks more like this: 1. Own-site optimization. Answer-first content, schema, entity clarity. Table stakes. 2. Digital PR. Data-driven story pitches to journalists in your category. A single link from a mainstream outlet often outweighs 20 owned pages. 3. Reddit and community presence. Answer questions in the subreddits your customers use — helpfully, not as spam. Reddit is over-represented in LLM training data and in live retrieval. 4. Expert quote programs. HARO, Qwoted, Featured, Terkel — get your subject-matter experts quoted in industry roundups and "expert says" pieces. 5. Podcast appearances. Transcripts are indexed and quoted. 6. Wikipedia and Wikidata. If your organization is genuinely notable, ensure the entry is accurate and well-sourced. Do not create one artificially. ## How much to shift Onely's default framework is 80–90% traditional SEO / 10–20% GEO on the budget line, with enterprises targeting closer to 85/15 ([Onely](https://www.onely.com/blog/geo-vs-seo/)). Inside the GEO bucket, the 89/23 ratio implies weighting off-site work heavily. A defensible split inside the GEO bucket: - ~50% off-site authority (PR, communities, expert programs). - ~30% content restructuring on your own site. - ~20% technical GEO (schema, rendering, discovery, monitoring). ## The uncomfortable implication If your marketing team has one PR-shaped person and ten content-shaped people, you are structurally under-invested for GEO. Digital PR skill is now closer to the middle of a marketing plan than the edge. ## Checklist - [ ] A named owner of digital PR with a monthly placement target. - [ ] Journalist-query response cadence (HARO / Qwoted / Featured / Terkel). - [ ] Subject-matter expert(s) available for quotes with 24-hour turnaround. - [ ] Presence in the top three subreddits your customers read. - [ ] Podcast pitch pipeline for your named experts. - [ ] Wikipedia/Wikidata monitoring for your organization and key products. ## Sources - [Onely — GEO vs SEO: What's the difference in 2026?](https://www.onely.com/blog/geo-vs-seo/) - [Firebrand Communications — GEO best practices 2026](https://www.firebrand.marketing/2025/12/geo-best-practices-2026/) - [Search Engine Land — Mastering GEO in 2026](https://searchengineland.com/mastering-generative-engine-optimization-in-2026-full-guide-469142) ------------------------------------------------------------------------ # Measuring GEO — the four metrics that replace rankings URL: https://geohandbook.io/articles/measuring-geo/ Published: 2026-08-20 Summary: Four practical metrics to measure Generative Engine Optimization — AI citation rate, share of voice, referral traffic from AI engines, and Bing/Google AI performance reports. Rankings alone are a lying metric in 2026. When 68–72% of searches end without a click and 83% of AI Overview queries produce zero clicks, "we rank #1" no longer implies "we have distribution" ([Onely](https://www.onely.com/blog/geo-vs-seo/)). GEO needs its own instrument panel. Four metrics carry the load. ## 1. AI citation rate The percentage of AI responses to a defined prompt panel that cite your domain. Strapi formalizes it as `SUM(is_cited) / COUNT(*)` over a rolling 30-day window ([Strapi](https://strapi.io/blog/generative-engine-optimization-geo-guide)). How to run it. Build a prompt panel of 20–40 real customer questions — brand, category, comparison and problem prompts. Once a month, run every prompt in ChatGPT, Perplexity, Google AI Overviews, Claude and Copilot. Record whether your domain is cited and which URL. Do this manually the first quarter — the tooling landscape is not yet trustworthy enough to skip the eyeballs. Target. Citation rate is category-dependent. What matters is whether your rate is rising quarter over quarter. ## 2. Share of voice Your citations divided by total citations across responses for a category. This is the GEO equivalent of market share — and a much better business metric than raw ranking position, because it captures competitive dynamics. How to run it. For each prompt category (brand, category, comparison), sum citations per domain across the month's runs. Your share is `your_citations / total_citations_in_category`. Target. Aim for share of voice matching or exceeding your traditional market share in categories where AI answer engines already dominate discovery. ## 3. AI referral traffic Direct traffic from `chat.openai.com`, `perplexity.ai`, `gemini.google.com` and `copilot.microsoft.com` in GA4. Small volumes today, growing fast — Onely reports 34% CAGR growth for AI-answer channels ([Onely](https://www.onely.com/blog/geo-vs-seo/)). How to run it. Segment referrals by source in GA4. Track absolute clicks and, more importantly, conversion rate — Onely reports GEO traffic converting at 27% vs 2.1% for traditional SEO ([Onely](https://www.onely.com/blog/geo-vs-seo/)). Watch for. Referral traffic is a lagging indicator: users often act on AI answers without ever clicking through. High citation rate with low referral traffic is normal — do not treat low referral clicks as failure. ## 4. AI Performance reports (Bing, Google) Wikipedia and Google's own documentation both point to Bing Webmaster Tools' AI Performance report and Google Search Console's Search Generative AI performance report as the officially supported view into how your content is used by AI features ([Wikipedia](https://en.wikipedia.org/wiki/Generative_engine_optimization), [Google Search Central](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide)). How to run it. Verify your domain in both tools. Review monthly. Bing's report currently gives more granular AI-answer data than Google's, but Google's report is expanding as AI Overviews rollout continues. ## What to skip - Vendor "AI visibility scores" that don't publish their methodology. They vary wildly from month to month. Use them as a directional signal only. - Raw ranking position as a standalone metric. Track it, but don't optimize to it. - Snapshot-only citation checks. Once-off audits look impressive but tell you nothing about trend. ## The reporting cadence that works - Weekly. Traffic dashboards; segment AI referrals from other sources. - Monthly. Prompt-panel rerun; citation rate and share of voice by category; Bing and Google AI performance reports. - Quarterly. Full audit — prompt panel refresh, competitor citation comparison, playbook adjustments. ## Sources - [Onely — GEO vs SEO 2026](https://www.onely.com/blog/geo-vs-seo/) - [Strapi — Generative Engine Optimization guide](https://strapi.io/blog/generative-engine-optimization-geo-guide) - [Wikipedia — Generative engine optimization](https://en.wikipedia.org/wiki/Generative_engine_optimization) - [Google Search Central — Optimizing for AI features](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) ------------------------------------------------------------------------ # About GEOhandbook URL: https://geohandbook.io/about/ Summary: About GEOhandbook — our editorial standards, sourcing policy, and why this site exists. ## Why this handbook exists Every marketing outlet is now publishing on GEO, most of it interchangeable. GEOhandbook exists to be the practical, cited, opinionated reference site that treats you like a professional — the kind of resource a marketing lead can send to their team and their developer and know both will get value. ## Editorial standards - Every non-obvious claim is linked to a primary source. Traffic-decline percentages, citation ratios, AI Overview reach numbers — all cited inline, no aggregated "according to reports." - We source the sources. When we cite Onely citing Gartner, we say so. - We update, not rewrite. Pages get a visible "Last updated" date and dated change notes when a claim changes. - We admit uncertainty. Where the industry disagrees (for example on whether `llms.txt` moves the needle), we say so on both sides. ## Who this is for - Marketers who need a defensible narrative for their leadership team. - Webmasters and developers who need to ship the technical work without wading through vendor pitches. - Content teams who need to change how they write to survive the shift to answer engines. ## Editorial disclosure GEOhandbook shows Google AdSense advertising to support the site. Ad placements are labelled and never influence editorial coverage. See our [disclosure page](/disclosure/) for detail. ## Get in touch - General: `hello@geohandbook.io` - Tip a source or correction: `editor@geohandbook.io` - RSS: `/feed.xml` The site is built with [Eleventy](https://www.11ty.dev/) and deployed on Cloudflare Pages. The source is intentionally boring — HTML, CSS and a tiny amount of JavaScript — because GEO rewards boring HTML. ------------------------------------------------------------------------ # Editorial disclosure URL: https://geohandbook.io/disclosure/ Summary: Editorial disclosure — advertising, affiliate links, sponsorships, and editorial independence at GEOhandbook. ## How this site is funded GEOhandbook is funded by: 1. Google AdSense. Ad placements are labelled "Advertisement" and are set apart from editorial content. Ads are chosen by Google, not by us. 2. Occasional affiliate links. Where a tool is recommended and an affiliate program exists, we may use an affiliate link. Affiliate status never determines whether or how prominently a tool is covered — and affiliate links are always disclosed inline where they appear. ## What we will not do - We do not sell coverage. - We do not run "sponsored" articles inside our editorial sections. If a sponsor ever supports a piece, it will be clearly labelled "Sponsored" and separated from editorial content. - We do not let advertisers see or influence editorial before publication. ## Corrections We update pages when we get something wrong. Substantive corrections are noted in a dated change log at the bottom of the affected page. ## Contact Coverage suggestions or corrections: `editor@geohandbook.io`. ------------------------------------------------------------------------ # Privacy URL: https://geohandbook.io/privacy/ Summary: The privacy policy for GEOhandbook, including analytics and advertising cookies. ## Analytics GEOhandbook uses privacy-preserving analytics to count pageviews and referrers. We do not build user-level profiles for advertising. ## Advertising This site displays Google AdSense advertising. Google and its partners may use cookies to serve ads based on your prior visits to this and other sites. You can opt out of personalised advertising at [Google's Ads Settings](https://www.google.com/settings/ads) and manage cookies through your browser. Ads are always labelled with the word "Advertisement" and are visually separated from editorial content. ## Cookies The core site does not require cookies. Cookies that appear are set by Google AdSense and, if you interact with them, embedded services like YouTube or X. Blocking these cookies has no impact on your ability to read the site. ## Contact Questions about privacy or data requests: `privacy@geohandbook.io`.