GEO resources
Learn generative engine optimization (GEO), improve one useful page and evaluate how AI answers use your content. Start with the beginner path or find a guide for the question you have now.
Featured guide · Practical GEO guides
How to write content that can support an AI citation
Rewrite vague claims as specific answers with clear limits and supporting sources. Includes a worked example and publishing checklist.
New to GEO? Start here
Follow these three guides in order: check your starting point, improve your content, then review the evidence.
1Check the basics
Start with the beginner checklist ›
Choose one page and work through access, content and evidence checks. Save your progress as you go.
2Improve one page
Write a clear, verifiable answer ›
Replace vague claims with specific facts, useful qualifications and links to supporting evidence.
3Review the results
Check whether citations support the answer ›
Inspect the claims and cited pages before treating a mention or link as a success.
Find a GEO guide
Search by topic or choose a category. Each summary tells you what the guide will help you do.
11 guides
Practical GEO guides
Improve access, plan useful pages and publish trustworthy content.
A practical GEO optimization checklist for beginners ›
Work through six beginner steps for one page, from checking access to reviewing AI answers. Track your progress with the interactive checklist.
How to write content that can support an AI citation ›
Rewrite vague claims as specific answers with clear limits and supporting sources. Includes a worked example and publishing checklist.
An access and eligibility checklist for AI search ›
Check public access, crawler rules, indexing and snippet eligibility. Learn to distinguish a blocked page from one an assistant simply did not select.
Build a GEO content plan around real customer decisions ›
Turn real customer questions into a prioritized content plan. Decide what to create, what to improve and where evidence is missing.
Create comparison pages that answer real buying questions ›
Build a fair comparison page around buying decisions. Define criteria, verify differences and disclose versions and limitations.
Publish original evidence without manufacturing authority ›
Publish real tests, customer examples and first-party observations with enough context for readers to verify the evidence.
A practical workflow for refreshing GEO content ›
Find outdated or conflicting facts, make focused corrections and review the results without assuming a new date improves citations.
Deeper technical explainers
Understand how retrieval, context and citations fit together.
From retrieval to citation: where evidence can be lost ›
Trace the path from finding a page to citing it in an answer. Learn where useful evidence can be dropped or misrepresented.
Context windows, passages and the limits of long documents ›
Understand what a context window can hold and why long pages may still be only partly used. Make excerpts retain important qualifications.
Research and measurement
Review citation quality and measure changes without overstating the results.
How to audit whether AI citations support the answer ›
Review an AI answer claim by claim and check its cited pages. Separate citation counts from actual support and factual correctness.
Design a GEO study without overstating the results ›
Plan a repeatable visibility study with comparable questions and clear denominators. Report observed changes without claiming unproven causes.
GEO visibility: compare the same outcome
These are GeckoCheck measurement definitions, not measured model scores. Compare assistants within matched cohorts, not by combining unrelated studies.
| Metric | Consistent definition | Cross-model comparison rule |
|---|---|---|
| Unprompted brand rate | Successful discovery answers naming the brand / successful discovery answers. | Use matched prompt intent, language and time window. Branded verification prompts are a separate cohort. |
| Owned-source rate | Successful relevant answers citing an owned public page / successful relevant answers. | Separate inline citations, related links and Sources-panel references. Exclude supplied files. |
| Citation share | Owned-domain citation occurrences / all eligible citation occurrences in the defined cohort. | Declare duplicate handling and which link types count. Not a share of all users or all possible prompts. |
| Citation support rate | Reviewed citation–claim pairs that support the claim / reviewed citation–claim pairs. | Apply the same review rubric. A large citation count does not prove accurate answers. |
| Qualified referral outcomes | Identifiable visits and meaningful actions, reported with raw counts. | Report attribution gaps. Mentions and citations are not visits or conversions. |
Source: Aggarwal et al.: GEO (KDD 2024)
Period: Definitions prepared October 12, 2026; research context: KDD 2024.
Sample: No collected assistant observations. Operational definitions for a future matched sample.
Limitation: GeckoCheck conventions; not universal provider metrics. Different denominators cannot be compared directly.
Model capabilities and access
Compare the assistant experience, not only the underlying model. A documented capability does not guarantee that every answer uses it. On smaller screens, scroll the table horizontally.
| Assistant / surface | Discovery capability | Source presentation | Access and scope | Evidence |
|---|---|---|---|---|
| ChatGPT Search | Automatic or manually selected web search; may rewrite queries. | In-answer citations and a Sources panel, which also includes other relevant links. | Documented across Free, Go, Plus, Pro, Business, Enterprise, Edu and signed-out use; usage and workspace limits apply. | Provider documentation ›Live documentation checked October 12, 2026 |
| Gemini Apps | Public-web information and optionally supplied or connected material. | Sources and related links may appear; not every response includes them. | Record the app, research mode and connected sources; do not substitute API behavior. | Provider documentation ›Live documentation checked October 12, 2026 |
| Google AI Overviews | Search summaries; may use multiple related searches (fan-out). | Supporting website links; not triggered for every search. | An eligible supporting page must be indexed and snippet-eligible in Google Search; this does not guarantee inclusion. | Provider documentation ›Live documentation checked October 12, 2026 |
| Google AI Mode | Search experience for exploration and complex comparisons; may use fan-out. | Supporting links; results can differ from AI Overviews. | Google Search eligibility applies to supporting pages; record the actual mode and locale. | Provider documentation ›Live documentation checked October 12, 2026 |
| Perplexity | Web discovery and synthesis; Pro Search and Research are distinct experiences. | Numbered citations linking to original sources. | Free users have limited Pro searches; subscriptions extend access. Record the selected search experience. | Provider documentation ›Live documentation checked October 12, 2026 |
| Claude | Web search and direct web fetch; behavior depends on the available experience. | Direct citations and source links in search responses. | Supported models and usage limits apply; Team/Enterprise owners enable the workspace capability. | Provider documentation ›Live documentation checked October 12, 2026 |
| Copilot (Microsoft 365 scope) | Generated Bing queries can ground answers when web search is enabled. | Web-search query citations; inspect source support separately. | Organizational documentation: administrator controls and eligible user toggles apply. Not a consumer-wide specification. | Provider documentation ›Live documentation checked October 12, 2026 |
| Grok | May decide to search public X posts and the live web. | Review the actual answer and cited material; the overview does not guarantee links on every answer. | Record the product surface and search behavior; an API tool configuration is not the same as consumer chat. | Provider documentation ›Live documentation checked October 12, 2026 |
| Meta AI | Web search described in the app launch announcement. | Launch announcement does not establish a universal citation format. | Historical product announcement; verify present availability in the observed app and region. | Provider documentation ›April 29, 2025 announcement |
| Mistral Le Chat | Announced mix of web search, journalism, social media and other sources. | Evidence-backed research described; source availability varies by experience. | Historical assistant announcement, not a specification for every Mistral API implementation. | Provider documentation ›February 6, 2025 announcement |
Source: ChatGPT Search · Gemini Apps · Google AI Overviews · Google AI Mode · Perplexity · Claude · Copilot (Microsoft 365 scope) · Grok · Meta AI · Mistral Le Chat
Period: Checked October 12, 2026; historical announcement dates are shown per row.
Sample: Ten documented assistant surfaces; documentation review, not a performance experiment.
Limitation: Plans, regions, limits and modes can change. Organizational Copilot, Gemini Apps and historical launch announcements have explicitly different scopes.
Evidence by study, not a universal leaderboard
Each study answers a different question. Do not rank current assistants using historical citation errors, SEO overlap and experimental uplift as if they measure the same thing.
| Study / surface | Reported result | Period and sample | Interpretation limit |
|---|---|---|---|
| Tow Center / Perplexity | 37% of article-identification queries answered incorrectly. | Published March 2025; 200 selected excerpts per tool, 20 publishers. | News-excerpt retrieval task, not general accuracy or brand visibility. Historical tool configuration. |
| Tow Center / ChatGPT Search | 134 / 200 article identifications incorrect (67%, calculated from the reported count). | Same 200-excerpt task; eight tools, 1,600 total queries. | Cannot convert this into a current citation likelihood or GEO ranking. |
| Tow Center / Grok 3 beta | 94% of article-identification queries answered incorrectly. | Same March 2025 study and selected news excerpts. | Beta version at study time; do not generalize to current Grok. |
| Ahrefs / Perplexity | 28.6% of cited URLs appeared in Google’s top 10 for the original query. | Published August 11, 2025; study used 15,000 long-tail queries. | URL overlap, not accuracy. Query rewriting and link-type differences affect comparisons. |
| GEO / experimental engines | Reported visibility improvements up to 40% in the evaluated conditions. | KDD 2024; GEO-bench contains 10,000 queries; experiments use defined subsets. | An upper reported improvement, not a typical result or a current assistant leaderboard. |
Source: Tow Center: AI Search Has a Citation Problem · Aggarwal et al.: GEO (KDD 2024) · Ahrefs: AI search overlap (August 11, 2025)
Period: 2024–2025 publications; reviewed October 12, 2026.
Sample: Study-specific samples listed in each row; they are not pooled.
Limitation: No verified, current, matched brand-visibility dataset across all ten assistants. Missing evidence is unknown, not a zero score.
Historical comparison: incorrect news-article identification
Share of 200 queries per tool marked incorrect in the Tow Center study. Lower is better for this specific task.