Reference

AI search visibility, documented.

24 AI visibility platforms scored on the same five dimensions. 19 engine profiles covering how each one retrieves, grounds, and cites. 44 defined terms. The funding map behind the category. Every figure sourced and dated.

Two disciplines sit underneath all of it. Answer Engine Optimization (AEO) covers the extractive surfaces: featured snippets, voice, AI Overviews. Generative Engine Optimization (GEO) covers the grounded generative engines: ChatGPT Search, Perplexity, Claude, Gemini, Bing Copilot. They share a foundation; the success metrics differ.

The core idea

How an AI engine picks its sources

1. Step 1

   A question arrives

   The engine decides whether the answer needs the live web.

2. Step 2

   Candidates are retrieved

   Only pages its crawlers can reach make the pool.

3. Step 3

   They are ranked

   Clear structure, freshness, and sourcing win.

4. Step 4

   Two or three get cited

   Citations are the visibility you can measure.

Platforms profiled

24

Engines profiled

19

Defined terms

44

Disclosed capital

$771.2M

Long-form editorial on the disciplines, implementation, and measurement.

• Foundations · AEO + GEO · 12 min

  AEO vs GEO: how Answer Engine Optimization and Generative Engine Optimization actually differ

  Two disciplines, one shared foundation, meaningfully different KPIs, and why most practitioners blur them incorrectly

• Implementation · GEO · 11 min

  The llms.txt specification, in depth

  What llms.txt actually is, what llms-full.txt adds, who's adopting it, and how to ship one that AI engines might actually use

• Implementation · AEO + GEO · 13 min

  Schema.org for AEO and GEO: which structured data actually matters

  Schema.org has hundreds of types. Only a small subset moves the needle for AI engines, and the priority order is not obvious

• Foundations · AEO + GEO · 11 min

  Citation-worthy content patterns: writing for both extraction and grounding

  The editorial patterns that survive both AEO extraction and GEO grounding, and the patterns that lose in both

24 AI visibility and citation-tracking platforms scored on the same five dimensions. Includes GrackerAI with explicit author-conflict disclosure.

The capital map of AI search visibility: the first unicorn round, the VCs placing repeat bets, Adobe's $1.9B acquisition of Semrush, and the bootstrapped indies. Sourced and dated.

Free interactive tools: llms.txt generator, schema validator, AI Visibility Scorecard, Citation Tracker Quickstart. No email gate; everything runs in your browser.

How each AI search engine grounds, what it cites, and what it rewards, covering both the extractive AEO surfaces and the grounded GEO engines.

• OpenAI

  ChatGPT Search

  OpenAI's web-grounded mode inside ChatGPT, surfaced as 'Search' and used for queries the model judges to benefit from fresh information. Grounded answers cite web sources alongside the assembled response. OpenAI retired GPT-4o in ChatGPT on 2026-02-13. OpenAI is also testing ads beside answers for Free and Go users and states that answers are independent of ads.

• Perplexity AI

  Perplexity

  Perplexity is a search-first AI engine: answers are grounded against retrieved sources by default, with a citation-forward UI. Users can choose among models from several providers, including GPT and Claude. Many SEO teams read it as a leading indicator of GEO signal.

• Anthropic

  Claude

  Anthropic's Claude is a conversational AI with web search. Claude searches when it judges the question benefits from it; there is no user toggle documented for that decision. Anthropic documents three crawler roles: ClaudeBot (training only), Claude-SearchBot (search index), and Claude-User (fetches made on a user's behalf).

• Google

  Gemini

  Google's Gemini app is grounded against Google Search when the conversation calls for current information. Distinct from Google AI Overviews: Gemini is the conversational interface, AI Overviews is the SERP-integrated answer block. Google states that blocking the Google-Extended robots.txt token affects Gemini Apps (training and grounding).

• Google

  Google AI Overviews

  AI Overviews is the SERP-integrated answer block Google began rolling out in 2024 (formerly 'Search Generative Experience' or SGE). It appears at the top of search results for queries Google judges to benefit from an assembled answer, with links to supporting sources. Google states that no special schema or AI text files are needed to appear, and that AI feature traffic is counted under 'Web' in Search Console. It is the most consequential AEO surface for sites that already rank well in traditional Google search.

• Microsoft

  Bing Copilot

  Microsoft Copilot Search in Bing is an AI answer experience grounded against Bing's search index. It is increasingly important as Microsoft pushes Copilot through Edge, Windows, and Microsoft 365 surfaces, and as enterprise buyers consolidate on Microsoft AI tooling.

• Google

  Google AI Mode

  Google AI Mode is the conversational AI surface inside Google Search, distinct from AI Overviews. Where AI Overviews is an answer block above the organic results, AI Mode is an active chat experience: the user opts in, asks multi-turn questions, and gets a conversational answer with links to sources. Google documents query fan-out for AI Mode, so retrieval can reach beyond what a single classical query would return. Google states that no special schema or AI text files are needed, and that AI feature traffic is counted under 'Web' in Search Console.

• SpaceXAI (formerly xAI)

  Grok

  Grok is the chatbot from SpaceXAI (renamed from xAI on 2026-07-06), available inside X and at grok.com. Its distinctive feature is access to real-time X posts as a grounding source, which gives it citation behaviour unlike other major engines. Strongest on breaking news, weaker on long-tail editorial content where X conversation is sparse.

• Meta

  Meta AI

  Meta AI is Meta's assistant, embedded in Facebook, Instagram, WhatsApp, Messenger, and the standalone meta.ai web app. Meta's Muse Spark model (from Meta Superintelligence Labs, announced 2026-04-08) is reported to power it in place of Llama. Search grounding was historically through Bing; Google results are reported to have been added alongside Bing.

• DeepSeek (China)

  DeepSeek

  DeepSeek is a Chinese AI lab whose chat interface at chat.deepseek.com gained global prominence in early 2025 when DeepSeek-R1 launched at a fraction of the cost of comparable Western models. The current model family is V4 (Pro and Flash); V4.1 Flash was announced on 2026-09-10. Strong in Chinese-language queries, with growing English-language usage as a free alternative to ChatGPT and Claude.

• Mistral AI (France)

  Mistral Le Chat

  Le Chat is Mistral AI's conversational interface. Mistral is headquartered in Paris and operates under EU jurisdiction, with GDPR alignment as a stated differentiator. Current models include Mistral Large 3, Medium 3.5, and Small 4. Mistral announced a content agreement with AFP in January 2025 for news in Le Chat. Strong adoption in European enterprise and public-sector contexts.

• Alibaba Cloud (China)

  Qwen

  Qwen is Alibaba Cloud's conversational AI, powered by the Qwen model family. The consumer app is reported to have been renamed from Tongyi to Qwen. Reporting indicates recent models are in the Qwen3.7 and 3.8 range; check the AI Models reference for current versions. Strong in the Chinese market and across Asia, with web search grounding. Open-source Qwen variants are widely deployed across the Chinese AI ecosystem.

• Baidu (China)

  Baidu AI Search

  Baidu is the dominant search engine in mainland China. Baidu's ERNIE 5.0 was released on 2025-11-13, and Baidu has described rebuilding its search results page as an AI application, with a large share of rich media. Baidu also operates ERNIE Bot as a standalone assistant. For any publisher targeting Chinese-language audiences, Baidu's AI surfaces are a consequential AI search channel: Google has effectively no presence in mainland China.

• Moonshot AI (China)

  Moonshot Kimi

  Kimi is Moonshot AI's chat product, distinguished early by very long context windows (in 2024, 2 million Chinese characters). The current model is Kimi K3 (released 2026-07-16). Kimi's agent mode shows the sites it visited and the sources it used. Strong adoption in Chinese knowledge-work and academic contexts.

• Yandex (Russia)

  Yandex Alice AI

  Alice AI is Yandex's AI assistant, built on the Alice AI LLM family. Yandex announced on 2025-10-28 that it is built into Yandex Search, the dominant search engine in Russia and several CIS countries. For Russian-language publishers and Western publishers with Russian-language content targeting CIS markets, it is a meaningful AI search channel (Google has reduced presence in Russia since 2022).

• Naver (South Korea)

  Naver AI Search (HyperCLOVA X)

  Naver dominates South Korean search. Naver's AI Tab reached full launch on 2026-06-26, with an AI Briefing entry point from July, and is based on HyperCLOVA X. For any publisher targeting Korean-language audiences, Naver's AI surfaces are the primary AI search channel.

• Brave (US)

  Brave Leo

  Leo is the AI assistant integrated into the Brave browser, with privacy-first defaults: no account is required and chats are not retained. Brave lists Mixtral, Claude, and Llama among the models available, with no single default model. Less consequential for raw citation share than the major engines, but meaningful for publishers reaching privacy-conscious or developer audiences.

• Apple

  Apple Siri and Spotlight web answers

  Apple's Siri and Spotlight can answer questions using web content. Apple documents that Applebot, its web crawler, is used for Siri and Search answers that include links to sources. Applebot-Extended is a separate control that lets sites opt out of having their content used to train Apple's foundation models; it does not crawl.

• Yahoo

  Yahoo Scout

  Yahoo Scout is an AI answer engine from Yahoo, launched on 2026-01-27. Yahoo's announcement presents it as a new AI answer engine at scout.yahoo.com. Sports and finance experiences were reported on 2026-06-03.

The vocabulary of AI search, defined, dated, and citable.

• GEO (Generative Engine Optimization)

  The discipline of getting your content picked up, grounded against, and cited by generative AI search engines like ChatGPT Search, Perplexity, Claude, Gemini, and Google AI Overviews.

• AEO (Answer Engine Optimization)

  Answer Engine Optimization (AEO) is the practice of structuring content so that answer engines return your page as the direct answer to a user's question.

• RAG (Retrieval-Augmented Generation)

  The architecture pattern where an AI engine retrieves source documents from a corpus before generating an answer, conditioning the response on the retrieved context. The technical foundation underneath both AEO and GEO.

• AIO (AI Overviews / AI Optimization)

  Two distinct meanings sharing the same acronym. Most often refers to Google's AI Overviews: the SERP-integrated AI answer block. Sometimes used as a catch-all for AI Optimization across all engines (a near-synonym for GEO + AEO).

• llms-full.txt

  A companion to llms.txt that contains the full Markdown content of a site's priority pages concatenated into a single file, intended for AI engines that prefer single-fetch corpus ingestion.

• Citation share

  The GEO equivalent of organic share of voice: the percentage of AI engine answers in a query set that cite your domain. The primary KPI for measuring AI visibility programs.

• Grounding model

  The specific architecture an AI engine uses to retrieve and condition its answer on external sources. Differs meaningfully across engines and is the variable that determines which content gets cited.

• Schema.org for AEO and GEO

  The Schema.org structured data types worth using on pages you want AI engines to understand. Structured data is general hygiene that must match the visible page; Google states no special schema is needed for its AI features.

Methodology, in public

How GEO Compass evaluates engines, scores AI visibility, and what counts as a citation. No hidden scoring; the whole framework is one page.

Read the methodology →