AI Search

AI search is reshaping how people find information online, and how content gets discovered and cited when the answer itself, rather than a list of links, is the product.

What is AI search?

AI search refers to the generation of direct answers by large language models (LLMs) and AI-powered search engines, in place of (or alongside) the traditional ranked list of blue links. Systems like Google’s AI Overviews, Perplexity, ChatGPT Search, and Microsoft Copilot synthesise information from across the web and present a conversational response, often with citations.

This shift changes what it means to “rank”. Visibility in AI search is about whether your content is retrieved, parsed, trusted, and cited by the system generating the answer.

The practice of optimising for these surfaces goes by several names: Generative Engine Optimisation (GEO), Answer Engine Optimisation (AEO), LLMO, and AI SEO are all in use. They describe the same underlying goal. This pillar uses GEO as the primary term; the GEO guide covers how the terms relate.

Core AI search elements

These break into five groups: how AI search works, how to earn citations (GEO), platform-specific optimisation, the emerging agentic web (the shift to agentic AI, where autonomous agents read and act on content), and using AI to do SEO work.

How AI search works

• How AI Search Works. The mechanics behind AI answer generation: training data versus real-time retrieval, grounding (RAG), why crawlability feeds into citation potential, and why new sites can appear in AI answers.
• RAG and SEO. How Retrieval-Augmented Generation determines which passages AI search engines cite, and how to structure content for retrieval.
• Query Fan-Out. How AI Mode expands a single query into multiple parallel sub-queries, and why topical coverage across a full cluster matters more than matching a single keyword.
• Knowledge Graph and knowledge panels. What the Knowledge Graph is, how entities get into it, what triggers a knowledge panel, and how to influence your entity representation in search results.
• AI crawler user-agents. A reference to active AI crawlers, who operates them, whether they are used for training or live retrieval, and how to verify legitimacy. Most operators run training and retrieval crawlers as separate user-agents, so blocking one leaves the other untouched; Googlebot, Bingbot and Applebot do both jobs, which is why a CDN block on training can stop search crawling too.

Earning citations (GEO)

• Generative Engine Optimisation (GEO). The practice of structuring content for retrieval and citation by AI answer engines. What moves the needle overlaps with SEO almost entirely; the page separates what the platforms say, what has been measured, and what nobody has measured.
• Entity SEO. Why a site, author or product that is named unambiguously and corroborated across the web is easier for any system to attribute, and how to establish that.
• AI Visibility Measurement. Citation rate, share of voice, and prompt coverage: the metrics that capture AI search visibility where traditional rankings do not.

How each platform retrieves and cites

The tactics do not change by platform: no AI search company publishes citation guidance beyond “be indexable and answer clearly”. What differs is the retrieval path, which index each one reads from, which crawler needs allowing, what the company has documented, and where citations can be measured. That is what these pages cover.

• AI Overviews. Google’s generative answer feature, shown at the top of many results pages. Google sets no requirements beyond its Search fundamentals: a page has to be indexed and eligible to show with a snippet, and what it takes to be cited beyond that is undocumented.
• Google AI Mode. Google’s conversational search tab, how it differs from AI Overviews, and what content it retrieves and cites.
• ChatGPT Search optimisation. How ChatGPT’s web search retrieval works, what OpenAI documents and does not, and what the stored snippet means for your H1.
• Perplexity citation strategies. What Perplexity documents about its search index (passages ranked one by one, stale content filtered), which of its two crawlers to allow, and the one study that measured what it cites.
• Claude Search optimisation. How Claude appears to retrieve web content via Brave Search, which two of Anthropic’s three crawlers affect citation visibility, and why everything after retrieval is undocumented.
• Microsoft Copilot optimisation. How Copilot retrieves content from the Bing index, the three things Microsoft has said, none of them measured (structured content helps its AI answers, freshness via IndexNow, and headings mark out the pieces Copilot splits a page into), and how to track citations across Copilot and Bing’s AI features in Bing Webmaster Tools’ AI Performance report.
• Grok (xAI) search optimisation. How Grok blends live web search with real-time X data, its two retrieval tools (web_search and x_search) with DeepSearch running multi-step research on top of them, and why an active X presence is the one lever that is Grok-specific.

Agents and the agentic web

• What are AI agents?. What AI agents are, how they differ from AI assistants and chatbots, and how the main categories (interactive vs background) map to search and content visibility.
• Agentic search. How AI agents retrieve content across multi-step research tasks, why ranking still governs which pages they reach, and why the idea that consolidated pages cut agent multi-hopping is unproven.
• Gemini Spark. Google’s personal AI agent: how it monitors topics and compiles digests from web content, and what it means for publishers whose content is synthesised without a click.
• Agent-readiness. The infrastructure layer that determines whether AI agents can discover, read, and act on a website. Covers the five infrastructure layers, what Cloudflare, Shopify, and WordPress each check for, and what content sites should implement now.
• WebMCP. The browser-native protocol that lets websites declare callable tools for AI agents. Where agentic search optimises content for retrieval, WebMCP declares what agents can do on a site. Early priority for e-commerce, travel, and booking sites.
• NLWeb. Microsoft’s open-source protocol that converts existing schema.org markup into queryable AI interfaces, exposing /ask (REST) and /mcp (MCP server) endpoints. Built by R.V. Guha (creator of Schema.org, RSS, and RDF). Most relevant for commerce, media, and documentation sites with deep content catalogues; informational sites should monitor rather than implement.
• llms.txt. A proposed convention for declaring an AI-readable index of a site’s content. Google’s AI optimisation guide confirms¹ it is not necessary for visibility in Google Search; publish it for browser-based agent navigation, the case Chrome gives for its Lighthouse check, or to give your audience curated documentation.
• Markdown Pages for AI Agents. Cloudflare and others offer tools to serve Markdown versions of pages to any client that asks for one. No evidence supports a citation benefit; it is useful where AI coding tools or agents are the main audience, such as developer documentation, and for readers pasting a page into an AI assistant.

Using AI for SEO

• Can ChatGPT do SEO?. What ChatGPT can and cannot do across SEO tasks: where it helps (ideation, drafting, data wrangling) and where it misleads (fabricated data, outdated tactics, no live ranking access).
• Agentic SEO. How AI agents execute SEO workflows autonomously: keyword research, content briefs, technical audits, and performance reporting. What tasks agents handle reliably, where human oversight remains essential, and what orchestration tools are available.
• AI Content Risks. Why reckless AI content production tends to reduce search and AI citation visibility: Google’s spam policy, hallucination, E-E-A-T erosion, and the irony that the volume strategy sold for AI visibility is a named spam policy violation on Google’s AI features.
• AI Watermarking. How SynthID and C2PA Content Credentials mark AI-generated content at the point it is made, why anyone can now check an image, video or audio file but nobody outside the provider can check AI text, and when the EU AI Act requires you to label AI-assisted text.

Why does AI search matter?

AI search has crossed into mainstream behaviour. A Pew Research Center survey conducted in February 2026 found around 60% of US adults have read AI-generated summaries in search results, and 49% have used an AI chatbot, with searching for information among the most common uses.² The share of searches that return an AI-generated answer is growing rapidly, and AI Overviews appear on a large and growing share of informational results pages. A page can hold its traditional ranking and still lose clicks when an AI Overview answers the query above it.

AI search also introduces zero-click risk at scale. If an AI summarises your content and provides the answer directly, users may never visit your site. Citation and brand visibility in AI responses is a new form of organic reach that matters independently of click-through rate.

AI search and E-E-A-T

E-E-A-T is Google’s quality framework, applied by its search quality raters and reflected in its ranking systems. Google’s AI features are “rooted in our core Search ranking and quality systems”,¹ so the same framework applies to them. Beyond Google, no platform documents an authorship or expertise signal; what is measured is that the pages AI systems cite overlap heavily with the pages that rank, which the GEO guide sets out with its sources.

The brands and publications that dominate AI citations over the next few years will be those that invested in genuine authority, not those that optimised purely for keyword placement.

Understanding how AI search works, from retrieval mechanisms to entity recognition, is the foundation for these investments. AI content risks, including hallucination and E-E-A-T erosion from unreviewed AI-generated production, degrade the quality Google’s systems reward, and Google names AI-generated volume aimed at its generative responses as scaled content abuse.¹ For sites that serve transactional queries, agent readiness, the infrastructure layer that lets AI agents discover and act on a site, is the next consideration beyond content quality alone.

Footnotes

1. Optimizing your website for generative AI features on Google Search — Google Search Central ↩ ↩² ↩³

2. Americans and AI 2026: Chatbots, Smart Devices and Views on Impact — Pew Research Center ↩