How AI search engines work

Understanding how AI tools generate answers is the prerequisite for optimising for them. There are two main architectures:

Training-based knowledge

Models like GPT-4 and Gemini are trained on vast corpora of web content. They "know" information from their training data. This means brands and entities that have significant, accurate web presence before a model's training cutoff are more likely to be represented in the model's baseline knowledge.

Retrieval-Augmented Generation (RAG)

Newer tools - including Perplexity, Google AI Overviews, and ChatGPT's web browsing mode - combine LLM capabilities with real-time web search. The AI searches the web, retrieves relevant pages, and synthesises an answer using that retrieved content. This is where GEO has its most direct, actionable impact: your pages can be retrieved, and your content can be cited right now.

What makes AI tools cite a source?

Through testing across dozens of domains and analysis of AI citation patterns, we've identified the primary signals that influence whether a page gets cited:

1. Topical relevance and specificity

Pages that provide precise, specific answers to specific questions perform better than broad, general pages. A page titled "What is CAC and how do you calculate it?" with a clear formula will be cited more readily than a generic "Marketing Glossary" page.

2. Authoritative domain signals

AI retrieval systems, like traditional search engines, favour pages from authoritative domains. A high Domain Rating (DR) or strong topical authority - dozens of interlinked pages on a subject - increases the likelihood of retrieval.

3. Structured, extractable content

AI tools extract specific passages. Content that is structured for easy extraction performs significantly better:

• Clear H2/H3 headings that describe the section content precisely
• Short, self-contained paragraphs that make one clear point
• Numbered or bulleted lists for processes and comparisons
• Definition patterns: "[Term] is [definition]" or "[Term]: [definition]"
• Tables for comparisons and data

4. Factual density and citations

Content with statistics, named sources, specific examples, and verifiable claims outperforms vague, generalist content. AI tools are more likely to cite a page that makes a specific claim they can verify than one that makes a general assertion.

5. Entity consistency

AI models recognise entities - companies, people, products, concepts. If your brand name, author names, and product names appear consistently across many credible sources, the model has stronger confidence in your entity and is more likely to surface you in responses.

The GEO strategy framework

Our recommended GEO approach is a 4-part framework:

Part 1: Content audit for GEO readiness

Review your top 20 pages for GEO signals. Score each page on: factual density (0-3), structural clarity (0-3), entity mentions (0-2), and citation potential (0-2). Pages scoring below 5/10 are GEO-upgrade candidates.

Part 2: Topic-by-topic authority building

Pick 3-5 core topics where you want AI citation. For each, create a comprehensive "pillar" article (the definitive resource on the topic) and 8-12 supporting articles that link to and from it. This topical cluster signals to both Google and AI tools that you're the authoritative source.

Part 3: Entity footprint expansion

Systematically build your brand's entity presence: consistent NAP (Name, Address, Phone) in directories, author pages on industry publications, your own Wikipedia entry if warranted, schema markup across your entire site, and PR campaigns placing your brand name in authoritative media.

Part 4: Structured data implementation

Implement schema markup - at minimum: Organization, WebSite, Article, FAQ, HowTo, and BreadcrumbList schemas. These help AI retrieval systems correctly categorise and extract your content.

Measuring GEO performance

Traditional rank tracking tools don't capture AI citations. Current best practices for measurement:

• Manual prompt testing: regularly query ChatGPT, Perplexity, and Google AI Overviews for your target topics and track when your domain is cited.
• Google Search Console: AI Overview impressions now appear in GSC - monitor "search appearance" filters for AI Overview data.
• Brand mention tracking: tools like Mention or Brand24 can track brand mentions in AI-generated content that gets published to the web.
• Perplexity referral traffic: Perplexity cites sources and those citations drive real traffic - monitor your analytics for perplexity.ai referrals.