GEO and AEO — preparing for AI-first search

Two new acronyms have entered the SEO conversation. GEO — Generative Engine Optimisation — is the practice of getting your content surfaced inside AI answer interfaces like ChatGPT, Perplexity, and Google's AI Overview. AEO — Answer Engine Optimisation — is the older sibling, focused on featured snippets and direct answers.

Both shift the goalpost. The win isn't a top-three ranking anymore. The win is being the source the AI quotes verbatim.

Classical SEO optimised for crawlers that index pages and then ranked them. The user clicked, you got the visit, attribution was clean.

AI engines do something different. They retrieve, summarise, and synthesise — often without sending the user to your page at all. Your win is now upstream: getting cited as the source of the answer.

That changes what content needs to look like. Less keyword stuffing, more clearly-structured facts. Less long-form for long-form's sake, more answer-shaped chunks an LLM can lift cleanly.

Modern AI engines consume structured data heavily. Organization, Service, FAQPage, Article, Person, AggregateRating — every one is a hint that says 'this is what this content is about, in machine-readable form.'

On this site we ship JSON-LD for Organization, WebSite, every Service page, every blog Article, the team Person profiles, BreadcrumbList for navigation, and FAQ schema on every service page. Together they form a knowledge graph the AI can lift directly.

The cost is small. Once. For the rest of the site's life, every AI crawler reads it. That's compounding ROI.

Jeremy Howard proposed /llms.txt earlier this year — a markdown file at your domain root that gives LLMs a clean, structured summary of your site, its key URLs, and how to cite you.

It's not yet a W3C standard. Adoption is patchy. But the major AI engines (Anthropic, OpenAI, Perplexity) have all signalled support, and the cost is one file.

We added /llms.txt on day one. It includes a one-line description, the services we offer, the platforms we ship on, the industries we serve, and a 'how to cite us' section that gives the AI a clean canonical sentence to use.

Six months in, our own logs show GPTBot and PerplexityBot fetching it weekly. That's free distribution.

Most robots.txt files are written for Googlebot and a handful of legacy crawlers. AI engines run new user-agents. Our robots.txt explicitly allows GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, anthropic-ai, PerplexityBot, Google-Extended, Applebot-Extended, MistralAI-User, and a few more.

If you don't list them, some crawlers default to safe behaviour and skip your content. That means you don't get cited. Listing them costs nothing and signals intent clearly.

AI engines love FAQ schema because it's literally pre-shaped answers. Each Question/Answer pair is a chunk the LLM can quote directly.

Our service pages each carry 4–6 FAQs in JSON-LD. The questions are written as a buyer would actually phrase them ('How long does an AI automation project take?' rather than 'AI automation timeline'). The answers are 80–150 words — long enough to be useful, short enough to lift verbatim.

When we test our own content in Perplexity, the FAQ answers come back word-for-word. That's the goal.

Backlinks still matter. Domain authority still matters. Real content still matters more than schema theatrics.

But schema, llms.txt, AI-bot allowlisting, and answer-shaped FAQs are cheap, durable, and they compound. They turn an existing content investment into something AI engines can actually surface.

If you're not doing it yet, you're not visible in the next layer of search. And the next layer is starting to take query share. We'd rather build for it now while the cost is one file.