At 9:42am on 7 March 2026, I searched Google for "aao optimisation southampton" on my phone. Position 1 organic: seostrategy.co.uk/llms.txt. The snippet pulled directly from the file: "SEO Strategy is a specialist SEO & LLM optimisation consultancy based in Southampton, Hampshire." Not my homepage. Not a service page. My llms.txt file — the thing I built to tell AI agents what I do and where to look. That is what AAO readiness looks like in practice.
The Shift Nobody Is Talking About Clearly Enough
LLM Optimisation — getting your brand cited when people ask AI systems questions — was phase one. It mattered enormously and still does. But phase two is already here, and it changes the game more fundamentally than most practitioners have acknowledged: AI agents that don’t just answer questions but take actions on behalf of decision-makers.
OpenAI’s Operator, Anthropic’s computer use capabilities, Google’s Gemini agents, Microsoft Copilot Actions, Apple Intelligence — these are not research prototypes. They are production tools that enterprises and consumers are adopting now. When a procurement manager asks an AI agent to "research the top five managed file transfer solutions and compare them on security, compliance and pricing," the agent doesn’t show a list of links. It visits websites, reads documentation, evaluates claims, cross-references sources, and delivers a structured recommendation. When a marketing director asks an agent to "find me an SEO consultant who specialises in healthcare IT and has demonstrable AI visibility expertise," the agent runs the same pipeline — and either your business makes the shortlist or it doesn’t.
AI Agent Optimisation (AAO) is the discipline of ensuring your digital presence is structured, accessible and authoritative enough that AI agents can find you, understand you, trust you and recommend you. It builds on the foundations of LLM Optimisation, Entity SEO and structured data — but adds specific requirements for how autonomous agents navigate, evaluate and compare businesses that go beyond what conversational AI demands.
Agentic AI refers to AI systems that act autonomously on behalf of users — planning tasks, evaluating options and executing decisions without waiting to be prompted at each step. As these systems begin selecting tools, services and providers directly, optimisation must move beyond citations to decision readiness.
The Paradigm Shift: The Buying Funnel Moves Inside the Agent
For twenty-five years, the buying funnel happened across your website and the search engine together. A potential client searched, clicked through to your site, browsed your services, read a case study, compared you against a competitor, and eventually made contact. The search engine was a traffic source. Your website was the conversion environment. The human did the evaluation.
Under AAO, the entire funnel — awareness, consideration, evaluation, shortlisting — happens inside the AI agent before the human ever sees a result. The agent becomes aware of your business through its retrieval process. It considers you against alternatives by parsing your site and cross-referencing sources. It evaluates your credibility through verifiable evidence. It shortlists you (or doesn’t) based on how well your digital presence matches the evaluation criteria. Then it presents the human with a recommendation — often a single recommendation, or at most a shortlist of three.
The human never visited your website during this process. Your role is no longer to attract visitors to a funnel on your site — it is to be the answer when the agent runs its own funnel internally.
How AI Agents Actually Discover and Evaluate Businesses
Stage 1: Query Formulation and Retrieval
The agent receives a task and decomposes it into search queries, retrieving results from search engines, its own knowledge base, and structured data feeds. This is where traditional technical SEO and GEO still matter enormously. If your site doesn’t appear in the agent’s retrieval set, you are eliminated before evaluation begins.
Stage 2: Site Visit and Content Parsing
The agent visits the most promising websites and reads their content systematically — parsing page structure, extracting claims, and building an internal representation of each business. It reads your service descriptions, your pricing, your case studies, your team credentials, your structured data. This is where content architecture and structured data become critical. Comprehensive schema allows an agent to extract information in milliseconds without interpreting marketing prose. A competitor’s site without structured data forces the agent to infer the same information from unstructured paragraphs — slower, less reliable, and lower confidence.
Stage 3: Cross-Reference and Verification
The agent verifies claims by checking multiple sources — what your website says versus what LinkedIn says, what Google Business Profile says, what review platforms say, what case studies claim, and what industry publications mention. Inconsistencies reduce confidence. Corroboration increases it. This is where entity SEO and entity corroboration work pays off — the signals built outside your own site.
Stage 4: Comparative Evaluation and Selection
The agent compares all evaluated businesses against the criteria it has determined for the task, weights relevant factors, ranks candidates, and presents a recommendation. Every stage of this pipeline is optimisable — and each stage has specific requirements that go beyond traditional SEO or conversational AI optimisation.
Push vs Pull: The Architectural Shift That Changes Everything
For two decades, search worked on a pull model: you published content, search engines crawled it on their schedule, and you waited. The push layer is emerging. Technologies like IndexNow, the Model Context Protocol (MCP), and structured data feeds allow you to push information directly to the systems that evaluate your business, rather than waiting for them to come and find it.
For AAO, the push layer matters because AI agents operate on tight timeouts. GPTBot, ClaudeBot, PerplexityBot and other AI crawlers typically timeout within one to five seconds. A slow site isn’t just a ranking disadvantage — it is invisible to the agents doing the evaluating. Your Core Web Vitals investment is not just an SEO and UX improvement. It is an AAO requirement.
On the emerging frontier: Chrome 146 introduced an experimental flag for WebMCP — a browser-level implementation of the Model Context Protocol. When enabled, websites can advertise an MCP server endpoint via a <link rel="model-context-protocol" href="/mcp"> element in the HTML <head>. AI agents using Chrome-based rendering can detect this link relation and query your MCP endpoint directly, accessing structured, real-time data without parsing HTML. See WebMCP: The Fourth Floor Is Being Built for the full picture on what this means commercially and how to assess your readiness.
The Seven Pillars of AI Agent Optimisation
1. Agent-Accessible Content Architecture
AI agents don’t browse — they parse. Every critical piece of information should be accessible within two navigation steps from your homepage. Can an agent find your complete service list in two clicks? Is pricing published and machine-readable? Does your URL structure communicate your topical hierarchy?
2. Comprehensive Structured Data
Structured data is the primary communication layer between your website and AI agents. Organisation schema with knowsAbout properties, Service schema for each offering, FAQPage schema on Q&A content, HowTo schema for processes, and Review schema for social proof. The goal: an agent can extract your complete business profile from structured data alone without reading unstructured prose.
3. Verifiable Claims and Quantified Evidence
AI agents are increasingly sophisticated at detecting unsubstantiated claims. Verifiable, quantified evidence is the currency of agent trust. Case studies should include specific metrics that agents can cross-reference with public data. Name clients where possible. Each verified claim increases agent confidence in your entire profile.
4. Cross-Platform Entity Consistency
Agents verify information across multiple sources. Inconsistencies between platforms reduce trust. Your business name, address, founder name, service descriptions and specialisms must be consistent across your website, Google Business Profile, LinkedIn, Wikidata, industry directories, and the sameAs URLs declared in your Organisation schema.
5. Transparent Pricing and Engagement Models
When AI agents compare vendors, pricing transparency is a competitive advantage. Businesses that publish pricing give agents the data to include them in comparison tables. Businesses that hide pricing behind "contact us" force agents to either exclude them or flag the opacity. At minimum, publish indicative ranges with Offer schema.
6. Push Architecture and Machine-Readable Interfaces
Ensure your robots.txt permits AI crawlers (GPTBot, ClaudeBot, PerplexityBot). Render structured data server-side, not via client-side JavaScript. Optimise page speed to load under two seconds. Consider IndexNow integration for real-time Bing notification. Track WebMCP specification developments — the browser-native MCP layer that lets agents query your endpoint directly.
7. Agent Testing and Continuous Monitoring
Ask ChatGPT, Claude, Perplexity and Copilot to perform the research tasks your potential clients would delegate. Document who appears, how each business is described, what evidence is cited. Run these tests monthly across at least three AI platforms. This is the AAO equivalent of rank tracking.
Who Needs AAO First?
The businesses that should move first are those in considered-purchase categories where buyers research extensively before engaging: B2B technology, professional services, SaaS, healthcare IT, financial services, legal services, and any category where the buying process involves discovery → evaluation → shortlisting → decision. If your sales cycle involves those stages, AI agents will increasingly mediate the first three.
For law firms specifically: the near-term AAO opportunity is not autonomous legal advice. It is agent-ready intake and routing — the agent identifies the right firm, checks service fit, confirms the offence type, and initiates a safe next action like a consultation booking. Agent-ready intake. Human-delivered advice. See Law Firm SEO and what legal regulators are missing for the full framing.
AAO, the AI Discovery Stack, and Why the Order Matters
AAO is most accurately understood as the terminal layer of a single unified discovery pipeline — Layer 5 of the AI Discovery Stack: Understanding (entity recognition), Retrieval (indexing and access), Selection (content structure), Recommendation (brand authority), and Action (agentic decision).
An agent acting at Layer 5 has already passed through Layers 1–4. It has confirmed your entity (Layer 1). It has retrieved your content (Layer 2). It has evaluated your paragraphs as extractable candidates (Layer 3). It has verified your brand authority through external sources (Layer 4). Only then does it act. AAO work that focuses only on Layer 5 signals without ensuring Layers 1–4 are solid will produce no results.
The execution infrastructure that sits above Layer 5 — enabling an agent to act with your business once it has selected you — is WebMCP. Where AAO determines whether you appear on the shortlist, WebMCP determines whether an agent can do anything with you once it gets there. Selection precedes execution.
The AI Provider Selection Pipeline, the CITATE Framework, the OARCAS Framework, and the AI Discovery Stack are not separate frameworks. They are a single system that determines whether your business is found, understood, trusted, selected, and acted upon. Remove one layer and the system fails.
For practitioners covering AEO, GEO and AAO as integrated disciplines, our GEO agency selection guide covers the cross-platform evaluation criteria that apply to any consultancy claiming AI citation capability.
