Sean Walsh
Sean is a Director at Intelligency heading up our digital marketing and client services operations. Sean has 15+ years experiencing working both in-house and agency with brands including Lloyds, Alstom, Hitachi, Lufthansa, Viaplay, DFDS Seaways and Mercedes-Benz.
Search behaviour is changing in a way that has direct consequences for how websites should be built and written. AI-powered tools are increasingly doing the searching on behalf of users. Google’s AI Overviews, ChatGPT search and Perplexity all fetch pages, read the content and formulate a response without the user ever clicking through to your site. For that process to produce an accurate, useful answer that includes your business, your content has to work for a very different kind of reader.
This is not entirely new territory. Google has been signalling for some time that content quality and clear audience focus are what separates sites that perform from those that do not. The shift now is that quality alone is not enough. The structure and economy of that content is becoming just as important as its substance.
Addy Osmani, Director of Engineering at Google Cloud AI, published a detailed framework this month on what he is calling Agentic Engine Optimisation, or AEO. The core argument is that websites need to be structured not just for human readers but for AI agents that fetch, parse and act on page content within strict processing limits. It is a practical discipline, not a theoretical one, and the implications for how pages are written and organised are significant.
What AI agents are actually doing when they visit your site
A human reader landing on a page will scroll, skim headings, jump to the section that interests them and ignore the rest. An AI agent operating within a limited context window has to make rapid decisions about what to read, what to skip and what to use. That context window is essentially a ceiling on how much the system can process in a single pass.
If a page is poorly structured, heavily padded or buries its key information deep in lengthy paragraphs, the AI agent may truncate it, skip sections or draw on incomplete information. The result is inaccurate or partial answers attributed to your business, or no citation at all. Neither outcome is useful.
The framework Osmani outlines treats token count as a practical constraint. Concretely, introductory or quick-start sections should sit at roughly 15,000 tokens or fewer. Longer conceptual content should be proportionate to what it genuinely needs to cover. Pages that pad out their word count with preamble, repetitive summaries and hedging caveats are less reliably processed and cited.
Why does this change the case for shorter, tighter content
For years, the conventional wisdom in SEO was that longer content ranks better. That was never the full picture, but it did lead to a culture of padding that is now actively counterproductive. In an AI-driven search environment, a well-structured 900-word page will frequently outperform a sprawling 3,000-word equivalent that takes too long to get to the point.
The issue is compounded by the volume of generic, low-quality content that has flooded the web as AI writing tools have become more accessible. Content that follows a template, repeats itself and adds little original value is increasingly easy for both human readers and AI systems to identify and discount. The bar for what counts as genuinely useful content is rising, not falling.
Length should serve clarity. A page that answers a specific question concisely, uses clear headings and organises its information logically will perform better in AI retrieval than one that inflates its word count in the hope of appearing more authoritative.
What this looks like across different sectors
The practical implications vary by sector, but the pattern is consistent. Most website content has been written primarily for human browsers using a format that does not always translate well to AI retrieval.
- Dental and healthcare practices. Treatment pages often carry a great deal of clinical reassurance and general background before they reach the specific details of cost, process and expected outcomes. That structure works for an anxious patient reading carefully. It is less effective for an AI agent trying to extract a direct answer to ‘what does dental implant treatment cost in London’. Moving key information closer to the top of the page benefits both audiences.
- Aesthetics and beauty clinics. Sites in this sector frequently describe the same treatment across multiple pages with minor variations in wording. That creates ambiguity for AI systems trying to identify the most authoritative version of a page. Consolidating treatment content and making each page more definitive reduces that ambiguity and improves citation potential.
- Education and training providers. Course pages that include structured information about learning outcomes, duration, entry requirements and course content are well suited to AI retrieval. Where they tend to fall down is in using internal terminology, acronyms or jargon that an AI system may not parse correctly.
- Professional services and B2B businesses. Service pages that are vague about what is offered, who benefits and what the process involves are difficult to interpret for human readers and AI alike. Being specific and outcome-focused is a straightforward improvement that helps on both fronts.
Practical changes to prioritise
None of the following requires a site rebuild. These are targeted adjustments that improve how content is processed by AI systems without compromising the experience for human readers.
- Front-load the key information. Answer the question being asked within the first two paragraphs. Extended scene-setting before reaching the substance increases the risk of truncation in AI-driven environments.
- Use descriptive, specific headings. Headings are one of the primary signals AI agents use to understand what a section covers. ‘Our approach’ or ‘How we help’ are much harder to interpret than ‘What the treatment involves’ or ‘How long results typically last’.
- Remove unnecessary repetition. Pages that repeat the same information across multiple sections, or that include lengthy disclaimers before any substantive content, dilute the signal-to-noise ratio for AI retrieval systems.
- Make specific facts easy to extract. Prices, timings, eligibility criteria, qualifications and outcomes are the details AI systems actively try to pull. Burying them inside long paragraphs reduces the accuracy of what gets cited.
- Review structured data. Schema markup for FAQs, services, reviews and local business information provides AI systems with a cleaner, more reliable layer of data to draw from. If this has not been reviewed recently, it is worth prioritising.
AEO extends SEO; it does not replace it
Agentic Engine Optimisation is not a reason to abandon the fundamentals of technical SEO. Strong site architecture, well-structured content, authoritative backlinks and clear crawlability all remain important signals. Research consistently shows that AI Overviews and similar systems draw heavily from pages that already rank well organically. The work already done on search optimisation is not wasted.
What is changing is the standard applied to the content layer. A page that ranks well but communicates poorly in structured, concise terms may start to lose ground to a better-optimised competitor. That gap will widen as AI-driven search becomes more central to how people find information. As AI platforms continue to develop their search and advertising infrastructure, the stakes for being cited accurately and prominently will increase.
The question worth asking of any page on your site is not whether it reads well but whether an AI agent can extract what it needs quickly, accurately and without ambiguity. For many sites, the honest answer is that it cannot, and that is a gap worth closing sooner rather than later.
