Codegarden 2026: the technical picture, and what it means for your site
Something shifted at Codegarden this year. The conversations between developers were less about features and more about how the web is being read, by humans and by machines. Specifically, by AI.
It was not doom and gloom. It was practical. What do we need to change about how we build sites, structure content, and think about visibility in a world where a growing portion of your audience never actually visits your website at all?
Here is what we took away from the technical sessions, and what is worth doing about it now.
The way people find websites is changing
For years, SEO meant ranking on Google. Page one, position one, done. That is still relevant, but it is no longer the whole picture.
More and more, people are asking ChatGPT, Perplexity, Gemini or Claude a question instead of typing into a search box. And those tools do not serve up a list of links. They give an answer, often citing a source or two, often without the user clicking anywhere at all.
This is not a niche behaviour. It is accelerating fast. Younger audiences in particular are using AI tools as their default starting point for research, product discovery, and decision making. For businesses that rely on organic search traffic, that shift has real implications.
The question is not whether this is happening. It is whether your website is positioned to benefit from it or be bypassed by it.
SEO is not dead, but it has changed
Search engine optimisation still matters. Google is still the dominant starting point for a lot of queries, and ranking well is still worth pursuing. But the goalposts have moved.
Google's own AI Overviews now sit above the traditional search results for many queries. That means even a position one ranking can be pushed below the fold by an AI-generated summary. And that summary is pulling from sources Google considers authoritative, well-structured, and clear.
The implication is that the old approach of targeting keywords and building links is necessary but no longer sufficient. The sites that win in this environment are the ones that are genuinely useful, clearly structured, and easy for both humans and machines to understand.
AEO: optimising to be the answer
AEO stands for Answer Engine Optimisation. Where SEO focuses on search rankings, AEO focuses on being the source an AI tool cites when someone asks a question your business should own.
Think about the questions your customers ask before they buy. What does X cost? How does Y work? What is the difference between A and B? If your website answers those questions clearly and directly, you are a candidate to be cited. If your content is vague, buried in jargon, or written primarily for keyword density, you are not.
The fundamentals are not complicated, but they do require intention. Clear, direct content written to answer specific questions. Proper heading structure so the hierarchy of information is obvious. Schema markup so machines can parse what your page is actually about. Authoritative, well-linked pages that AI systems learn to treat as trustworthy sources over time.
One thing worth noting is that AEO rewards depth. A single thorough page on a topic tends to outperform several shallow ones. If you have been building out thin category pages or duplicating content across multiple URLs, that is worth reviewing.
A lot of Umbraco sites already have good bones for this. It is usually a case of auditing what is there and making targeted changes rather than a full rebuild.
GEO: making your content machine-readable
GEO, Generative Engine Optimisation, is closely related to AEO but focuses specifically on how large language models parse and represent your content when generating responses.
The question GEO asks is: if an AI tool scraped your website today, would it come away with an accurate, complete understanding of who you are, what you do, and why you are credible?
That means thinking about entity clarity. Is it obvious from your homepage copy who you are and what you do, without requiring prior knowledge of your brand? AI tools do not have context. If your site assumes the reader already knows you, a model will struggle to summarise you accurately.
It means structured data. Schema markup for your products, services, people, locations, FAQs, and reviews gives machines a reliable way to read your content without having to infer it from prose. This is not new technology, but most sites are still under-using it.
It means content depth. AI tools prefer sources they can learn from. Pages that go three levels deep on a topic are more likely to be cited than pages that skim the surface.
We built Glint and Beacon specifically to analyse this. Glint surfaces the technical visibility gaps, missing schema, broken structured data, weak entity signals. Beacon looks at GEO readiness alongside traditional SEO signals and shows you where you stand relative to how AI tools are likely to read your site.
LLMs and your website: the technical basics
There is a relatively new file convention called llms.txt, similar in concept to robots.txt, that tells AI crawlers what they are allowed to read and how to interpret your site. Adoption is still early, but having it in place now is sensible and straightforward to implement.
Beyond that, the key question is whether your content is clear enough for a language model to understand without context. If a page requires a reader to already know your company to make sense of it, an AI tool will not know what to do with it either. Writing for clarity has always been good practice. Now it is also a technical requirement.
Site speed and crawlability matter here too. AI crawlers behave differently to Googlebot, and a site that is slow to load, poorly linked internally, or structured in a way that makes navigation difficult is harder for any automated system to index accurately.
AI on your website: practical ideas worth considering
Codegarden reinforced that AI on the front end of a website is becoming expected rather than experimental. Here are a few things worth thinking about for your own site, roughly in order of complexity.
AI-powered search replaces a basic keyword search with one that understands intent. A user types a question in natural language and gets a relevant result rather than a list of keyword matches. For sites with a large content library or product catalogue, this is often one of the highest impact changes available.
An AI assistant trained on your own content can answer questions about your products, services, or documentation without the user having to dig for it. Not a generic chatbot, but one that actually knows your business. Done well, this reduces support load and keeps users on your site longer.
Content generation inside the CMS is now live in Umbraco via the AI Copilot. For teams producing a high volume of content, whether that is product descriptions, news articles, or landing pages, the time saving at the editor level is real. The approval workflow stays in place, so quality control does not change.
Personalised content uses signals from how someone is browsing to surface more relevant pages or calls to action. If a user has been reading about a specific product category, showing them related case studies or a targeted CTA rather than a generic one tends to convert better. Umbraco has the infrastructure for this already.
Playing our part
We have been looking at all of the above through Eddi, our AI content layer for Umbraco. If any of it looks relevant to your setup, we are happy to talk through what is realistic and where it makes sense to start.