llms.txt and Structured Data: GEO Checklist 2026

If AI systems can’t read your content properly, they won’t cite you either. Good writing alone isn’t enough. It’s the technology that determines whether your answers are even seen. This checklist will bring your site up to scratch.

Many content teams invest a great deal of time in producing good copy, whilst overlooking the fact that an AI system cannot accurately process this content in the first place if the technical foundation is lacking. This checklist goes through the various stages that ensure content can actually be read, understood and, ultimately, cited – from the simplest to the most frequently underestimated aspect.

What llms.txt is and what it is intended to do

llms.txt is a voluntary Markdown standard, which is stored as a file in the root directory of a website and is intended to provide AI systems with a structured overview of a page’s key content. You can find the full specification at llmstxt.org.

The idea behind it is simple: rather than an AI system having to laboriously click its way through an entire website, the llms.txt file provides a concise, machine-readable summary of the most relevant pages and resources. This is particularly useful for extensive documentation or product pages, where the content that is actually relevant might be buried deep within a complex, branched structure.

Important to note: llms.txt is not an official standard that is officially supported by the major AI providers. Google states in its documentation on AI features in search It is explicitly clear that no additional AI text file is required for inclusion in AI Overviews. You should therefore treat llms.txt as an experimental, optional supplementary signal, not as a mandatory component without which nothing works.

Nevertheless, the file can be worthwhile, particularly for large websites with a lot of technical or complex content. Providers such as Anthropic or the documentation tool Mintlify are already actively using llms.txt to make their own content more accessible to AI systems. Anyone setting up an llms.txt file should view it as a complement to a solid technical foundation, not as a replacement for it.

The structure of an llms.txt file is based on a simple logic: a brief description of the service, followed by a structured list of the most important subpages with concise explanations of what each one contains. It is therefore similar to a sitemap, but in a text format that is particularly easy for language models to process, rather than the XML format of traditional sitemaps. Anyone who already has a well-maintained sitemap and a clear information architecture can often produce a first draft of the llms.txt file in just a few hours, without having to start from scratch.

Even if an AI provider is not currently analysing the file, setting it up causes no harm, as it is simply stored in the root directory as an additional, optional resource. At most, there is a risk of investing time in maintaining a file that is currently only taken into account by some of the AI systems. For most companies, the effort is still worthwhile, as it is minimal compared with the other items on this checklist.

Using Structured Data Correctly for AI Citations

When it comes to AI citations, the schema types ‘Article’, ‘FAQPage’ and ‘Organisation’ are particularly important, as they present content in a format that search engines and AI systems can clearly categorise. Structured data is therefore not a new topic in GEO, but a tried-and-tested SEO foundation that is becoming even more important for generative systems.

The official guide to getting started is provided by Google’s introduction to structured data. For articles, the ‘Article’ type is suitable for clearly marking the author, publication date and main topic. For question-and-answer sections, the FAQPage type is suitable for marking up individual questions and their answers as standalone, extractable units. At the organisational level, the Organisation type provides basic facts such as name, logo and contact details in a structured format.

Mixpanel’s well-designed Q&A section is a good example of what a layout suitable for an FAQ page might look like.

Each question is clearly displayed on its own, with the answer directly linked to it and expandable. It is precisely this pattern that can be mapped one-to-one using a suitable FAQPage schema, so that both people and search engines encounter the same clear structure.

It is important not to view structured data in isolation. Technically correct schema markup on a page with thin, unhelpful content is of little use, because the markup merely describes what is already there; it does not create any substance. Conversely, an article with strong content but without appropriate markup is more difficult for systems to categorise unambiguously. The two go hand in hand: good content with a clear, machine-readable description of what it actually is.

In most CMS systems, a one-off, clean configuration at template level is sufficient for the technical implementation, rather than having to mark up every single page manually. For example, anyone working in Storyblok or a similar system can embed Article, FAQPage and Organisation markup directly into the relevant components, so that every new page is automatically marked up correctly without editors having to worry about technical details.

Crawlability and rendering: the unseen prerequisite

Crawlability comes first, because even perfect structured data is ineffective if a system cannot reliably access or render the page in question. In practice, this prerequisite is often underestimated because it operates invisibly in the background – or fails to operate at all.

Particularly in the case of websites that rely heavily on JavaScript, it is worth checking carefully whether content is visible even without full client-side execution. Many crawlers, including those used by AI systems, process JavaScript only to a limited extent or with a delay. Server-side rendering – or at least a clean pre-rendering of the most important content – provides a reliable foundation in this regard.

Equally important is a clean robots.txt configuration that does not inadvertently block relevant AI crawlers. A common, easily overlooked error occurs during website relaunches: generic robots.txt rules from a staging environment are accidentally carried over to the live site, where they block entire sections from being accessed by any crawler, including AI systems. A quick check of the robots.txt file immediately after every launch should therefore be included in every deployment checklist, regardless of how experienced the team in question is.

Loading time also plays a role that is often underestimated. Crawlers and rendering services usually operate within a limited time budget per page. A page that only loads its actual content after several seconds runs the risk that this very content will not even be captured during rendering. Anyone suspecting bottlenecks here should try visiting their own page with JavaScript disabled and check how much of the actual content remains visible – a simple but revealing practical test.

Page structure, headings and ‘answer-first’ snippets

A clear page structure with unambiguous headings and questions answered directly increases the likelihood that content will be extracted and cited by AI systems. Structure is therefore just as important a ranking factor for GEO as it is for traditional SEO.

The extent of the impact is illustrated by a Analysis by Search Engine Land: Structured content such as lists and tables is shown to be cited as evidence in AI responses around 2.5 times more frequently than comparable content without this structure. These figures highlight why it is worth making the effort to present content clearly, regardless of the page’s actual topic.

In practical terms, this means: one H1 per page, H2 sections clearly organised by topic and, where appropriate, lists or tables for comparative content. The first sentence after each heading should already contain the section’s key message, rather than taking several sentences to get to the point. This ‘answer-first’ approach makes it easier both for people to skim the text and for systems to extract individual blocks of information.

It is also helpful to keep paragraphs deliberately short and to focus on just one idea per paragraph. Long, convoluted paragraphs containing several statements are more difficult for an AI system to unambiguously assign to a single aspect, which is why content teams would be well advised to critically review their current text lengths and, where appropriate, break them down into clearly defined units.

The timeliness of the content also belongs in this section, even if, strictly speaking, it is not purely a structural issue. Generative systems favour information that is recognisably up to date, particularly when it comes to dates, prices or statistics. A page containing an out-of-date date or an obsolete figure is more likely to be avoided by a system than one that is visibly kept up to date. A fixed schedule for checking key figures and data is therefore just as much a part of the technical foundation as clean markup.

The technical GEO checklist to tick off

The technical basis for GEO can be summarised in a concise list that can be worked through regardless of company size or CMS.

1. Check crawlability: review robots.txt, render important content on the server side

2. Implementing Structured Data: Article, FAQPage and Organisation as a basis

3. Streamline page structure: one H1, clearly structured H2s, answer-first paragraphs

4. Use lists and tables specifically where content is presented in a comparative or enumerative manner

5. Add to llms.txt if necessary, particularly for large, technical websites

6. Ensuring information remains up to date: regularly check for out-of-date data and figures relating to specific years

7. Make AI referral traffic distinguishable separately in the Analytics setup

In our own practice, we consistently apply precisely these layers when implementing Storyblok, from neatly configured structured data right through to a page structure designed with a ‘response-first’ approach from the outset. This technical foundation is the part of GEO that offers the least creative scope but yields the most immediately measurable results.

A practical tip for using this list: you don’t have to work through it all in a single go. It makes more sense to start with the ten to twenty most important pages on a website, put them through all seven points thoroughly, and only then apply the checklist to other sections of the website. This way, you’ll quickly see visible progress on the pages that already account for the largest share of traffic and potential AI citations, rather than getting bogged down in the details.

For more in-depth information on the strategy behind this checklist, you will find The GEO Framework in six steps, and you can read about why internal linking plays an often underestimated role in this in Why internal linking matters for GEO too. We’ll help you with the technical implementation in your own setup via Technical GEO implementation with pechschwarz.

In the end, this work pays off twice over. A technically sound website not only helps with GEO, but generally improves traditional SEO as well, because both disciplines are based on the same fundamentals: crawlability, a clear structure and unambiguous markup. Once you’ve gone through this checklist carefully, you’ll simply need to maintain it – there’s no need to reinvent the wheel.