The Technical Signals That Determine Whether AI Cites Your Website

Being referenced by AI tools in response to questions such as on Google AI Overviews, ChatGPT Search, or Perplexity is not just a matter of having good content. It’s a technical SEO issue masquerading as something else. Websites that get referenced in AI responses share common characteristics that make them more likely to be easily parsed, verified, and attributed by large language models.

When your agency clients ask why their competitors always seem to pop up in AI responses but not them, the answer is seldom “because they have better content.” It’s more likely a technical SEO issue.

Structured Data Is the Clearest Signal You Can Send

Language models learn from content on the Web, but AI citation systems, especially that of Google, require real-time access to information at the moment when the system fetches the webpage in question. When an AI-based system uses a website to make up its mind whether to cite it, structuring information allows the system to understand what the page is all about.

The most relevant schema types when considering AI citation are not the ones that the majority of agencies would consider first. Certainly, there is Article and FAQ schema. However, the most relevant schemas are those that identify the entities’ identity and authority – these are Person schema (with the sameAs pointing to LinkedIn, Wikipedia, or Wikidata), Organization schema with its foundingDate and knowsAbout, and Speakable schema, which is a less utilized schema but which tells AI about the sections on a webpage which can be quoted.

The speakable schema labels certain CSS selectors or XPaths as “the elements of this web page that should be spoken out or cited.” In the documentation of Google, it is mentioned that speakable was created for use with voice assistants; however, the purpose serves equally well for AI in selecting excerpts that can be quoted.

Page-Level Authority Signals That AI Retrieval Systems Use

AI search runs on a chain of trust. An AI-referred web page must undergo an immediate validation process that occurs during retrieval, not during training. The validation process considers numerous technical factors that represent authority:

Author identity. Web pages with a clear author identity, including their name, credentials, and link to their author biography page with structured data, receive consistent citations in AI search more frequently compared to those without authors or organization bylines. From a client perspective, all blog posts and articles must include human authorship and a linked schema-structured author biography page.

Date of publication and last update. An AI system is very sensitive to timeliness, especially for rapidly changing fields such as SEO and AI optimization. Your Article schema’s datePublished and dateModified properties should match reality, and if there have been any updates to the article, the dateModified field should be updated as well. Web pages that have not been updated after being published years ago are not preferred for citation when there are new-state searches.

E-E-A-T signals for entities. Experience, Expertise, Authoritativeness, and Trustworthiness form the evaluation criteria used by the AI Overviews system employed by Google as a citation filter. Technical considerations for E-E-A-T signals include: schema of the author with links to the outside profile, Organization schema with sameAs link to Google My Business, NAP consistency within the website, and internal links that concentrate PageRank on the pages you want to cite.

Site Architecture Decisions That Help or Hurt AI Citation

The structure of the website also plays an important role in terms of the completeness of the AI’s creation of a knowledge graph about the content on the website. Topic clusters, in which there is a pillar page linking to supporting pages and vice versa, provide the AI with information about the comprehensive coverage of the subject by the website, as opposed to having a single article on the topic.

When websites have isolated content that is not clustered, only individual pages get cited because the retrieval system cannot understand the topical authority based on the link graph. With topic clusters on a website, several pages on the same topic can be cited due to topical authority inferred through the link graph.

It implies that when creating topic clusters for agency clients, each content created must belong to a certain topic cluster. There must be bidirectional links between pages and a high-level summary of the cluster’s topic on the pillar page.

Crawlability and Citation: The Most Overlooked Connection

Citation by AI bots requires the page to be available first of all. It seems self-evident, yet the technical considerations involved here cannot always be considered. Thus, pages with disallowed robot instructions, strict rate limits, and render-blocking JS will prevent full page content rendering by an AI crawling system, which means the citation of only the visible page content.

Potential pages for citation via AI must pass the tests for full content availability in the non-JS rendered HTML, no robot instructions that would hinder AI bots in crawling the site, and fast loading within less than 2.5 seconds when correlated with increased crawling frequency.
The list of user-agents for the AI crawling systems must include PerplexityBot (Perplexity), GPTBot (OpenAI), and Google’s extended bots (Googlebot) used for crawling AI-friendly pages.

Many websites inadvertently prohibit access to AI bots through general rules applied to all user-agents. Verification of permissions for AI bots in robots.txt is now a part of the SEO audit process.

The LLMs.txt Opportunity

There is a new protocol in development which is very similar to robots.txt except that it will be tailored specifically for AI-based systems: the /llms.txt file. This plain text file, sitting at the root of the website, will give AI systems the means to know what the key pages on the website are, the subjects which the website speaks with authority about, and how it should go about representing the website’s content.

For agency clients operating in niche markets which face competition, becoming one of the first websites to have implemented the llms.txt file gives them a strategic structural advantage in that the AI systems reading that file get access to the key claims of authority made by the website. Implementing such a file is a low-cost activity which takes under an hour to implement.

Additional Resources: How To Get Cited In Google Ai Overviews | Introduction to structured data markup in Google Search