Digital Signal · AI discovery
AI discovery: what machines can understand, retrieve and cite
People now find organisations through conventional search, AI features within search, and assistants that answer a question without sending anyone to a website. Your information has to be understandable and retrievable by machines as well as useful to people.
This is the discoverability lens of Digital Signal: what we can see from the outside, what the evidence supports, and where the market is ahead of it.
Discovery now happens before the visit
AI search features retrieve content from several sources, synthesise an answer and cite some of what they used. Google describes AI Mode and AI Overviews issuing a “query fan-out” — a set of related searches across subtopics — to assemble a response.
That rewards content an engine can find, understand and trust for each of those related questions. It also means an organisation can lose visibility in an answer without any change to its own site, because a competitor explains the subject more clearly.
| Aspect | A person on your website | A system retrieving your content |
|---|---|---|
| Entry point | A search result, a link, navigation | A query, often one of several related queries generated for one question |
| What it uses | The page, with its design, imagery and context | Passages of text, structured data and the relationships between them |
| How it judges trust | Brand, design, familiarity | Consistency, clarity, provenance and corroboration across sources |
| What breaks it | Confusing navigation, slow pages, unclear next steps | Information locked in images or PDFs, contradictions, blocked crawling, unclear entities |
| How you can measure it | Analytics on your own site | Partial views: Search Console’s generative AI report shows impressions only, and assistant visibility tools sample answers |
What the evidence supports
The market is full of GEO advice. We sort it by how much weight it can bear, starting with what Google has documented in its guide to generative AI features in Search.
Act on this
Established
Documented by the search engine itself.
- Pages must be indexable and eligible to show with a snippet to appear in Google’s AI features
- Sites must be included in the Search generative AI setting in Search Console, which is the default
- Crawl access, internal links, text content and structured data that matches the page still apply
- Google Search does not use llms.txt or other special AI files, and needs no special schema
Useful practice
Well-reasoned, consistent with how retrieval works.
- Answer related questions clearly, so content serves several parts of a fanned-out query
- Name products, services and the organisation consistently
- Show provenance: who wrote it, when, and on what evidence
- Keep facts current and consistent wherever they appear
Unproven
Widely claimed, without support in primary evidence.
- A universal ranking formula for AI answers
- Rewriting or “chunking” content specifically for AI systems
- Any third-party tool’s insight into Google’s internal ranking
- That llms.txt improves visibility in assistants
What Digital Signal examines
The discoverability signals we can observe
- Indexability, crawl paths and what the rendered HTML actually contains
- Semantic structure and information architecture
- Answerable content: whether key questions are answered clearly, and where
- Entity clarity across products, services, people and the organisation
- Metadata and structured data, and whether they match the visible page
- Provenance and authority signals: authorship, dates, sources
- Duplication, thin content and freshness
- Topic coverage against competitors, and internal linking between related content
- A sample of how AI answers describe you and your competitors for the questions that matter, taken at the time of the assessment
Example finding · Discoverability
Key facts live only in documents
- What we can observe
- Eligibility criteria, specifications and pricing terms are published as PDFs, while competitors publish them as structured, linked pages.
- Why it matters
- Those facts are harder to retrieve, compare and cite, so answers about the subject may draw on competitors’ explanations.
- What it may indicate
- It is consistent with a content model that was never designed to publish these facts as content, or with a regulatory review process built around documents.
- What we cannot know from the outside
- Whether AI systems are citing competitors for these queries at scale, or why the documents are the source of truth internally.
Example finding · Discoverability
One product, four names
- What we can observe
- The same product appears under different names across product pages, support articles and structured data.
- Why it matters
- Machines and people may not connect the pages, which weakens the organisation’s authority on its own product.
- What it may indicate
- It may indicate that product naming is owned in several places, or that older content was not updated after a rename.
- What we cannot know from the outside
- Which name customers actually use, and who owns the decision internally.
SitecoreAI
Where Sitecore fits
Sitecore has added AI-discovery capability in two places.
Content SDK 2.4, released on 10 September 2026, lets Next.js applications serve an llms.txt file managed in SitecoreAI configuration alongside robots.txt, and its Next.js starter templates generate page metadata and Open Graph tags by default. Sitecore’s release note has the detail. From version 2.5, Content SDK for Next.js also renders Schema.org structured data configured in SitecoreAI without custom JSON-LD code; the current release is 2.5.1. Support for llms.txt makes the file easy to publish; it does not make it a Google ranking requirement, and Google states that Search ignores it. The file can still be useful for agents and developer tools that read it.
Signals, powered by Scrunch, launched on 1 September 2026 and has been generally available since 2 October 2026. It monitors how AI platforms cite and mention a brand, surfaces significant changes on the SitecoreAI Strategy page, and suggests page-level changes to apply in Page builder. It requires a Scrunch subscription.
That is continuous monitoring of AI visibility. A Digital Signal is a senior assessment at a point in time: why your content is or is not retrievable, how that compares with your competitors, and what to invest in to change it.
AI visibility is an observable consequence of content quality, structure, authority, technical accessibility and operating discipline.
Explore Digital Signal
The signals we look for
Seven lenses, what each can and cannot tell us from the outside, and how the patterns across them inform a decision.
A sample Digital Signal
An anonymised extract from a real snapshot for a UK travel company, annotated to show where evidence ends and interpretation begins.
Related articles
AI · 21 October 2026
GEO after the hype: what actually makes content visible in AI search?
GEO, AEO and AI search optimisation name one problem. Here is what the primary evidence supports, what is sensible practice, and what nobody can yet prove.
SitecoreAI · 7 October 2026
SitecoreAI Signals and Scrunch: closing the loop from AI visibility to content action
Signals, powered by Scrunch, puts AI-visibility monitoring and page-level recommendations inside SitecoreAI. What that changes for content teams, and the judgement it leaves with them.
FAQ
AI discovery questions
The term was popularised by a 2023 research paper on generative engine optimisation. Google’s own position is that optimising for its generative AI features is still SEO, because they are built on its core ranking and quality systems. Other assistants retrieve differently, so we look at both.
Not for Google Search, which states that it ignores llms.txt. The file can help agents and developer tools that choose to read it, and no major assistant has publicly committed to using it. It is cheap to publish, and worth doing once the content itself is well structured.
Partly. Search Console’s generative AI performance report, rolled out worldwide by 31 August 2026, shows impressions in Google’s AI features by page, country, device and date. It does not separate clicks. Tools that sample assistant answers give a directional view of other platforms.
As a snapshot. We sample how assistants answer the questions that matter to you and your competitors at the time of the assessment, and read that alongside the structural evidence on your site. Answers vary by prompt, time and user, so the sample shows direction. Tracking visibility over time is a job for monitoring tools such as Signals, powered by Scrunch.