Skip to content

Research

Researching how businesses get discovered by AI.

We test how AI systems find, understand, cite and recommend businesses, then publish what we learn.

The programme

What we're studying

Six questions we keep returning to as we test how businesses appear across search and AI systems.

Local AI recommendations

What makes one business appear when someone asks an AI assistant for a recommendation, while another doesn't?

What we're seeing

AI recommendations appear to depend on much more than traditional ranking position. Systems need to understand what a business does, where it operates, whether it is relevant to the exact request, and whether enough trustworthy information exists across the web to support recommending it. Businesses with clear service information, strong third-party signals and consistent entity data tend to give an AI system more confidence than businesses whose online presence is fragmented or ambiguous.

Citation sources

Which websites and sources do AI systems rely on when deciding what businesses to mention?

What we're seeing

A company's own website matters, but it is rarely the whole picture. AI systems can draw on directories, review platforms, publications, industry sites and other pages that independently describe the business. This means citation visibility is partly an authority problem: the more consistently credible sources confirm who a business is and what it is known for, the easier it becomes for an assistant to substantiate a recommendation.

Cross-platform visibility

Why can a business be highly visible in ChatGPT but almost absent from Gemini or Perplexity?

What we're seeing

There is no single AI ranking. Different systems retrieve information differently, use different sources and may weigh those sources differently. A business can therefore have strong visibility on one platform and weak visibility on another. We think the useful goal is not to optimise for one model in isolation, but to build an online presence that is sufficiently clear and authoritative to travel across multiple discovery systems.

Review signals

How much do review volume, rating, recency and reputation influence AI recommendations?

What we're seeing

Reviews appear to matter both directly and indirectly. They provide evidence of reputation, reveal what customers repeatedly associate with a business and create signals across platforms that AI systems can retrieve. Raw review volume alone is unlikely to explain visibility, though. Recency, consistency, subject matter and the authority of the platform carrying those reviews can all change what the wider web appears to know about a company.

Entity consistency

Does consistent information about a business across the web make it easier for AI systems to understand and recommend it?

What we're seeing

Probably, and this is one of the clearest principles behind our work. When a company's name, services, locations, positioning and other core facts are described consistently across its own site and third-party sources, there is less ambiguity for a system to resolve. Conflicting or thin information makes it harder to establish what the business actually is and when it should be considered relevant.

Search and AI overlap

How closely does traditional Google visibility correlate with visibility in AI-generated answers?

What we're seeing

The two are connected, but they are not the same thing. Strong search visibility can make useful information easier to discover and can strengthen the web signals AI systems rely on, but appearing prominently in search does not guarantee recommendation inside an AI answer. AI discovery adds another layer: the system must understand the business well enough, and trust the available evidence enough, to actually include it in the answer.

Method

How we investigate AI discovery

We test real customer questions across major AI platforms, record which businesses are recommended and which sources are cited, then compare those results with signals across websites, search visibility, reviews and the wider web. We repeat tests over time and across different prompts to separate one-off answers from patterns that appear consistently.

As the evidence grows, we use those observations to refine our own work and publish the findings that are useful enough to share. Where we make a stronger claim, we want to be able to show what we tested and how we reached it.

Curious what your own prompt set looks like?

The free audit is the small version of this: your questions, your category, and who is currently being recommended.