Perplexity on AI Search

An interview with Perplexity AI has offered fresh insight into how AI-powered search works and what it means for SEO professionals adapting to answer-based search engines.

The discussion followed a conversation with Jesse Dwyer from Perplexity, focusing on how search behaviour is shifting as AI tools become more personalised and context-aware.

One of the key points raised was that AI search is no longer a simple competition for a single ranking position. Instead, different users asking the same question may now receive very different answers.

This happens because AI search tools can factor in personal context, such as prior interactions or stored memory, when generating responses. Platforms like Perplexity and ChatGPT do not treat all users as identical.

As a result, visibility in AI search is no longer about appearing in a fixed set of results. Two users may see different sources and interpretations even when the query itself is the same.

Despite these changes, Dwyer noted that many traditional SEO principles still matter. AI search systems continue to rely on underlying indexes to decide which content can be accessed in the first place.

Perplexity, for example, is understood to use link-based signals similar to PageRank, meaning authority, relevance, and credibility still influence whether content is retrieved.

However, what happens after content is retrieved is where AI search begins to differ significantly from classic search engines.

In traditional search, entire web pages are indexed, ranked, and displayed as complete documents. AI tools built on this structure typically gather a selection of top results and then summarise them.

This approach is often described as generative engine optimisation, or GEO, where AI is layered on top of conventional search systems.

By contrast, newer AI-first systems rely on what is known as sub-document processing. Rather than indexing whole pages, they store and retrieve much smaller pieces of information.

These fragments consist of short sections of text that have been converted into numerical data using transformer-based models. The system retrieves tens of thousands of these fragments at once.

The aim is to completely fill the AI model’s context window with the most relevant information available. When that space is fully occupied, the AI has less room to generate incorrect or invented details.

This process helps improve accuracy, shifting the AI’s role from creative generation to factual synthesis. It also explains why AI answers can feel more precise and grounded.

Personal context plays a major role here, as AI systems can use what they already know about a user to decide which fragments to prioritise within the context window.

Dwyer explained that much of Perplexity’s competitive edge lies in how it selects and refines these fragments. Techniques such as query reformulation and adaptive computing help improve relevance.

The result is a search experience that feels more tailored, more detailed, and more accurate than traditional models, marking a clear move towards true answer engine optimisation rather than classic SEO alone.