Cluster 2 · How AI works

2.4 Citations and source logic

Why does an AI system cite one source and not another? Nobody outside the platform builders knows the precise weighting — and it differs per channel. But five factors return consistently in platform documentation, research and our own measurements. Understand those, and you know where optimising pays off.

Factors influencing citation

Relevance

The content has to align semantically with the intent behind the question. This is the most fundamental factor.

Authority

In practice we often see sources in citations that are broadly considered reliable — the internal weighting nobody knows. What plausibly contributes: quality of external references, consistency of information across sources, visible expertise on sensitive topics.

Structure and readability

Clearly structured content — with explicit headings, direct answers, logical build-up — is easier to retrieve. Vague, descriptive texts perform worse.

Uniqueness

Content that says something nowhere else expressed quite that way — a specific number, a unique comparison, an original definition — has a higher citation chance. Copied or generic content has less value as a source.

Freshness

In retrieval answers, recent, dated pages appear strikingly often. A documented rule it is not — keeping content current pays off for more reasons than citation alone.

What can you do?