Why is Intercom AI using outdated help center content?

It can happen. Intercom AI relies on indexed versions of your knowledge base, not always the live page state.

Correct. Most AI support systems use retrieval pipelines that periodically re-index content. If the indexing process hasn’t refreshed, older embeddings may still be referenced.

That’s where retrieval-based systems can become risky. The AI doesn’t know which article sections are operationally sensitive. It simply ranks what appears most relevant from its indexed dataset.

If older versions have stronger semantic similarity or more historical weight, they may outrank the updated version temporarily.

Exactly. Retrieval pipelines prioritize similarity, not recency, unless explicitly configured otherwise.

In channel-native setups, control is limited. More advanced architectures introduce validation layers that approve or constrain which content sources the AI can reference, especially for high-risk categories like pricing or billing.

That’s why some teams move toward orchestration-based AI systems that separate retrieval from decision logic. Instead of trusting similarity ranking alone, they validate content freshness and enforce source rules before responses are generated.