Ask the document

Ask the document a question

Ask

Answers come only from this document — not from the internet, and not from the model's own knowledge. If the answer isn't here, it will say so.

Try:

• What does the evidence actually say about verapamil?
• Why do CGRP antibodies fail in chronic cluster headache?
• How do I get high-flow oxygen in Australia?
• What has the patient community worked out before the researchers?
• What does the document say is still unknown?

How this works, and what it will not do

Your question is matched against the document, the closest passages are pulled out, and a language model is asked to answer using only those passages. It cannot search the internet, and it is instructed not to fall back on its own training data even when it is confident it knows the answer.

That restriction is the entire point. Ask a general chatbot about cluster headache and you will get a fluent, plausible answer — and on this subject it will often be wrong in ways you cannot detect from the answer itself: guidance that has since been superseded, the episodic figure quoted for the chronic form, a treatment described as established when the evidence for it actually failed. An answer here is worth something precisely because it could only have come from the document, and every claim links back to the passage it came from so you can check it yourself.

When the document does not cover something, it says so rather than improvising, and offers to add your question to the queue for the next research pass. Those questions are genuinely useful — they are a running list of what people need to know and this reference does not yet say.

Where it can still get things wrong

It can misread a passage. It can attach a figure to the wrong study. It can miss context that sits in a different part of the document from the passage it found. The evidence tags it reports are the document’s own, and a tag describes where a claim came from, not whether it is true.

So treat an answer as the fastest way to find the relevant passage, not as a replacement for reading it. The citations are there to be followed.

And the standing point, which applies here more than anywhere else on the site: this is not medical advice, and the assistant is not a clinician. It will tell you what the document reports about a treatment. It will not tell you what to take. That decision belongs with a doctor who knows your history — see the important notice.

If you are in the middle of an attack, or looking after someone who is, the crisis resources are listed and there is no shame at all in using them.