Pelvic pain is confusing. Your next step doesn’t have to be.

When you ask PelvIQ to research a question, it searches PubMed (the US National Library of Medicine's index of biomedical journals) — your main question and up to two related ones — plus Europe PMC for extra coverage. Guidelines, systematic reviews and randomized trials are searched first, then other study types. Up to 12 papers per search are read. It also checks each paper against Crossref for retractions and corrections, and lists related studies registered on ClinicalTrials.gov.

Full papers are read only when their licence allows AI processing (open licences such as CC BY) — often a minority. Every other paper is read from its abstract, and each result says which. It doesn't search guideline websites or Cochrane's full reviews (their terms don't allow it) or use preprints, so a search can miss relevant work. Coverage is broad, not complete.

The papers are then handled by a team of two AI agents. The researcher reads them, notes exactly what each study tested, sets aside papers that aren't really on the question, and drafts the answer. A separate reviewer then checks that no claim goes further than the studies — rewording or adding a caveat where needed, never removing a claim or making it sound more certain. Each result shows what the team did.

Your history is what you confirm: symptoms, treatments and how you responded, tests, findings. Nothing is added until you confirm it.

In conversation, PelvIQ uses that history (and research you've saved) to make replies specific to you. For a research question, a short list of related, date-free facts shapes the search and how papers are read. Every result shows which facts were used, with a link to correct each.

Research sources only ever receive generic medical search terms that you see and approve first — never your words, name or records.

• Every citation must be a paper the search actually retrieved. Anything else is removed, and claims left without a real source are dropped.
• Quotes are compared word for word with the abstract, or with the full-text passages that were actually read.
• Retracted papers are never used as support.
• Study type comes from PubMed's own labels, not from the AI.
• Obvious mismatches — a different condition, population, age group, or an animal study — are flagged.

These checks confirm a source exists and a quote matches. They don't establish that a conclusion is medically correct: whether a study supports an exact conclusion takes a person reading it, and nothing in PelvIQ is clinically validated.

Each claim carries a certainty level. Where studies disagree, or a result is mixed, that's listed separately — and so is what the evidence leaves unanswered.

When there isn't enough to go on, PelvIQ says so rather than guessing. Research describes groups of people; on its own it can't establish what is happening in your body or give a diagnosis. That takes a clinician who can examine you.

You can edit anything in your history. Research and summaries made before a change are marked for review, and you can research again with your current history.

In a conversation, “You missed something” lets you correct PelvIQ directly — that stays between you and the conversation. “Report a problem” is different: it shares the reply or result, and any context you choose, with the PelvIQ team. That review is about product quality, not clinical validation.

PelvIQ uses Anthropic's Claude models, with instructions specific to pelvic pain: how to take a history, weigh evidence, handle urgent symptoms, and never recommend medication changes. AI-written text is labelled throughout.

It isn't smarter than your clinicians and it can be wrong. What it adds is time and organisation: it keeps your history together, reads the research you ask about, and shows its working so you can question it.