The short version: we run deep research once, store every finding with its sources, publish a free report as a view over that research, and let you buy a precise, sourced answer to anything the report doesn't spell out — generated only from what we've actually verified.
Each topic is worked once, deeply, by a multi-agent research pipeline before anything is published. Agents don't write prose first — they extract structured claims, entities and numbers, each tagged with a source, a publication date, and a confidence level. Findings go into a database, not a document.
The free report you read is a curated view over that database — a way of presenting the corpus, not the corpus itself. That's why a fragment in the report can be traced back to the exact claims and sources behind it.
While reading, select any span of text and two options appear:
Most fragments also carry two or three seeded questions — things earlier readers found worth asking, or questions the research surfaced as natural follow-ups. They exist to show you, concretely, what the corpus is good at answering before you spend a credit.
The answer engine never sees the open internet and never falls back on the model's general knowledge. It sees only the retrieved claims and sources from our corpus that are relevant to your fragment and question. Every sentence in the answer is expected to trace to a specific claim, and every claim shows the sources behind it — how many, how primary, and how recent.
When sources disagree, the answer says so rather than picking a side silently. When a claim has a known expiry date, that date is shown, too — a verified number from three years ago is not the same as one from last month.
Every visitor gets 2 free answers, no account required, and 3 more as a one-time signup bonus. After that, a “go deeper” costs 1 credit and a question costs 1 credit, with a possible +1 credit surcharge for an unusually long or claim-heavy question. The exact cost is always shown before you accept — never deducted first and explained after. See credits & pricing for packs.
If your question goes beyond what we've researched, the system says so instead of guessing — “not researched yet.” You are not charged for a refusal. Your question is logged to our research queue, which is one of the main inputs into what we research next. An unanswered question today can become an answerable one for every future reader.
Each topic has its own refresh cadence based on how quickly it decays. Every report and every source shows a “last researched” date, so you can judge for yourself whether a number is current rather than take that on faith.
Answers are informational, not professional advice, and — like any AI system — can still be wrong even when every source it cites is real; check anything you intend to rely on. If an answer is off, tell us — per-answer refunds are available on request. See AI disclosure for the full explanation and refunds for the policy.