Citations
AI answers that cite the source — and scroll you to it.
Most AI tools sound confident and hope you don't check. Sidenote does the opposite: every answer is grounded in a passage from your document, and one click takes you to the exact sentence. If a claim can't be backed up, you never see it.
The problem
A confident answer isn't the same as a true one.
The failure mode of document AI is the plausible fabrication — a fluent answer that cites a page that doesn't say that, or no page at all. When the stakes are a runbook, a contract or a research claim, “probably right” isn't good enough.
Sidenote is built around verification, not vibes. The model can only show you a citation it can prove against the source — and it makes that proof one click away.
How it works
How Sidenote's citations work.
The differentiatorValidated, not vibesUnsupported citations are dropped server-side before you see them. Scrolls the real documentThe PDF, wiki page or article you're on — not a copy, not a screenshot. Highlights the exact spanAmber highlight on the cited sentence — multi-span when the answer needs it.
Scroll-to-source, not just a page number.
Plenty of tools print a citation. Far fewer take you to it. A page number still leaves you scanning; a footnote still leaves you trusting. Sidenote closes the loop — the citation is a link into the live document, landing on the exact sentence and highlighting it.
Everywhere you read
Cited answers across every source.
The same citation engine works wherever Sidenote reads — browser PDFs and articles, your wikis, your docs, and files you upload. Pick a source:
FAQ
Common questions about citations.
Does Sidenote hallucinate?
Every answer is checked server-side against the passages actually retrieved from your document. Any claim that can't be matched to a source has its citation dropped — so you only ever see citations the model could genuinely back up. Sidenote can still say it doesn't know, but it won't dress an unsupported claim up as a cited fact.
How are the citations verified?
When the model answers, it must attach the source chunk and the exact quote for each claim. The server then normalises that quote and matches it against the real document text — handling PDF artefacts, curly quotes, ligatures and line-break hyphenation — and emits the verbatim passage it found. Quotes that don't match anything in the source are dropped, never invented.
What does “scroll to source” actually mean?
Instead of a footnote or a bare page number, clicking a citation scrolls the live document — a PDF, a wiki page, an article — to the exact sentence and highlights it in amber. You verify the claim in context, in about a second, without leaving what you're reading.
Which documents does this work on?
Everything Sidenote reads: browser PDFs and web articles, Confluence, Notion, Google Docs, SharePoint and OneDrive, and files you upload (PDF, DOCX, PPTX, TXT, MD). Scanned PDFs are OCR'd first so even image-only documents can be cited.
What is multi-span highlighting?
When an answer draws on more than one sentence — say the first and last sentence of a paragraph — Sidenote highlights each cited sentence on its own and leaves the unused middle plain. You see exactly which parts of the source grounded the answer, not a vague block.
Is this really more trustworthy than other AI tools?
The claim is deliberately narrow: Sidenote won't show you a citation it couldn't match to the source, and it makes checking the source trivial. It can still be incomplete or wrong about interpretation — but you're never more than one click from the evidence, which is more than most tools offer.
See it for yourself
Don't trust it. Check it.
Add Sidenote to Chrome, ask a document a question, and click a citation. Watch it scroll to the line that proves the answer.
7-day Pro trial · No card required · Free tier forever