AI that answers from your own papers, with a citation for every claim
If you want an AI to answer from the papers you have actually read, and to show the passage behind each claim, here is how grounded chat works, how it differs from NotebookLM, Elicit and SciSpace, and what happens to your unpublished work.
Short answer: use a tool that answers from your own library rather than from what a model learned in training. In ScholarThread, chat searches the papers and notes in your library, answers from the passages it finds, and puts a numbered citation on each claim. Click a citation and the paper opens at the passage, so you can check the answer the way you would check any source.
General chatbots are good at many things, but ask one about the papers on your reading list and it either answers from whatever it absorbed in training or tells you it cannot see them. Neither helps in the middle of a literature review.
Three kinds of AI research tool
They are often lumped together, but they do different jobs.
| Your question | The kind of tool | Examples |
|---|---|---|
| "What has been published on this topic?" | Literature search and summary over the wider literature | Elicit, SciSpace, and the OpenAlex search inside ScholarThread |
| "What do these documents I uploaded say?" | A notebook grounded in the sources you add to it | NotebookLM |
| "What do the papers I have read say, and where exactly?" | Chat grounded in your whole library, next to your highlights and notes | ScholarThread |
Many researchers use one from each row: a search tool to find papers, and a library tool to read and question the ones they keep.
How ScholarThread's chat answers
- It searches your library first. Each question pulls the relevant passages from across your papers and notes, not just the paper you have open.
- Every claim carries a citation. Citations appear as numbered chips, and the sources are listed under the answer. Hover a chip to preview the passage; click it to open the paper there.
- It can stay on one paper. Start a chat about a single paper and it answers from that paper alone, which is useful for interrogating one argument closely.
- It tells you what it cannot see. If a paper is still being indexed, or only its abstract is available, the answer says so instead of guessing. It will not invent the contents of a paper you have not added.
- You choose the model. The standard models are Anthropic Claude and Amazon Nova. A few optional OpenAI models are also available after you consent, because they may run outside the EU.
The grounded chat guide walks through it in detail.
What happens to unpublished work
Researchers rightly ask this before uploading a draft, a dataset description or a paper under review.
- Storage: your library, notes and PDFs are stored in AWS's Stockholm region (eu-north-1).
- AI processing: chat and assisted notes run on Amazon Bedrock in AWS's EU regions (primarily Frankfurt), one request at a time. The optional global models are the exception, and they stay locked until you agree to a consent form that explains it. You can withdraw consent at any time.
- No training: we do not use your content to train models, and AWS does not use content sent to Bedrock to train models or share it with model vendors.
- On your device: on the iPhone and iPad app, search by meaning is computed on the device, and that content does not leave it.
- No advertising on any plan.
The privacy policy is the authoritative version.
Academic integrity
The assistant helps with research; it is designed not to write coursework for you. It has a set of safeguards, including a check on requests before they reach the model, and it keeps a disclosure log of AI assistance that you can export as Markdown or PDF to show your supervisor or examiner. See how the assistant stays on the right side of academic integrity.
What it costs
Chat is paid in AI credits. Pro includes a monthly allowance, and on Free or Pro you can buy extra credits as one-time top-ups that never expire. The pricing page lists what each model costs in credits.
Use your own AI assistant instead
If you already work with Claude Code, Claude Desktop or another assistant that supports MCP, Pro includes an MCP server that gives it scoped access to your library and concept notes. See connecting your AI assistant.
Questions people ask
Does it search the internet?
Chat answers from your library. To find new papers, use the OpenAlex tab in Search and import what you want; chat can then answer from them. See one search box for your library and the open web.
Can it read my handwriting?
No. Ink you write with Apple Pencil stays as drawn, and chat never sees it. Highlights and typed notes are part of your library.
What if a paper has no full text?
The paper's details show how much of it is indexed, with a button to index it or retry. A chat about a partly indexed paper says so in its header.
If something here is out of date or wrong, tell us.