Asking Questions
Get the most out of Scrapalot by learning how to ask effective questions and choose the right mode for your needs.
Choosing the Right Mode
Choosing What to Search
Knowledge Stacks
The Knowledge Stacks button in the chat toolbar controls what is searched. It works in one of two modes:
AI Routing (the default) — Scrapalot reads your question, picks the collections that look relevant and chooses a retrieval strategy for you. If you pin a collection here, routing searches only that collection. Choose Switch to Manual to take control.
Manual — pick one or more stacks yourself and tune retrieval:
- Similarity — Low, Medium or High: how closely a passage must match to be used
- Number of chunks — how many passages are retrieved
- Orchestrator and Strategy — the retrieval approach, from simple similarity search to graph search, query decomposition or multi-strategy orchestrators
- Knowledge augmentation — whether the AI may add general knowledge alongside cited sources: Strict (only cited sources), Augmented (always add a clearly labeled wider-context paragraph) or Auto (add wider context only when retrieval is incomplete)
- Leave this book out of this conversation — exclude a single book without changing the stack
Switch to AI Routing returns to automatic mode.
@-mentions
Scope a question from the chat box itself:
| Type | Result |
|---|---|
@ | Pick an individual file |
@/ | Pick a collection |
@/name/ | Browse inside a named collection |
Mentioned files and collections appear as chips above the input. Questions about a document you @-mentioned go straight to that document.
Seeing what was searched
The Search strategy panel on an answer shows which strategy ran (for example similarity search, hybrid search, graph search, query decomposition, step-back reasoning, HyDE or multi-query), which sources it used, and why it was chosen.
Modes and Options
The chat toolbar's options menu holds the modes that change how the assistant works.
Standard chat with your documents
Best for: Questions about your uploaded documents
How it works:
- Scrapalot retrieves the passages that bear on your question (semantic, keyword and graph search combined)
- Short matches are expanded to their surrounding section where that helps
- The model answers from those passages
- Every claim carries a citation — click it to jump to the exact passage
Example questions:
- "What is the main conclusion of this paper?"
- "Summarize the methodology section"
- "What does the document say about climate change?"
Simple Search (web)
Best for: Questions that need up-to-date information beyond your own documents
Adds live web results, so the answer can combine your documents with current sources. Web sources are cited like document sources.
Deep Research
Best for: Complex topics that need a structured, cited report
Runs a multi-step research workflow: it plans the topic, searches your documents, the web and academic sources over several rounds, validates sources, and writes a long report with citations.
How it works:
- Set Research Breadth (Focused … Comprehensive) and Research Depth (Surface … Deep)
- In Research Setup, optionally choose a template — Scientific Research, Literature Review, Concept Exploration, Fact Verification, Comparative Analysis, Alternative Research or PRISMA Review — or let the AI auto-detect the approach
- Answer the clarification questions, or skip them
- Review the research plan
- Follow the live activity timeline (planning, retrieval, web, verification, synthesis)
- Read the report — executive summary, sections, key conclusions, recommendations and bibliography — and download it as Markdown or visual HTML
Research depth tiers (shown in Research Setup):
| Tier | Scope | Typical time |
|---|---|---|
| Quick | ~50 papers | 2–3 min |
| Standard | ~150 papers | 8–10 min |
| Deep | 500+ papers | 30+ min |
Tick Let it work on its own in the background to research on the server for up to about 30 minutes; you can close the tab and follow progress in the header.
Deep Research is part of the Pro plan and above.
See: Deep Research Architecture
Thought Partner
The assistant never answers — it asks you 3–5 focused questions to sharpen your own reasoning.
AI Tutor mode
The assistant teaches Socratically, guiding you with questions and step-by-step explanations instead of giving the final answer straight away. On a collection with a detected topic structure, it walks you through lessons, checks and quizzes.
Incognito
The conversation is not saved to your second brain: Scrapalot neither reads what it knows about you nor learns anything new from it.
AI Thinking
For reasoning models, shows the model's step-by-step thinking as it answers.
Chat without documents
With no Knowledge Stack selected and nothing @-mentioned, the assistant answers from the model's general knowledge. Answers without citations to your documents are not grounded in them.
Other Ways to Ask
- Voice mode — talk to the assistant and hear it answer; it can search your library and the web while you talk (Pro and above)
- Ask about a selection — select text in an answer or a document and choose Ask Scrapalot to trace it back to its sources
- Annotations — in the PDF and EPUB viewers, highlight a passage and use Explain, Similar (similar passages in your library) or Cite
- Prompt Templates — reuse saved prompts from the chat toolbar
- Model settings — per-conversation model instructions, temperature and other generation parameters
- Attach Files — add a file to the conversation directly
Question Types & Techniques
Factual Questions
Good examples:
- "What is the definition of X in this document?"
- "What year was the study conducted?"
- "Who are the authors of this paper?"
Tips:
- Use specific keywords from your documents
- @-mention the document if you know where the answer is
Summary Questions
Good examples:
- "Summarize the main points of this document"
- "Give me a brief overview of chapter 3"
Tips:
- Specify the level of detail and format (bullet points or prose)
- For whole books, the document's Book Summary and chapter summaries are often faster
Comparative Questions
Good examples:
- "Compare the methodologies used in these two papers"
- "What are the similarities between documents A and B?"
Tips:
- @-mention each document, or select the collection that holds them
- Turn on Simple Search to compare against external sources
Analytical and Research Questions
Good examples:
- "What are the implications of these findings?"
- "What is the current state of AI safety research?"
Tips:
- Use Deep Research for anything that needs many sources
- Pick a research template that matches the task
- Use Standard or Deep tier for broad topics
Getting Better Answers
Do's and Don'ts
Do:
- Be specific and clear
- Specify the format you want (list, paragraph, table)
- Mention time periods, regions, or specific aspects
- Use follow-up questions to dig deeper — the conversation keeps its context
- Check citations to verify information
Don't:
- ❌ Ask vague or overly broad questions
- ❌ Expect answers about documents you haven't uploaded
- ❌ Use Deep Research for simple lookups
- ❌ Ask multiple unrelated questions in one message
Writing Effective Questions
Instead of:
"Tell me about this"
Try:
"Summarize the main methodology and findings from the 2024 climate report"
Instead of:
"Compare these"
Try:
"Compare the cost-effectiveness and implementation timelines of the three proposed solutions in documents A, B, and C"
Working with Answers
Each answer has actions to Read aloud, Copy content as Markdown, Regenerate as new branch, rate it (Good response / Bad response) and view response metrics. To share a whole conversation, use Share Conversation — anyone with the link can view it.
Troubleshooting
No relevant sources found
Causes:
- The question doesn't match the document content
- Similarity is set too high
- Documents are not fully processed
- The wrong stack is selected
Solutions:
- Rephrase your question using terms from your documents
- In manual mode, lower Similarity to Medium or Low
- Check the document's processing status
- @-mention the document you mean
Answers are too generic
Solutions:
- In manual mode, increase Number of chunks
- Make your question more specific
- @-mention specific documents instead of a whole collection
- Set Knowledge augmentation to Strict to keep the answer to your sources
Deep Research is taking long
Deep Research time depends on the tier: Quick takes a few minutes, Deep can take half an hour or more. Follow the activity timeline to see the current phase, or run it in the background and come back later.
Related Topics
- Deep Research Architecture - How Deep Research works
- RAG Strategy - Understanding retrieval strategies
- Uploading Documents - Prepare your document library
- Features Overview - All Scrapalot features
Tip: Start with simple questions to understand your documents, then use Deep Research for complex analysis when needed.