Skip to main content

What You’ll Learn

  • How file search works behind the scenes
  • How to get the best results from uploaded files
  • What types of questions work well
  • How to troubleshoot when the AI misses something

How It Works

When you upload a file:
  1. Chunking - The file is split into meaningful sections
  2. Embedding - Each section is converted to a vector representation
  3. Indexing - Vectors are stored for fast similarity search
  4. Ready - The file is available for AI reference
When you ask a question:
  1. Your question is converted to a vector
  2. The system finds the most similar file sections
  3. Relevant sections are included in the AI’s context
  4. The AI references specific content in its response
This is called semantic search - it finds content by meaning, not just keywords.

What This Means in Practice

You don’t need to specify page numbers or sections. Just ask your question naturally:
"What does the contract say about early termination?"
The system finds the relevant clause, even if it uses different terminology (like “cancellation provisions” instead of “early termination”).
"According to the research report, what's the projected market growth?"
The system locates the market projections section, even in a 100-page document.

Types of Questions That Work Well

Specific Fact Retrieval

"What was the revenue figure mentioned in the Q3 report?"
"Who is listed as the primary contact in the partnership agreement?"

Analysis Based on Document Content

"Based on the uploaded spec, what are the biggest technical risks?"
"Does our contract allow us to sublicense the software?"

Cross-Document Questions

"How does the pricing in our proposal compare to the competitor analysis?"
(Works when both documents are in the same project.)

Summarization

"Summarize the key findings from the research PDF."
"What are the main recommendations in the consultant's report?"

Tips for Best Results

Be specific in your questions

Better: “What does the contract say about liability limits?” Worse: “Tell me about the contract.”

Reference the file when helpful

"Looking at the uploaded financial model..."
"Based on the competitor report I shared..."

Ask follow-up questions

If the first answer doesn’t cover what you need:
"Is there anything else in the document about this topic?"
"Check section 4 of the uploaded file specifically."

Upload clean documents

  • PDFs with actual text (not scanned images) work best
  • Well-structured documents with headings are easier to chunk accurately
  • Remove cover pages and appendices that aren’t relevant

Troubleshooting

AI doesn’t reference the file

  • Make sure the file uploaded successfully (check the Files section)
  • Be more explicit: “Based on the uploaded [filename]…”
  • The question might not be related enough to trigger file retrieval

AI references wrong section

  • Ask more specifically: “In the section about [topic]…”
  • Rephrase your question with terms used in the document

Large documents feel incomplete

  • Very large files have more content than can fit in one context window
  • Ask focused questions about specific sections rather than “summarize everything”
  • Break very large files into smaller, topic-specific uploads if needed

Still Need Help?

Reach out to us at [email protected] or use the feedback button in the app.