Troubleshooting Document Uploads | TurnkeyGTM Support

Document stuck in processing

Documents typically finish processing within a few minutes. If yours seems stuck:

Check the processing stage. Visit the Upload page to see which stage your document is in (parsing, extracting, embedding, etc.). Some stages take longer than others.
Give it more time. Large documents, complex PDFs, and image-heavy files take longer to process.
Refresh the page. The status indicator may not have updated.
Re-upload the document. If the document has been stuck for more than 15 minutes with no progress, delete it and upload it again.

If the problem persists after re-uploading, the file may have a formatting issue. Try converting it to a different format (for example, save a DOCX as a PDF) and upload the new version.

Re-indexing after a processing failure

Re-indexing unchanged text reuses matching stored chunks, including chunks saved before an embedding failure. It preserves existing chunk IDs and metadata; it does not delete older data to start over.

If processing fails with “Ambiguous existing document chunks” or “Conflicting document chunk identity”, contact support with the document ID. The retry has stopped rather than choosing or overwriting a conflicting record. Repeated retries will not resolve this condition.

This retry protection is not a cleanup tool: existing duplicates are not removed. Syncing changed source text can retain historical chunks, so the latest attempt's chunk count need not equal all stored chunks.

No entities extracted

If your document finishes processing but produces no entities, here are some possible causes:

Document is too short. Very short documents may not contain enough context for meaningful extraction. Try combining related short documents into a single file.
Format issues. Scanned PDFs (images of text rather than actual text) may not parse correctly. Use OCR software to convert scanned documents to searchable text before uploading.
Unstructured content. Heavily visual documents like slide decks or infographics may not contain enough extractable text.
Content doesn't match entity types. Upstream looks for specific types of intelligence (Pain Points, Objections, Feature Requests, etc.). If your document doesn't contain this kind of content, fewer entities will be found.

What to try:

Paste the key text directly instead of uploading the file.
Break long documents into smaller, focused sections.
Add context or headings to help Upstream understand the content.

Low confidence scores

Low confidence scores (below 50) mean Upstream found possible entities but isn't certain about them. This usually happens when:

The language is vague or ambiguous.
The entity is implied rather than directly stated.
The surrounding context is limited.

How to improve confidence:

Use clear, specific language in your documents.
Provide more context around key points.
Upload additional documents that reinforce the same themes — repeated mentions across multiple sources increase confidence.

Review low-confidence entities in your Review Queue and approve the ones that are accurate.

Unsupported file type

Upstream supports the following file types:

PDF (.pdf)
Microsoft Word (.docx)
Plain text (.txt)

If your file type isn't supported:

Convert your document to one of the supported formats.
Copy and paste the text content directly.
Export from your source application as a PDF or DOCX.

Upload failed

If your upload fails before processing begins:

Check the file size. Very large files may exceed upload limits. Try compressing the file or splitting it into smaller documents.
Check your internet connection. A dropped connection during upload will cause a failure. Make sure you have a stable connection and try again.
Try a different browser. Occasionally, browser extensions or settings can interfere with uploads.
Clear your browser cache. Cached data can sometimes cause issues. Clear your cache and try again.
Try again later. If the problem continues, the service may be experiencing temporary issues.

Still need help?

Email us at hello@getupstream.ai — we read every message.