Indexing files for AI
Making a document searchable by the assistant, and why it is a separate step.
Uploading a file stores it. Indexing it reads its text and makes it searchable by meaning, so the AI assistant can find the relevant passage when you ask a question.
The two are deliberately separate. You decide which documents the AI is allowed to search.
Indexing an attachment
In the attachments table on the Files page, files that can be read and are not yet indexed show an Index for AI action. Choosing it indexes that file.
Files in the organisation library are indexed as part of uploading them.
Indexing does not spend AI credits.
What the AI can then do
- Search across indexed documents by meaning, not just keyword, and cite what it found.
- Read a specific file's extracted text, in pages, when it needs the detail.
Text is extracted from PDF, Excel, Word and PowerPoint. A scanned PDF with no text layer has nothing to extract, and will not become searchable.
Choosing what to index
Worth indexing:
- previous sustainability reports,
- policies, codes of conduct, supplier requirements,
- data summaries and calculation sheets behind reported figures.
Think twice about:
- documents containing personal data that does not need to be in a reporting context,
- commercially sensitive material that is not needed to answer a disclosure,
- very large raw data dumps, where a summary sheet is usually more useful than a million rows.
