---
updatedAt: 2026-01-20T13:20:01.000Z
agentTools:
projectIndex: https://docs.phacetlabs.com/llms.txt
---
# AI Source
Make your AI outputs verifiable
When using AI to analyze file-based inputs like contracts, reports, or invoices, trust and traceability of the results matter.
AI Source adds transparency to your AI workflows in Phacet by showing exactly which parts of a document were used to generate each response.
It highlights the relevant text snippets directly from your PDFs — so you can verify, debug, and justify every AI output with confidence.
## When to use AI Source
Whether you're extracting structured data, classifying content, or generating summaries, **AI Source makes the model’s reasoning auditable**.
Use it when transparency is important—especially if you’re using AI to extract, summarize, or classify content from complex documents.
AI Source is ideal for:
* **Legal and financial analysis**: Trace exactly which paragraph or clause supports a given classification or risk flag.
* **Contract parsing and review**: Extract key terms and pinpoint where they appear.
* **Invoice and document processing**: Show where invoice amounts, dates, or line items were picked up.
## How it works
AI Source relies on **visual grounding**, a method that links the model’s output to specific locations in a document.
When you activate **AI Source** in a column using the AI Tool, Phacet will:
* Parse the referenced input document (PDF).
* Identify which passages were most relevant to generate the output.
* Visually highlight those source snippets in the viewer.
🛠 AI Source **Works with the following output column types**:
* **Text**
* **Single Select**
* **Multi Select**
* **JSON**
Sources work only with the **AI Tool** (LLM) and require a file-based input. They are not available for Python Tool or manual entry.
## How to set up AI Source
To use AI Source, you first need to configure it at the column level, in the AI Tool parameters
1. Create a **Text**, **Select**, or **JSON** column.
2. Choose **AI Tool** as the generation method.
3. Add your extraction or classification prompt.
4. In your instructions, reference the **file input column** using `@FileColumnName`.
5. Toggle AI Source On
6. Click **Generate Values**.
## Reviewing Sources in a Document
Once AI Source is enabled and your output is generated, you can easily trace back the reasoning behind the result:
1. **Click on the cell** where the AI generation was applied.
2. **Click on\_See Sources\_**
3. A **side panel** will appear, displaying the source document used for analysis.
4. **Click on the numbered chips** to jump to the relevant snippets within the document.
5. The source passages are **highlighted directly** in the document for easy review.
The two screenshots below highlight how to access and review AI Sources
## Credit consumption and optimization
AI Source consumes more credits than a standard AI Tool generation, due to the additional computation required to trace and highlight sources.
**To Control credit usage**
* **Test before scaling**: start with a limited number of rows to validate the relevance of sources and the output quality before applying it to your entire dataset.
* **Write precise prompts**: the more focused your prompt, the more accurate the result — and the fewer unnecessary sources.