Skip to main content

Chat with documents

With Chat with files, you let AI-Corporate answer based on documents that you link to the chat yourself. This is suitable for a one-time analysis, a comparison, or a question about information that is not publicly available.

When to use Chat with files?​

Use this feature, for example, to:

  • summarize a report;
  • extract agreements, risks, or action points from a document;
  • compare multiple versions of an agreement;
  • ask questions about policy, research, or internal information;
  • analyze information from tables, forms, or presentations.

Documents that an administrator has published centrally do not need to be selected by you. For a relevant ordinary chat question, AI-Corporate can automatically search the central knowledge documents available for your organization or location. You can also find these sources under Knowledge Base → Organization → Documents (or under your own location). Administrators can read in Setting up and managing the Knowledge Base how to add and publish this knowledge.

Selecting documents​

  1. Open a new or existing chat.
  2. Click the paperclip next to the input field.
  3. Upload a new file or choose a previously uploaded file from File Management.
  4. Select up to ten files for Chat with files.
  5. Ask a concrete question and optionally specify the desired form of the answer.

The selected files remain active as long as the file mode is enabled. Before asking another question, check which files are still selected so that old sources do not unintentionally influence the answer.

You can also paste an image, such as a screenshot or a copied picture, straight into the input field with Ctrl+V (Cmd+V on a Mac); the image is uploaded right away and attached to your question, just like when you drag a file onto the input field. To put a screenshot straight on the clipboard, press Cmd+Ctrl+Shift+4 on a Mac (without Ctrl, the screenshot is saved as a file) or Windows key+Shift+S on Windows. If you copy text that also contains an image, such as cells from Excel or text from Word, the text is pasted as usual. In an assistant, pasting images only works if that assistant allows adding images.

Suitable language model

AI-Corporate checks whether the chosen language model supports document analysis. If not, a suitable model from the current model catalog can be automatically selected.

How document analysis works​

After uploading, AI-Corporate processes the file. The content is indexed into smaller fragments. For each question, the application first searches for the fragments that best match the content of the question. The language model receives the question together with that relevant context.

As a result, a targeted question usually works better than a very general instruction. For example, ask:

What three obligations apply to the supplier according to the contract? Name the relevant passage for each obligation.

Processing can take longer for large files, scans, images, and files with many pages. Wait until the file is shown as processed in File Management before querying it.

Text, tables, and visual information​

PDF, Word, and PowerPoint files are automatically processed into searchable content. This includes text, document structure, and where possible, relevant visual information.

This is especially useful for:

  • scanned documents without a good text layer;
  • tables where rows and columns are lost with ordinary text extraction;
  • presentations where text and images together convey meaning;
  • forms with checkboxes, circled choices, or handwritten notes;
  • charts and diagrams where the conclusion is not literally stated in the text.

Explicitly mention in your question that tables, charts, or images should also be included. Always verify important conclusions against the original file.

Supported file types​

AI-Corporate supports, among others:

  • PDF (.pdf);
  • Word (.docx);
  • PowerPoint (.pptx);
  • Markdown (.md);
  • text (.txt);
  • CSV (.csv);
  • JSON (.json);
  • common audio and video files, including .mp3, .mp4, .m4a, .wav, and .webm.

The exact processing differs per file type. A damaged, protected, or unusually structured file may not always be fully readable.

Comparing multiple documents​

Select the relevant documents at once and clearly formulate what you want to compare. Provide the desired criteria and output format.

Compare these three quotes on price, lead time, exclusions, and termination conditions. Create a table and then name the main risks.

If documents contradict each other, ask to name the differences per source. This prevents conflicting information from being merged into one seemingly unambiguous answer.

Process per file​

With Process per file, the same prompt is applied separately to each selected document. You then receive an individual result per file. This is useful for a series of contracts, applications, reports, or forms that need to be assessed on the same criteria.

A practical workflow:

  1. Optionally add a reference document under Chat with files.
  2. Select the documents to analyze under Process per file.
  3. Write one prompt with fixed evaluation criteria.
  4. Check the results per file and then compare them.
Maximum

For Process per file, a maximum of thirty files per task applies.

Audio and video​

Audio and video files are first transcribed via the configured transcription provider. If the chosen model supports this, the transcription can include time blocks and speaker labels. Then the transcription is made searchable in the same way as text from a document.

Check names, technical terms, numbers, and speaker labels: automatic transcription can misrecognize these. Long files may also hit provider-dependent limits for file size or duration.

Asking better questions​

For document questions, preferably use this structure:

  1. Task — what should AI-Corporate do?
  2. Scope — which documents, chapters, or periods are relevant?
  3. Criteria — on what should the answer be based?
  4. Output — do you want a table, summary, checklist, or advice?
  5. Traceability — ask for source references or relevant passages.

Example:

Summarize chapters 2 and 3 for management. Name decisions, outstanding risks, and owners in a table. Indicate for each point from which document and chapter it comes.

If an answer lacks information​

  • Check whether the correct file is selected and fully processed.
  • Ask a more specific question with recognizable terms from the document.
  • Explicitly ask to include tables, images, or attachments.
  • Check whether a scan is readable and not password protected.
  • Upload a corrected or smaller version if processing failed.
  • Compare the answer with the original source for important decisions.

Do you want to create a new document or fill in a Word template from the chat? Then read Creating documents.

Checking source usage​

The amount of source text used depends on the question, the document, and the chosen model. Check with the answer which sources were fully, partially, or not read. Read the explanation about file management and source usage.

On this page