Skip to content
  • There are no suggestions because the search field is empty.

Document AI-Copilot: Query Examples

Learn how to write effective questions and get the most out of the AI-Copilot.

This article complements the Document AI-Copilot Best Practices guide with concrete examples of queries that work — and ones that don't.

Query types

There are four kinds of queries the Copilot may receive. Understanding them helps you phrase questions that consistently return useful answers.

Type

Description

Invalid

Ambiguous, lacks context, or requests metadata the Copilot cannot access

Needs refinement

Has clear intent but not enough detail to retrieve a reliable answer

Simple

Well-defined, context-appropriate, asks for one specific piece of information

Complex

Well-defined but spans multiple contexts or combines several questions into one
Invalid queries

These queries will not return useful answers. The most common cases are:

Not asking anything:

  • hello
  • 157

Requesting metadata (the Copilot does not have access to document listings or file-level information):

  • show me this document
  • show me the latest…
  • is there any Question & Answers document?

When asked to show a document directly, the Copilot will respond:

I'm unable to provide the document directly. However, I can summarize the key points or answer specific questions you may have regarding its content.

Queries that need refinement

These queries show intent but lack the context needed for a reliable answer. Single words or short fragments fall into this category:

  • chocolate
  • cocoa powder
  • fibre claim
  • what about cereals?

Fix: Add context and write in full sentences. Before submitting, ask yourself:

  • What specifically would I like to know about this?
  • How would I explain this search to a colleague?

Example — query: chocolate

Example — improved query: What are the permissible levels of contaminants in chocolate products?

The Copilot's answer to the improved query would return specific regulatory limits for lead, arsenic, cadmium, mercury, copper, zinc, aflatoxin B1, and microbiological indicators.

Simple queries

Simple queries are well-formed and ask for one specific piece of information. They work best when asked in the correct context (e.g. inside a regulatory post, an FRG, or an insights document).

Examples:

  • What is mentioned about BPA and/or PFAs? (in a post)
  • What is the definition of vegan food? (in a post)
  • Are there any restrictions on low sugar claims? (in an FRG)
  • Is it allowed to use E477 in food? (in an FRG)
  • What minimum cocoa solids content should be in milk chocolate? (in an FRG)
  • What is a health benefit of Nutri-Score labeling? (in insights)

ℹ️ The same well-written question can fail if asked in the wrong context. Make sure you are inside the relevant document or section before querying.

Complex queries

Complex queries span multiple contexts or combine several sub-questions into one. The Copilot may struggle to answer these in a single response.

Examples:

  • Is the Swiss regulation on GMO aligned with EU regulations on GMO?
  • What are the regulations on PFAs or BPA in Latin America countries?
  • If the best-before date is not in the same field of vision as the net weight and legal designation, can a "see back of pack" be put in place?

Fix: Decompose complex queries into simpler ones, each asking for a single piece of information in the right context.

Example — complex query: Regulations on PFAs or BPA in Latin America countries

Example — decomposed queries:

  1. What PFAS regulations apply in Brazil? (in a relevant post)
  2. What BPA restrictions exist in Mexico? (in a relevant post)
  3. What is Peru's current position on BPA in infant products? (in a relevant post)

The Copilot can then answer each accurately, drawing from the specific documents where that information appears.