Sign in Subscribe
Knowledge Base

How to Name Your Tableau Data Source to Prevent AI Hallucinations (Tableau Pulse & Copilot)

Yehwan Hwang

• 2 min read
Share
How to Name Your Tableau Data Source to Prevent AI Hallucinations (Tableau Pulse & Copilot)

Many administrators who have introduced AI features like Tableau Pulse or Einstein Copilot into their Tableau environments are struggling with "hallucinations"—situations where the AI fetches the wrong data to answer user questions.

The biggest cause of these AI hallucinations lies in the messy 'Data Pane' hidden behind your beautiful dashboard screens. AI doesn't read the visual charts; it reads the text (field names) in the Data Pane on the left to analyze your data. For a successful implementation of Pulse & Copilot and solid data governance, here is a naming guide you should apply to your data sources.

💡
Scope of Application: This guide should be applied uniformly to all fields present in the Data Pane. This includes custom calculated fields you create, as well as the names (aliases) of existing dimensions and measures pulled directly from your database.

1. Avoid Abbreviations and Internal Jargon (Use Full Names)

Leaving column names like TXN_AMT or CUST_ID exactly as they came from the database for developer convenience will severely confuse Large Language Models (LLMs). To help the AI accurately map the inherent meaning of each field, you must rename your data source aliases using clear business language (full names) rather than abbreviations or machine language.

  • Recommended (Full Name): Total_Sales, Customer_ID
  • Not Recommended (Abbreviation/Machine Language): tmp_sales, TXN_AMT

2. Explicitly Express Field Attributes

Help the AI instantly grasp the nature of the data by explicitly stating at the beginning of the name whether it is a date or a logical (boolean) value indicating true/false.

  • Date Fields: Use the Date_ prefix. (e.g., Date_Order, Date_First_Purchase)
  • Boolean Fields: Use the Is_ or Has_ prefix. Instead of simply naming a field Active, naming it Is_Active_User clearly communicates to the AI that this is a true/false criterion determining whether a user is active.

3. Remove Unnecessary Version Tags and Clutter (Hide Unused Fields)

During data processing, it's common to append version numbers or temporary statuses to field names (e.g., calc_rev_01, temp_sales_backup). If these tags remain, the AI will get confused about which of the many dimensions and measures it should prioritize for its analysis.

  • Recommended: Total_Revenue
  • Not Recommended: Revenue_Data_Final_v3

Unused intermediate calculations or unnecessary columns from the original DB should be boldly Hidden. Leave only the cleanest, final fields visible for the AI to utilize in its analysis.

4. Leverage 'Default Properties ➔ Comment' as a Metadata Goldmine

💡
If it's difficult to capture the entire business context within just the field name, actively utilize Tableau's [Default Properties] ➔ [Comment] feature.

Entering a specific data dictionary definition here, such as "This field represents net sales excluding VAT," becomes the most powerful Knowledge Base. It enables the AI to interpret the data correctly and provide highly accurate answers to your end-users.

A systematic data source naming convention is more than just cleaning up text. In the upcoming era of AI agents, it is the crucial first step to ensuring data integrity and building an analytics environment your organization can truly trust. Start inspecting the field names of your 'Published Data Sources' shared on the server today.