Prompt variables are placeholders that insert row values into an AI prompt.

They let one prompt run across a whole spreadsheet.

For example:

Summarize this company in one sentence.

Company: {{Company Name}}
Website: {{Website}}
Description: {{Description}}

For each row, the variables are replaced with that row's values.

Why prompt variables matter

Without variables, you copy and paste data into a chat window. With variables, the prompt becomes a reusable workflow.

Prompt variables are useful for:

  • Running ChatGPT on CSV rows
  • Classifying text
  • Scoring leads
  • Translating product descriptions
  • Extracting data from scraped text
  • Researching one company per row
  • Creating personalized email lines

They keep the prompt stable while the input changes.

Common prompt variable patterns

Use one variable when the task is simple:

Rewrite this product description for clarity:
{{Description}}

Use several variables when the answer depends on context:

Score this lead from 1 to 5 for a B2B SaaS outbound campaign.

Company: {{Company Name}}
Website: {{Domain}}
Industry: {{Industry}}
Employee count: {{Employees}}
Recent job post: {{Job Description}}

Use variables with instructions for missing values:

If {{Company Website}} is empty, return "Missing website" and do not guess.

How to check that variables work

A variable must refer to an input column. Typing a company name directly into the prompt uses that same company for every row. In Datablist, use the property picker after typing / or {{ and select the column you want to insert.

Keep instructions above the row data so long descriptions cannot become part of a sentence that tells the model what to do. For example:

Summarize a company description

Summarize the supplied company description in one sentence. Use only the supplied text. If the description is empty or does not describe a company, return “No description”.

Description: /Description

Test a row with a short description, a row with a long description, and a row with an empty description. Check that the results change with the input and that missing data does not produce a guessed answer. The Datablist LLM enrichment reports an empty-variable error when none of the referenced inputs has a value; a missing-data instruction does not turn an entirely empty input into a model call.

Prompt variables in an LLM enrichment supply data for each row. In AI Editing, property references help generate a script that runs across your data. They do not mean the model reads and reasons about every row individually. Use AI classification or another row-based enrichment when each text needs its own model response.

Good variable names

Readable column names make prompts easier to maintain.

Prefer:

  • Company Name
  • Website
  • Job Title
  • Product Description
  • Review Text

Avoid unclear names such as col1, data, or misc.

📌 Keep prompts debuggable

If a result looks wrong, you should be able to read the prompt and understand which input field caused it.

Prompt variables in Datablist

Datablist uses variables such as {{Column Name}} in AI enrichments. You can map row data into prompts for Ask ChatGPT/OpenAI, Ask Claude AI, Ask Gemini, and other LLM providers.

Prompt variables also work in AI agent workflows when each row needs web research. For example, the agent can search for {{Company Name}} pricing page or read the website stored in {{Domain}}.

For more context, read LLM spreadsheet processing, batch LLM processing, and the guide on running ChatGPT on Excel and CSV rows.