Write effective AI prompts in Baserow

Great AI results start with clear prompts. Whether generating formulas, creating content in AI fields, or analyzing documents, specific instructions produce better outputs. This guide teaches you to write prompts that get accurate, useful results from Baserow’s AI features.

This guide covers prompt writing for:

  • AI prompt field - Generating text, classifications, and document analysis
  • AI formula generator - Creating formulas from descriptions
  • General AI interactions - Chatbots and assistants

Overview

Clear, specific prompts get better AI results. Include what you want, reference your data with {Field Name}, and specify output format. Test on a few rows before generating for your whole table.

AI models respond to the instructions you provide, your “prompt.” The quality of AI-generated content depends heavily on how well you communicate your intent. A vague prompt produces vague results; a detailed, structured prompt produces precisely what you need.

Prompt patterns

Every good prompt needs:

  1. Task - What you want done
  2. Input - Data to process (Field Name)
  3. Format - How to structure output
  4. Constraints - Rules and boundaries

Example:

Classify this review as Positive, Neutral, or Negative. Review: {Customer Review} Return only one word.

Classification

Classify {Input Field} as [Option1], [Option2], or [Option3]. Return only one of these options.

This prompt can be used for support ticket routing, sentiment analysis, and lead scoring.

Summarization

Summarize {Input Field} in [X] sentences focusing on [key aspects].

This prompt works effectively for summarizing insights from meeting notes, customer feedback, or long documents because it guides the model to focus on key information while maintaining context.

Extraction

Extract from {Input Field}: Item 1, Item 2, Item 3. If any field is missing, return "Not found"

Use for invoices, resumes, and forms. This prompt works because it focuses the AI on pulling structured, relevant data from semi-structured documents.

Generation

Write a [content type] for [audience]. Input: {Field Name}, Tone: [style]. Length: [constraint]

This generation prompt is ideal for creating product descriptions, email responses, and social media posts because it guides the model to produce clear, contextually relevant content.

Tips by feature

AI is collaborative, not telepathic. The more context you provide, the better your results.

For AI fields

✅ Do:

  • Reference fields: {Customer Name}, {Order Total}
  • Set appropriate temperature (0.1 for consistent, 0.8 for creative)
  • Use “Choices” output type for classifications
  • Test on 5-10 rows before bulk generation

❌ Don’t:

  • Use generic terms like “the name column”
  • Skip edge case handling (empty fields, unusual values)
  • Generate for 1000 rows without testing first

For formula generation

✅ Do:

  • Describe what you want: “Calculate 20% discount if quantity over 100”
  • Use exact field names: “Multiply Hours Worked by Hourly Rate”
  • Specify output format: “Round to 2 decimals”
  • Handle edge cases: “If quantity is 0 or blank, show 0”

❌ Don’t:

  • Describe syntax: “Use an IF function to check…”
  • Use generic terms: “the hours column”
  • Cram multiple calculations into one prompt

For document analysis

✅ Do:

  • List what to extract: “Invoice number, date, total, vendor name”
  • Specify format: “Return as bullet list”
  • Handle missing data: “If field not found, write ‘N/A’”

❌ Don’t:

  • Use vague requests: “Get the important information”
  • Assume AI knows document type
  • Skip format specification

Temperature guide

Setting Best For Examples
0 - 0.3 Consistent, factual outputs Classification, data extraction, summaries
0.4 - 0.6 Balanced General business writing, Q&A
0.7 - 1.0 Creative, varied Marketing copy, product descriptions

Common mistakes

Too vague

❌ “Analyze this data”
✅ “Identify the top 3 sales trends in the last quarter”

Missing context

❌ “Classify this: {Text}”
✅ “Classify this support ticket as Bug, Feature, or Question: {Ticket Description}”

No error handling

❌ “Extract invoice total from {PDF}”
✅ “Extract invoice total from {PDF}. If missing or unreadable, return ‘Unable to process’”

Overly complex

❌ “Analyze sentiment, extract topics, classify category, and generate response”
✅ Create separate AI fields for each task

Ready-to-use templates

Support ticket classifier

Classify this support ticket into exactly one category:

Technical Issue Billing Question Feature Request Account Help General Inquiry

Ticket: {Ticket Description} Return only the category name.

Lead scorer

Score this lead as Hot, Warm, or Cold. Hot = Budget + urgent timeline + decision maker Warm = Interest + vague timeline + needs approval Cold = Browsing + no timeline + unclear fit Lead notes: {Lead Notes} Return only: Hot, Warm, or Cold

Invoice extractor

Extract from this invoice:

Invoice #: [number] Date: [YYYY-MM-DD] Vendor: [name] Total: [amount with currency]

Invoice: {Invoice PDF} If any field not found, write “Not available”

Product description generator

Write a 75-word product description for online shoppers. Product: {Product Name} Features: {Features} Target buyer: {Target Customer} Tone: Persuasive but informative Include: Key benefit and unique differentiator

Email sentiment analyzer

Analyze this customer email: Sentiment: Positive, Neutral, or Negative Urgency: High, Medium, or Low Topic: (5 words max) Email: {Email Content} Format as: Sentiment: [X] Urgency: [X] Topic: [X]

Troubleshooting

“Why are my outputs inconsistent?” Lower the temperature to 0.1–0.3 and add explicit constraints to improve consistency.

“Why does the AI ignore parts of my prompt?” Simplify the prompt and place the most important instructions first.

“Why is the AI making up information?” Use only specific fields: “Only use information from Field. If not available, return ‘Not found’.”

“Why is the output in the wrong format?” Provide a clear example: “Format as: Label: Value.”

“Why does it work for some rows but fail for others?” Handle empty fields explicitly: “If Field is blank, return ‘N/A’.”

Need help? Visit the Baserow community or contact support.