AI Autofill can fill in your CRM data automatically using either your box timeline or web research. Adding a custom prompt is the single most effective way to improve your results — and is the first thing to try if autofill isn't returning what you expect.
New to AI Autofill? → Learn how AI Autofill works
Wondering about credit usage? → How AI credits work in Streak
What is a custom prompt?
When you configure an AI Autofill column, you can add custom instructions to tell Streak's AI exactly how to fill in that field. Think of it as adding specific guidance on top of Streak's built-in prompt — you don't need to explain what Streak is or what a CRM does. You're simply telling the AI what you want for this specific column.
You can add custom instructions by clicking the dropdown next to a column name from the pipeline view and then selecting AI Autofill.
From here, you'll be able to select whether or not you want the AI to use web research to use any publically available online info to find an answer for you, or use your box timeline (emails, call logs, comments, etc). Then, you're all set to write your custom prompt.
General tips for any prompt
1. Be specific about what you want. Vague prompts return vague results. The more context you give about the desired output, the better.
❌
Fill in the company industry✅
Identify the primary industry this company operates in. Choose from: SaaS, Professional Services, Healthcare, Real Estate, E-commerce, or Other.
2. Use dropdown or tag columns whenever possible. If you configure your column as a dropdown or tag field, the AI will match its output to your predefined options — giving you consistent, filterable data that works with saved views and automations. Avoid free-text fields when a structured output is possible.
3. Include an example output (for text fields). For text columns, providing a sample output helps the AI understand the format you want:
Summarize the needed deliverables from the engineering team now that the project has reached the design stage. Example: "Full architectural drawing, COGS list, risk assessment."
Note: Don't include example outputs for dropdown or tag fields — just define your options clearly.
4. Use "if not certain, leave blank". If accuracy matters more than coverage, add this instruction to your prompt:
If you are not certain, leave this field blank.
This prevents the AI from guessing when it doesn't have enough data.
5. Review the reasoning and refine. Streak shows you the AI's reasoning for every autofill result. Use this to understand why it returned a particular value — then update your prompt to correct it. Prompting is iterative: test, read the reasoning, and adjust.
Use AI to help write the perfect prompt
We trained ChatGPT to help you write good autofill prompts, try it here: Streak Autofill Prompt Builder
Troubleshooting
The AI returned the wrong value → Read the reasoning Streak shows. Update your prompt to be more specific about what you want.
The AI left the field blank → If you included "if not certain, leave blank," this is intentional. Try removing that instruction or checking whether the data source has enough information.
Results are inconsistent across boxes → Add more constraints to your prompt. Define the exact format, options, or conditions you want.