We recently released AI Workflow Builder for Starter, Pro and Enterprise Cloud customers, turning your natural language prompts into working automations.
Instead of launching into building a new workflow from scratch with a blank canvas, you can easily get your automation ideas out of your head and into a functioning workflow in minutes. You can even use it to debug and refine your existing workflows, all with real-time feedback of what AI Workflow Builder is doing at every stage.
Think of AI Workflow Builder as a thought partner, there to be used as a sounding board for validating ideas, a way of exploring nodes that you might not have considered, and an n8n expert ready to help fix or improve your workflows.
Building workflows with prompts is a natural and intuitive process, but getting the most of AI Workflow Builder can take time and a few iterations to find what works best for you.
Getting the most from AI Workflow Builder is a personal experience. Here are proven ways to improve your AI Workflow Builder prompts, based on feedback from the n8n community and our AI engineering team.
Best practices & examples
Think in iterations
Having a full plan of how you want the workflow to be triggered, what data and logic needs to be processed, and how it end is fantastic. However, giving AI Workflow Builder all of this in a single huge prompt might not get the best outcomes.
Think in iterations, especially for large, complex automations. Use AI Workflow builder to create a rough “map” of the finished workflow, then run, prompt, refine, and repeat.
Prepare for required parameters and credentials
When sending the first prompt to kick-start the workflow building process, almost all workflows will need you to manually enter additional parameters and credentials, especially if using APIs and tools that need access to data via a specific account.
When AI Workflow Builder finishes after your prompt, it will detail all manual steps you need to carry out to make sure this iteration works correctly.
As you’re planning your overall workflow, think about the tools and services you plan on connecting to and plan for any login information or required access in advance to cut down on delays in your workflow building.
Don’t rely on prompts created with other AI chat
It might be tempting to use another AI tool for helping build the perfect prompt, but in reality this can lead to overly-long prompts that might still face the same challenges as a much shorter prompt that lacks good planning for triggers, outputs and data flow.
Creating a prompt using another chat tool isn’t a bad thing, but don’t rely on it to create a single, comprehensive prompt that creates the perfect workflow in one go. Treat it in the same iterative way of prompting, testing, and refining, with a clear vision of what you are wanting the workflow to achieve.
Be specific about integrations and nodes
Don’t just say “get my emails” or “send results to a table”, get ready to be specific with exactly what you want AI Workflow Builder to make for you: vague prompts = vague workflows that will need more iterations and potential manual work to get them working.
Instead, use prompt language like “get the latest 10 emails from my Gmail account” and “send to a Google Sheet called XXX” which will help AI Workflow Builder create exactly what you were meaning, reduce build iterations, and return back more helpful responses for required next steps.
Clearly describe the flow of data
Just like being specific about integrations, try to be as specific as possible about the way you expect data to flow in your workflow.
In your prompts, tell AI Workflow Builder where it should find data, what it should do with it and where it should send it.
Rather than “Get the latest sales calls, summarize them, then save the summary”, use more specific prompting like “Get the last 100 sales call text summaries from Gong, then send them to OpenAI to summarize each one in 2-3 bullet points. Save the summaries in an n8n Data Table with the date of the call, company name and AI summary”.
Be clear about how the data should be formatted, and which fields should be sent from one node to another.
No need to role play
Unlike some general purpose AI chat tools that require you to explain to the agent how it should operate to get the best results, there’s no need to do that with AI Workflow Builder.
Don’t worry about telling it “You are an expert n8n workflow builder” - it already is, and will create the best possible workflows based on your prompts.
A deeper look into why a good prompt works well
Create an automation that checks the weather for my location every morning at 5 a.m using OpenWeather. Send me a short weather report by email using Gmail. Use OpenAI to write a short, fun formatted email body by adding personality when describing the weather and how the day might feel. Include all details relevant to decide on my plans and clothes for the day.
It calls out specific integrations
Here, the prompt is calling out specific tools like Gmail and OpenAI rather then “email” and “AI model”, helping it to create the right nodes for the task.
It clearly describes the flow from start to finish
Nothing is left for AI Workflow Builder to work out, the prompt clearly states what should happen at each stage of the automation.
No-fuss, straight to the point instructions
We don’t need to include additional AI-related instructions or “scene setting” to help with context, here the prompt is instructing AI Workflow Builder what to do with comprehensive instructions.
A small iteration ready for the next step
The prompt is <400 characters rather than a 1,000+ character input, giving you the ability to quickly validate ideas and make small changes as your workflow idea evolves.
Key takeaways
With AI Workflow Builder you really do get out what you put in: clear, well-defined, and specific prompts will create well-built workflows with clear flows from trigger to output. External tools to help you write your prompts might just add confusion and extra build time.
Have a idea in mind of what you want the workflow to achieve and which nodes and integrations you might want, helping you plan for other requirements like parameters, credentials and API endpoints.
Lastly, don’t be afraid to build in iterations: use AI Workflow Builder to create workflows in clear, short prompts rather than trying to achieve everything in one go. This allows you to quickly validate ideas and quickly change direction if needed. Being specific about your nodes, integrations and flow of data will help reduce the number of iterations needed to get to your perfect workflow.
Got any tips on how you're building with AI Workflow Builder? Let other community members know your best tips, tricks and prompts.
Now, check out the docs and get building with AI Workflow Builder!
