Storm's AI Templates Generate feature produces more accurate, ready-to-deploy templates when given clear, specific prompts. The quality of your output depends largely on the quality of your input. This article covers how to write better prompts, how to use preferences effectively, and what the Generate feature currently cannot do.

1. Be specific about the document type and context

Always name the document type and include industry or project context in your prompt. Common construction industry abbreviations — SWMS, ITP, JSA, PTW — are reliably identified, so you can use short names confidently without spelling them out in full. The more context you add (task type, location, crew type), the more tailored your output will be.

📓 Note: Storm identifies SWMS, ITP, JSA, and PTW as specific document types — not vague terms. Using a short name like "Create an ITP for structural steel" is sufficient to get accurate, industry-specific template output.

2. Select the right generator type

Storm has 23 specialist generator types covering safety, quality, and environmental workflows. Selecting the correct generator — rather than using a generic prompt — produces templates structured to industry best practice for that document type.

For example, selecting the SWMS generator and describing your task produces a better result than typing "create a SWMS" into a generic prompt field.

See the full list in What Can You Generate With AI Templates?

3. Select the correct country or region

Storm uses your region selection to apply local terminology, regulatory references, and document structure. Getting this right avoids needing to manually rename fields after generation.

4. Describe required sections and field types explicitly

Storm can generate tables, dropdowns, counter fields, approval sign-offs, and linked lists — but only if you indicate they're needed. If your prompt doesn't mention them, Storm may omit them.

Useful things to include in your prompt:

📓 Note: If your prompt contains a form or reference number pattern — such as "Form No.", "Permit No.", or a numbered prefix like "NCR-001" — Storm automatically detects this and generates a read-only Counter field with the correct prefix. You do not need to request a Counter field explicitly when the pattern is clear in your prompt.

5. Use preferences to guide structure

Before clicking Build, configure your preferences to match the output you want:

6. Answer clarifying questions when prompted
​

When Storm hits something ambiguous in your prompt or preferences, it will ask you a quick question before continuing — rather than guessing and producing the wrong output. Answer each question to help Storm generate a more accurate result. You can set your preferences upfront to reduce the number of clarifying questions.

7. Use voice note input for more natural, detailed prompts
​
Both the Generate and File Import flows support voice note recording. This can be especially useful when you need to describe a complex structure quickly, or when typing out a detailed prompt would be slow or awkward. The same principles apply to spoken input as typed — be specific, name the document type, and include industry or project context. Storm combines voice and text, so you can speak part of your instructions and type the rest in the same session.

8. Review and refine before deploying

Generated templates are a starting point, not a finished product. Always review before making a template live — particularly for:

Will Storm always produce a perfect template on the first try?

Not always. Complex workflows, unusual document structures, or vague prompts may require manual refinement. The more specific your prompt and preferences, the closer the output will be to ready-to-deploy.

Can I change preferences after generation starts?

No. Preferences are applied at the start of generation. If you want different settings, start a new generation request.

Does Storm generate Counter fields automatically?
​Yes. When your prompt contains a form or reference number pattern — such as "Form No.", "Permit No.", or a numbered prefix like "PTW-001" or "TOOLBOX-001" — Storm automatically generates a read-only Counter field with the correct prefix. Records are then auto-numbered in sequence (e.g. PTW-1, PTW-2 or TOOLBOX-1, TOOLBOX-2). You do not need to request a Counter field explicitly when the pattern is clear in your prompt.

Can I use abbreviations like SWMS, ITP, JSA, or PTW in my prompt?
​Yes. Storm reliably recognises common construction industry short names and produces accurate, industry-specific template output when you use them. You do not need to spell out "Safe Work Method Statement", "Inspection and Test Plan", or similar in full. A prompt like "Create a SWMS for excavation works" or "Build an ITP for structural steel" is sufficient to get the right template structure.

Why did Storm use the wrong generator structure?

If your prompt doesn't closely match the selected generator type, Storm retries automatically — first trying the generator that best fits your prompt. If that doesn't work, Storm builds a template using its own built-in knowledge. You will always land in the editor with a template. For the most accurate structure, always match your prompt to the generator type you selected.

Can I queue multiple generation requests?

Yes. As soon as a generation request begins processing, you can start a new one immediately.

Do generated templates require review before use?

Yes. Safety workflows, compliance processes, and approval-heavy forms should always be reviewed and refined before operational deployment.

What happens if Storm refuses my prompt?
Storm screens your prompt before generation begins. If a prompt is clearly off-topic, unintelligible, or violates content policy, Storm will refuse it — a plain-language banner appears explaining why the prompt was not accepted, and no template is created.

The following prompt types are not blocked by screening:

If your prompt is refused, read the plain-language banner for guidance on how to rephrase. This banner is distinct from the generic “Unable to create template” error, which indicates a technical failure rather than a content-screening outcome.