As of 13 August 2026, AI can only partly update spreadsheets from online forms.

What to actually do

1. Hand it to a person

   The route this page recommends

   A person who owns the outcome does this end to end, worth it when the failure is dear.

2. Use a tool built for this

   Second choice

3. Do it yourself

   The distant third

   A chat interface, power-user skill, and roughly 1 hour until you can act on the result.

   How to actually do it

   1. Open the online form and spreadsheet, then copy the exact form field names, spreadsheet column headings, platform names and intended matching key into the prompt.
   2. Paste the completed prompt into an AI assistant and ask it to produce the workflow design, code or native automation configuration without inventing missing platform details.
   3. Create a test copy of the form and spreadsheet, and apply the proposed field mappings and permissions there rather than to the live records.
   4. Use dummy submissions to test a normal response, blank fields, duplicate submissions, an amended response and invalid data, then compare each resulting row with the expected outcome.
   5. Inspect the run history, error log and access settings for failed runs, repeated runs and unauthorised access, and correct the configuration against the platform documentation.
   6. Ask a colleague who owns the spreadsheet or process to check the mappings, duplicate behaviour, permissions, recovery steps and sample outputs.
   7. Deploy the tested workflow to the live form and spreadsheet, send one controlled real submission, and confirm that the expected row was created or updated before relying on it.

   Prompt

   Design a reliable workflow that updates a spreadsheet from an online form.
   
   Context:
   - Form platform: [Google Forms, Microsoft Forms, Typeform or other]
   - Spreadsheet platform: [Google Sheets, Microsoft Excel or other]
   - Form fields: [list each field and its exact name]
   - Spreadsheet columns: [list each column and its exact name]
   - Desired behaviour for a new submission: [add a new row or update an existing row]
   - Matching key, if updating existing rows: [field name]
   - Rules for blank, duplicate, amended or invalid submissions: [rules]
   - People who need access: [roles, not names]
   - Data that must not be exposed: [data types]
   
   Produce a practical implementation plan for this exact setup. Prefer the platform's native automation where it is sufficient; otherwise provide complete code or configuration instructions. Do not invent platform features, field names, permissions or API details. Mark anything that must be confirmed in the platform documentation.
   
   Include:
   1. The exact trigger, actions, field mappings and error handling.
   2. Authentication and minimum required permissions.
   3. Duplicate prevention, retries, logging and handling of partial failures.
   4. A test plan using dummy submissions, including normal, blank, duplicate, amended and invalid data.
   5. A rollback and monitoring plan.
   6. A final checklist of what I must configure and what a colleague must verify before this is used with real submissions.
   
   Do not use real personal data in examples. Keep the design limited to the information supplied above.

   Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.