REDA One Help Center
EN
Language
English EN Español ES Português PT Italiano coming soon Français FR Deutsch coming soon العربية coming soon
reda.one ↗
Home / REDA AI / Prompt Management / Creating & Editing Prompts

Creating & Editing Prompts

Custom Prompts Fine-Tune Agent Behavior Creating and editing prompts is straightforward. Each prompt has a type, content, and settings that control how it's assembled with other prompts.

How to Create a New Prompt

  1. Open the Agent Definition

    In the REDA AI app, navigate to Agent Management and click the Agent Definition you want to add a prompt to.

  2. Scroll to Related Lists

    Find the "Custom Prompts" related list on the Agent Definition record.

  3. Click New Custom Prompt

    Click the "New" button in the Custom Prompts related list to create a new prompt.

  4. Select Prompt Type

    Choose the Prompt Type from the dropdown: Persona_Override, Custom_Instructions, Schema, Action, or Guardrail. (Required)

  5. Write Prompt Content

    Enter the actual prompt text in the Content field. Be clear and specific about what you want the AI to do. (Required)

  6. Set Sort Order (Optional)

    Enter a number in the Sort Order field (packaged default: 100). This controls the prompt's position in the assembled sequence — across all prompt types, not just its own. Lower numbers come first.

  7. Decide Whether It Starts Active

    New prompts are active by default — the Is Active checkbox comes pre-ticked, and the prompt takes effect on the agent's next conversation as soon as you save. Untick it if you want to create the prompt disabled and enable it later.

  8. Save the Prompt

    Click Save. The prompt is now linked to your agent and will be used if Is Active is checked.

Prompt Configuration Fields

Field Type Required? Notes
Prompt Name Text No Not sent to the AI, but it identifies the prompt in the Runtime Preview's Edit Loadout exclusion list — give each prompt a distinctive name. Example: "Sales Tone Override".
Prompt Type Picklist Yes One of: Persona_Override, Custom_Instructions, Schema, Action, Guardrail.
Content Long Text Area No — but a prompt without it does nothing The actual prompt text sent to the AI. Be clear and specific. Can include context variables like {USER_NAME}.
Is Active Checkbox No Checked by default — a newly created prompt is live immediately. Untick it to create the prompt disabled; you can deactivate and reactivate at any time without deleting.
Sort Order Number Yes Packaged default: 100. Controls the prompt's position across all prompt types. Lower numbers first. Use spaced values (10, 20, 30) for easy reordering.
Channel Multi-select picklist No Which channels the prompt loads on (Internal, External Widget, Community, API).
Context Objects Text No Optional object API names. When set, the prompt loads only in conversations running in the context of one of those objects.
Complexity Estimate Number No Estimated token size used against the agent's Prompt Complexity Budget during assembly. Blank counts as 200.
Summary Text On exclusive prompts One line saying what the prompt covers. It has two readers: an administrator choosing prompts to assign to an agent, and — for an On Demand assignment — the agent itself, deciding whether to fetch the prompt. Write what the procedure is for, not how it works.
Exclusive Prompt Checkbox No Off by default. Tick it and the prompt stops being matched by agent, channel and context altogether; it then reaches an agent only through an Exclusive Prompt assignment.

Editing an Existing Prompt

  1. Open the Agent Definition — Go to the Agent Definition that contains the prompt.
  2. Find the Prompt in Related List — Locate the prompt in the Custom Prompts related list.
  3. Click the Prompt Name — Click the prompt's name to open its record.
  4. Edit Fields — Modify the Content, Prompt Type, Sort Order, or Is Active as needed.
  5. Save Changes — Click Save. Changes take effect immediately for new AI interactions.
  6. Test the Agent — Open the agent chat and verify the changes have the desired effect.

Cloning Prompts Across Agents

If you want to use similar prompts across multiple agents, you can clone a prompt:

  1. Open the Custom Prompt record you want to clone.
  2. Click the Salesforce "Clone" button (or select Clone from the dropdown).
  3. Change the Agent field to the target agent.
  4. Click Save to create the cloned prompt on the target agent.

Alternatively, you can copy the Content text, create a new prompt on the target agent, and paste the content. This gives you a fresh prompt without linked history.

Testing Prompt Changes

After creating or editing a prompt, always test it:

  1. Activate the Prompt (if not already active) — Make sure Is Active is checked.
  2. Open the Agent Chat — Deploy the agent in a sandbox or use the preview feature.
  3. Send Test Messages — Send messages that should trigger the prompt's behavior.
  4. Observe Results — Does the agent follow the prompt's instructions? Are responses formatted correctly?
  5. Iterate — If the behavior isn't what you expected, edit the prompt and test again.
  6. Deploy When Ready — Once satisfied, deploy the agent to production.

Deactivating and Reactivating Prompts

Rather than deleting a prompt, deactivate it:

  • Open the Custom Prompt record
  • Uncheck the Is Active checkbox
  • Click Save

The prompt is now disabled and won't be sent to the AI, but the record remains in your database. You can reactivate it anytime by checking the box again. This is useful for:

  • A/B Testing — Test two versions of a prompt by deactivating one, testing the other, then switching.
  • Seasonal Behavior — Disable prompts outside their relevant season and reactivate when needed.
  • Experimental Changes — Create a new prompt, disable the old one, test, then commit or revert.

Best Practices for Creating Prompts

Keep Each Prompt Focused

Each prompt should have ONE clear purpose. Don't combine multiple behaviors into a single prompt. It's better to have 3 focused prompts than 1 mega-prompt.

Be Specific, Not Vague — Instead of "Be helpful", say "If the user asks about pricing, respond with a link to the pricing page and offer to schedule a demo."

Name Your Prompts Clearly — Use descriptive names like "Response Format - Bullet Points" or "Guardrail - No Personal Data" so admins can find and manage them easily.

Use Sort Order Strategically — If certain instructions should take priority, assign them lower Sort Order numbers so they're assembled first.

Avoid Redundancy — Don't repeat the same instruction across multiple prompts. Consolidate into one prompt and reference it in comments if needed.

Include Context Variables — Use {USER_FIRST_NAME}, {TODAY_DATE}, etc. for personalization. This makes interactions feel more tailored.

Common Prompt Types & Examples

Prompt Type Use Case Example Content Snippet
Schema Describe Salesforce objects and fields "Opportunity.StageName has values: Prospecting, Qualification, Decision Makers, Closed Won, Closed Lost"
Custom Instructions Behavioral rules "Always format lists with bullet points. Keep responses under 150 words. Use friendly tone."
Action How to perform specific actions "When creating a task, set the priority to 'High' for urgent items. Use the subject format: 'Follow-up: [customer name]'"
Guardrail Restrictions and safety rules "Never discuss employee salary information. If asked, respond: 'That's a question for HR. I can help with other topics.'"

Troubleshooting Prompt Issues

Prompt Not Taking Effect

Verify that Is Active is checked. If active, check that the correct Agent is set on the prompt. Clear your browser cache and refresh the chat widget.

Agent Ignoring Prompt — The AI may ignore instructions if they conflict with the persona or other prompts. Use Guardrail prompts for hard rules. Avoid vague language — be explicit about what you want.

Too Many Tokens Used — Long prompts consume tokens. Consider splitting large prompts into multiple focused ones, and use Schema prompts instead of querying tools.

Configure reusable business instructions

Exclusive Prompts let administrators assign reusable instructions to selected agents. Review the Exclusive Prompt flag and Summary on Custom Prompt, then the assignment's Agent Definition, Custom Prompt, Active and Load Mode. Use the intended load mode; On Demand needs the context-prompt tool in the agent's loadout. Record-type brackets use shared scope rules, including in preview and prompt administration. Prompt and tool ordering is deterministic, and runtime preview can use an actual record ID. Updated language instructions follow the message on API/external channels; user placeholders apply only where the authenticated user context is intended. Preserve custom prompts when merging the current defaults.

Frequently Asked Questions

Yes. Changes to prompts take effect immediately. New conversations will use the updated prompt. Existing conversations continue with the version they started with (unless the session resets). Test changes carefully to avoid disrupting live agents.

The Persona field on Agent Definition is the base system prompt. Persona_Override Custom Prompts refine or override it. Use the Persona field for core identity, and Persona_Override prompts for adjustments.

Yes, the AI understands markdown. You can use *, -, # for lists and formatting in prompts. The AI will interpret and follow markdown conventions.

Each prompt consumes tokens. Keep individual prompts under 10,000 characters. If your prompts are getting very long, break them into multiple focused prompts and use Sort Order to control assembly order.

Next Steps

Learn prompt best practices to optimize your prompts, or explore context variables for personalization.

← Previous: Understanding Prompts
Next: Prompt Best Practices →