JSON Output Mode

JSON Output Mode lets one Use AI column return a structured object instead of plain text.

Use it when one prompt should generate multiple related fields together, for example:

• subject + body
• score + reasoning
• headline + summary + cta

This is the fastest way to avoid dependency chains like "draft email -> subject extractor -> body cleanup" when the fields belong together.

When to Use It

Use JSON mode when:

• The outputs come from the same reasoning step
• You want those outputs to stay in sync
• You want to cut credits by replacing multiple AI columns with one
• You can reference nested fields downstream, for example {{Outreach Draft}}.subject

For simple one-field outputs, standard text mode is still fine.

How to Enable It

1. Open your workflow table
2. Click Add Column -> Use AI
3. Write your prompt as usual
4. In Output Type, choose JSON - Structured JSON output
5. Click Generate from prompt or paste your own JSON Schema
6. Click Validate
7. Save the column and test it on a few rows first

The schema must use an object at the top level. Keep it simple: strings, numbers, booleans, arrays, and nested objects work best.

Example: Subject + Body in One Run

Prompt

Write a journalist outreach draft for {{Name}} at {{Publication}} about {{Story}}.

Use this context:
- Recent coverage: {{Recent Articles}}
- Pitch angle: {{Pitch Angle}}

Return:
- subject: one subject line under 50 characters
- body: a concise email body in plain text, 2 short paragraphs, no signature, no "Subject:" label

Schema

{
  "type": "object",
  "properties": {
    "subject": {
      "type": "string",
      "description": "Email subject line"
    },
    "body": {
      "type": "string",
      "description": "Email body in plain text"
    }
  },
  "required": ["subject", "body"]
}

Sample output

{
  "subject": "A story angle on newsroom workflow",
  "body": "Hi Emily,\n\nI noticed your recent coverage on newsroom tooling and thought this data point might fit what you report on. We found that smaller PR teams are adopting AI workflow tools faster than larger teams because the operational lift is lower.\n\nIf helpful, I can send over the full findings and a short breakdown of what changed this quarter."
}

Another Prompt Example

Create a follow-up email for {{Name}} at {{Publication}} about {{Story}}.

Rules:
- subject: make it feel like a natural follow-up, under 45 characters
- body: 1 short follow-up email, friendly but direct
- do not repeat the original opening line word-for-word

Context:
- Previous angle: {{Pitch Angle}}
- Recent articles: {{Recent Articles}}

Return only the fields defined in the schema.

You can reuse the same two-field schema:

{
  "type": "object",
  "properties": {
    "subject": { "type": "string" },
    "body": { "type": "string" }
  },
  "required": ["subject", "body"]
}

How the Output Maps Back to Columns

JSON mode does not split one AI run into multiple columns automatically. Instead, downstream fields reference the JSON properties they need.

Formula columns cost 0 credits, so they are the cheapest way to split JSON output into separate visible columns after the AI run.

Credit Savings

Each Use AI run is billed per row. JSON mode still counts as one run, even if it returns multiple fields.

That means subject + body together usually cuts the AI cost in half.

The same percentage savings applies if you pick a higher-cost or lower-cost model. One JSON-mode run is still cheaper than two separate AI columns.

Practical Tips

1. Keep the schema tight. Only ask for fields you will actually use.
2. Put formatting rules in the prompt for each field, not just the schema.
3. Test on 5 to 10 rows before running the full table.
4. If you need visible split fields, add Formula columns instead of more AI columns.
5. If the model keeps returning messy structure, simplify the schema before adding more fields.