Creating Personalizations

Personalization columns let you generate custom content for each row in your table — pitch angles, subject lines, opening hooks, or full email drafts. Hand-matched, not mass-blasted.

How It Works

1. Add a Use AI column to your workflow
2. Write a prompt that references other columns (e.g., {{Recent Articles}}, {{Beat}})
3. Run the column — AI generates personalized output for each row

Example Prompts

Pitch Angle

Based on this journalist's recent articles: {{Recent Articles}}

Suggest a pitch angle for our product that aligns with their coverage. 
Be specific and reference their actual work.

Subject Line

Write a compelling email subject line for a pitch to {{Name}} at {{Publication}}.
They cover {{Beat}} and recently wrote about {{Recent Articles}}.
Make it specific, not generic.

Opening Hook

Write a 2-sentence opening for a pitch email to {{Name}}.
Reference their article "{{Most Recent Article Title}}" and connect it to our announcement about your product launch.

AI Email Columns

When you need both subject and body, the cleanest setup is one Use AI column in JSON Output Mode.

Recommended pattern: one AI run -> reference the fields downstream

1. Outreach Draft (Use AI, JSON mode): generate subject and body
2. Gmail/Outlook draft column: use {{Outreach Draft}}.subject and {{Outreach Draft}}.body
3. Optional: add free Formula columns if you want separate visible subject/body columns

Why this works: subject and body come from the same reasoning step, stay in sync, and cost one AI run instead of two.

If you stay in plain text mode, the older draft -> extract subject -> clean body pattern still works. That fallback is useful when you already have legacy text prompts or need a cleanup pass on messy output.

Text-mode fallback: subject extraction prompt

Extract only the subject line from this draft.
Return one line of plain text only.

Draft:
{{Draft Email}}

Text-mode fallback: body cleanup prompt

Clean this email body for journalist outreach.

Rules:
- Remove "Subject:" lines, markdown section breaks, and divider lines
- Replace em dashes with commas or periods
- Remove generic openers like "Dear client" and "Been following your work"
- Keep it concise and natural
- Do not add a signature
- Only output the cleaned up version

Draft:
{{Draft Email}}

Removing AI Giveaways

Common tells to remove before sending:

• Section breaks (---, ###, **)
• Em dashes (—) used repeatedly
• Generic openings like "Dear client"
• Flattery filler like "Been following your work"

Prompt tips for cleaner first-pass output:

1. Ban specific phrases directly in the prompt ("Do not use 'Dear client' or 'Been following your work'")
2. Specify tone ("write like a concise human PR pitch, no template language")
3. Force strict output mode ("only output the final email body, no notes or labels")

Signature Handling

Do not include signatures in AI-generated body columns.

Set signature behavior in your campaign send column instead:

1. Open your Gmail Send Email or Outlook Send Email column config
2. Open compose settings
3. Set Signature to Default signature or a Custom signature
4. Keep the AI body column signature-free

Use No signature only when you plan to add one manually outside Medialyst.

Best Practices

Be specific in your prompts. "Write a good pitch" produces generic output. "Reference their recent article about X and connect it to Y" produces evidence-backed content.

Include context columns. The more data we have (beat, recent articles, publication), the better the personalization.

Review before sending. AI output is a starting point. Edit for accuracy and tone before reaching out — journalists notice the difference.

Quality Control Workflow

Use two AI passes plus manual review:

1. Generator column creates the first draft
2. Validator/Cleanup column rewrites for tone, clarity, and banned phrases
3. Manual review confirms facts, names, and publication fit before sending

Validation prompts reduce mistakes, but they do not eliminate hallucinations. A model can still confidently keep or rephrase incorrect claims if the source data is wrong or incomplete.

Cost Optimization

Personalizations use credits per row. To stay efficient:

1. Filter your list to the 10-20 journalists most likely to care about your story
2. Use Run first 10 rows to test your prompt before running on all rows
3. Start with smaller models for simple tasks, larger models for complex research