How to use Draftship with Google Gemini

Gemini's edge is long context. You can paste your entire brand archive (every past email you've shipped) into a single prompt and ask for a new draft that matches. With over a million tokens of context, the voice anchor problem mostly disappears.

Three Gemini surfaces

• Gemini app (gemini.google.com): consumer chat, fastest path for non-developers.
• Gemini API: developer access via Google AI Studio. Standard endpoints.
• Vertex AI: enterprise path with VPC, audit logs, BYOK. Required for regulated industries.

Long context as the differentiator

For email drafting, the big win is the ability to paste 50 or 100 past emails as voice anchors instead of just two or three. The voice consistency across drafts improves dramatically.

Prompt template:

txt
You're writing a marketing email for {{ company }}.

Past emails (this is the full archive; match this voice exactly):
[paste 30 to 100 past emails, separated by ---]

Brief for the new email:
- Goal: {{ what action }}
- Audience: {{ who reads this }}
- Constraints: subject under 50 chars, 2-4 paragraphs, one CTA.

Return as JSON:
{
  "subject": "...",
  "preheader": "...",
  "blocks": [...]
}

The model picks up patterns from the archive that abstract instructions can't capture: phrasing tics, sentence rhythm, even how you transition between paragraphs.

Structured output via Vertex AI

For batch generation, Vertex AI supports response schemas similar to OpenAI's:

bash
curl -X POST https://us-central1-aiplatform.googleapis.com/v1/projects/PROJECT/locations/us-central1/publishers/google/models/gemini-2.5-pro:generateContent \
  -H 'Authorization: Bearer $(gcloud auth print-access-token)' \
  -d '{
    "contents": [{ "role": "user", "parts": [{ "text": "Your prompt" }] }],
    "generationConfig": {
      "responseMimeType": "application/json",
      "responseSchema": { ... }
    }
  }'

The responseSchema enforces shape. Map the parsed JSON onto Draftship blocks programmatically.

Image generation via Imagen

Google's Imagen (accessible via Vertex AI or Gemini app) generates hero images. Workflow:

1. Prompt Imagen for a 16:9 hero, no text. 2. Download the result. 3. Upload to your CDN or ESP image library. 4. Paste URL into Draftship's Image block.

Imagen tends toward photographic realism. For illustration-style hero art, Midjourney or DALL-E often produce better results.

Safety filter quirks

Gemini's safety filters block prompts they read as harmful, including some legitimate marketing prompts. Examples that get flagged:

• "Create urgency around the deadline" (read as manipulative)
• "Persuade users to upgrade" (read as deceptive)

The fix: rephrase neutrally. "Communicate the upgrade benefits" instead of "persuade." Vertex AI offers tighter control over filter aggressiveness for enterprise accounts.

Test send checklist

• Validate JSON structure before importing to Draftship.
• Run output through the spam word checker.
• Run the size checker on the final HTML.
• Send a real test through your ESP.

When Gemini wins over Claude or GPT

The long-context use case: you have hundreds of past emails and want voice consistency. Gemini handles the archive in a single prompt. Claude and GPT can do this too with Project Knowledge or RAG, but Gemini's raw context window is the simplest path.

For Claude prompt patterns, see Use Draftship with Anthropic Claude.