Create emails with Launchpad — prompting best practices

If emails generated by Launchpad require structural fixes, tone corrections, or missing elements before they are ready to send, the prompt is usually the root cause. This guide covers seven practices to specify upfront — reducing manual rework and getting production-ready output from the first draft.

1. Use exact reference language, not approximation language

Approximation language — for example, "...like [template name]" — signals inspiration, not structure. Launchpad interprets reference instructions very literally. Small wording changes produce very different outputs.

Do: "Maintain structure, style, and color palette from @[template name] exactly as-is — replace only the content below." Avoid: "Create a new email based off the designs like [template name]."

Use the table below as a translation guide.

Reference language: translation key

No reference available? Add: "Do not add any sections or structural elements beyond what is described below." This prevents Launchpad from making its own layout decisions.

2. Configure universal elements once using Account Instructions

Any element that appears in every email — header, footer, promo bar, nav bar, brand rules — belongs in Account Instructions, not in individual prompts. Repeating these per prompt is unnecessary overhead.

Do: Add standard headers, footers, brand hex values, CTA conventions, and guardrails to Account Instructions once. Avoid: Repeating structural rules and brand constraints at the start of every prompt.

What to configure:

• Standard header, footer, and promo bar — names, positions, and any required HTML wrappers (for example, a centered <table> with max-width: 600px)
• Brand colors — primary, accent, and hierarchy hex values
• CTA button styling conventions
• Navigation bar structure and style
• Standing guardrails — for example, no emojis in HTML, dynamic copyright year using {{ 'now' | date: "%Y" }}

Also, define the sequence your assets appear in — order matters. Launchpad will place assets where it infers they belong if no order is specified, and getting it wrong requires full layout restructuring to fix.

If you use a reference template (best practice 1), Launchpad extracts shared assets directly — no need to list them in the prompt.

3. Always include exact CTA URLs

Do not rely on Launchpad to infer destination URLs. If URLs are not provided, Launchpad will guess — and will choose incorrectly.

Do: Provide the exact canonical URL for every CTA. Map each URL to its section if the email has multiple CTAs. Avoid: Leaving URLs unspecified or describing the destination in general terms.

If specific image assets are required, include their CDN URLs in the prompt. Without them, Launchpad may use placeholder images.

4. Define tone and objective before generating content

Without guidance, Launchpad defaults to your account-level brand voice. If that voice is authoritative but the email needs to be warm, urgent, or conversational, the subject line and body copy will require rework. Specify tone and objective upfront — Launchpad will generate the subject line, preheader, and copy from them.

Do: Specify objective (awareness, conversion, trust-building, urgency), tone register, audience context, and key stats to include. Let Launchpad generate the subject line and copy. Avoid: Writing the subject line yourself in the first prompt, or leaving tone unspecified.

If the output needs adjustment, iterate in the same conversation. Override specific elements only when you have a concrete requirement.

5. Label reference data separately from content

When a prompt includes research, raw data, or background documents alongside a content brief, Launchpad has to infer what to use and what to ignore. Without clear labels, it may mix source language into body copy, overshoot the intended length, or miss specific figures entirely.

Do: Label each section of your prompt — FOR REFERENCE ONLY, Stats — use these numbers only, Body copy brief — so Launchpad knows its purpose. Avoid: Pasting a research dump and a content brief together with no distinction between them.

Three labels to use:

• FOR REFERENCE ONLY — background text Launchpad draws context from, but does not copy verbatim into the email
• Stats — use these numbers only — specific figures for a statistics section; list only what should appear
• Body copy brief — the tone, angle, and specific facts to write from; this is the active instruction

6. Specify personalization explicitly — in both directions

Personalization requires explicit direction both when you want it and when you don't. Find your attribute keys in your Blueshift account schema before writing the prompt.

Do: Specify the attribute, logic, and asset for each personalization variant. Or explicitly state that no conditional logic should apply. Avoid: Leaving personalization intent ambiguous — Launchpad will make its own decision.

When you want conditional personalization:

Find the relevant attribute key in your account schema, then specify the attribute, logic, and exact asset for each variant. If the reference template already uses location or behavioral personalization, explicitly tell Launchpad to carry it over — it will not do this automatically.

Personalize the hero image based on [your_attribute_name]:
  - If value is '[A]' → use this image: [Image CDN URL]
  - All other values (default) → use this image: [Default Image CDN URL]

When you do not want conditional personalization:

Use this single image as the hero: [CDN URL].
Do not apply conditional logic — all users must see the same image.

Personalized preheaders: By default, Launchpad sets the preheader as a static string. For per-segment preheader text, specify this explicitly:

Include a personalized preheader in a hidden HTML div at the top of the body.
Use {{ user.firstname }} and {{ user.last_location_state }} where available,
with appropriate fallbacks for unknown values.

7. Avoid overstatements in claims and statistics

Launchpad defaults to confident, assertive language. For regulated or high-accuracy contexts — finance, insurance, healthcare, or editorial content — the default output may require legal or accuracy correction. Instruct Launchpad to qualify claims upfront.

Do: Instruct Launchpad to use "up to," "may," "sometimes," and "could" for statistics and product claims. Anchor statements to their source where required. Avoid: Letting Launchpad default to absolute language — "always," "best," "guaranteed," "typically" — for claims that require qualification.

Prompt examples by scenario

Use these as starting templates — replace placeholders with your own references, content, and URLs.