Build Casper's AI Prompt Guardrail Doc

Marketing Academy · Field Work●Content Marketing

All projects·From: AI Content Operations & Governance

MiniBuild the Asset· 30 minutes

Casper Sleep

Objective: Given Casper's brand voice notes, build a short guardrail document listing banned AI-tell words and two prompt rules that keep AI drafts on-brand and legally safe.

Casper's content team just approved using AI for first drafts of blog posts. Nobody has written the guardrail doc that goes into every prompt yet.

Write a banned-words list and two guardrail rules the AI must follow, grounded in Casper's brand voice and health-product compliance needs.

Before you start

What you'll need

Free path (everything below is enough to finish)

Google Docs

FreeDraft and store the reusable prompt guardrail document

Free, easy to paste into every future AI prompt as a shared reference

The process

1 step

Step 01 of 01

Guardrails and brand safety

The lesson recommends a digital style guide fed into every prompt, including a banned-words list for common AI-tell phrases like 'leverage,' 'game-changer,' and 'delve,' plus rules a gateway can enforce automatically.

Casper sells a health-adjacent product (sleep). What banned words and what two guardrail rules go into the prompt style guide before any draft ships?

Google Docs— Create a 'Casper AI Prompt Guardrails' doc with a banned-words section and a rules section.

Procedure

List five AI-tell words to ban outright ('leverage,' 'game-changer,' 'tapestry,' 'delve,' 'unlock')
Write Rule 1: no sleep-health claim ships without a cited study or an internal SME sign-off
Write Rule 2: no customer testimonial or statistic may be invented; only pull from an approved source doc

Sample output

BANNED WORDS: leverage, game-changer, tapestry, delve, unlock

RULE 1: Any claim about sleep quality, health, or medical benefit requires a cited study or SME sign-off before publish.
RULE 2: Never generate a customer quote, statistic, or testimonial. Pull only from the approved source doc provided in the prompt.

Healthy

A short, specific doc an editor can paste into every prompt, with rules a compliance reviewer could actually check against.

Unhealthy

A vague instruction like 'sound like Casper and don't lie,' which gives the AI and the editor nothing checkable.

What this means

Specific, checkable rules are what a compliance gate or an automated guardrail can actually enforce; vague tone notes cannot.

So what do I do about it?

Symptom	Action	Effort
Editors keep finding invented statistics in AI drafts	Add Rule 2 to the prompt template so the AI is told explicitly not to generate stats	5 min

YouYou can do this yourself, no engineering access required.

Final deliverable

A reusable 'AI Prompt Guardrails' doc with a banned-words list and two checkable brand/compliance rules.

See a reference example

Sample output

Zomato AI Prompt Guardrails (excerpt)

BANNED WORDS: leverage, revolutionize, delve, tapestry
RULE 1: No delivery-time or pricing claim ships without a cited internal ops number.
RULE 2: Never generate a restaurant rating or review quote; pull only from the verified reviews database.

Success criteria

You're done when you can:

Banned-words list has at least five real AI-tell words
Both rules are specific and checkable, not vague tone guidance