DEV Community

draftkit
draftkit

Posted on

10 AI Prompts I Use for Growth Marketing (Experiments, Activation Funnels, Retention Loops)

I've spent the last year running growth experiments for SaaS products using AI as a thought partner — not a content factory. The difference matters: most people ask AI to write growth copy. I ask it to think through experiments with me.

These 10 prompts have survived real use across activation funnels, retention analysis, win-back campaigns, and experiment readouts. Each one follows the same structure: Role → Context → Constraints → Output. That structure is why they work — the constraints do the heavy lifting, not the role assignment.

If you want the full library (200+ tested prompts across marketing, operations, sales, and product), I keep them at my Developer Product-Launch Prompt Pack ($9) and the larger SaaS Marketing Copy Pack ($49).


1. The Growth Hypothesis Framer

Most growth "experiments" are actually guesses dressed up in hypothesis language. This prompt forces a real hypothesis with a prediction, a falsification signal, and a kill criterion — before you spend a single hour building.

You are a growth marketing strategist. I will describe a feature change
or experiment idea. Frame it as a testable growth hypothesis using this
exact format:

HYPOTHESIS: We believe that [action] for [segment] will result in
[measurable outcome].
PREDICTION: We expect a [X]% change in [metric] within [timeframe].
FALSIFICATION SIGNAL: If [metric] does not move by at least [Y]%, the
hypothesis is false.
KILL CRITERION: Stop the experiment if [specific early signal] by day [N].
ASSUMPTION RISK: The riskiest untested assumption underlying this is [assumption].

Constraints:
- Do not use the words "leverage," "synergy," or "optimize."
- The falsification signal must be a number, not a feeling.
- The kill criterion must be actionable within 7 days, not "monitor and revisit."

Feature change: [DESCRIBE YOUR CHANGE]
Enter fullscreen mode Exit fullscreen mode

Why the constraints matter: Banning "optimize" forces specific language. The 7-day kill criterion stops experiments from drifting into "we need more data" limbo. The falsification signal being a number prevents the post-hoc rationalization where you declare victory on a secondary metric.


2. Activation Funnel Diagnostic

You are a growth analyst. I will paste our activation funnel steps with
conversion rates between each step. Identify the single biggest drop-off
and diagnose the likely cause.

For the biggest drop-off, provide:
1. THE BOTTLENECK: Step X → Step Y ([Z]% drop)
2. THREE POSSIBLE CAUSES, ranked by likelihood:
   - Cause A (most likely): [specific behavioral explanation]
   - Cause B: [specific behavioral explanation]
   - Cause C: [specific behavioral explanation]
3. ONE DIAGNOSTIC QUESTION per cause that would confirm or rule it out
   (a question you could ask 5 users in a 10-minute call).
4. THE FIX THAT IS NOT "ADD A TOOLTIP."

Constraints:
- Do not suggest "improve onboarding" or "add guidance."
- Each cause must reference a specific user action or inaction, not a vague "confusion."
- The diagnostic questions must be answerable by a user, not by analytics alone.

Funnel data: [PASTE STEPS + RATES]
Enter fullscreen mode Exit fullscreen mode

Why the constraints matter: "Improve onboarding" is not a fix — it's a category. Forcing specific behavioral explanations (e.g., "users complete email signup but never click the confirmation link because the value of confirming isn't stated on the success page") produces actionable diagnoses, not platitudes.


3. Retention Loop Designer

You are a growth marketer specializing in retention loops. I will describe
our product's core value action (the action that correlates with retention).

Design a retention loop around this action. Output:

1. THE TRIGGER: What external event or internal signal prompts the user
   to perform the core action again? (Must be specific — not "a notification.")
2. THE ACTION: The core value action, stated as a verb.
3. THE INVESTMENT: What the user puts in that increases their switching cost
   or makes the next trigger more effective.
4. THE LOOP VELOCITY: How fast does this loop complete? (Daily, weekly, monthly?)
5. ONE WAY THE LOOP CAN BREAK: The most likely failure point.

Then identify the SINGLE highest-leverage change to tighten the loop.

Constraints:
- Do not use "gamification" or "engagement" as solutions.
- The trigger must be an observable event, not a feeling.
- The investment must be something the user explicitly does, not passive data accumulation.

Core value action: [DESCRIBE]
Enter fullscreen mode Exit fullscreen mode

4. Win-Back Campaign Writer (No Guilt, No Discounts)

You are a retention marketer. I will describe a churned user segment and
when they last used the product. Write a 3-email win-back sequence.

Email 1 (Day 0): The "something changed" email — reference a specific
product improvement released since they left. No discount. No "we miss you."
Email 2 (Day 4): The "one specific use case" email — show ONE workflow
that solves a problem they likely still have. Include a 30-second setup.
Email 3 (Day 10): The "honest question" email — ask what would need to
be true for them to return. One sentence. No CTA button, just a reply-to.

Constraints:
- Subject lines must be lowercase, under 6 words, and must not contain
  the word "back," "miss," or "return."
- Zero discounts across all three emails.
- Email 3 must not have a button — only a plain-text reply prompt.
- Each email under 100 words.

Churned segment: [DESCRIBE + LAST ACTIVE DATE]
Enter fullscreen mode Exit fullscreen mode

Why no discounts: Discount-based win-backs train users to churn for deals. This sequence rebuilds the value proposition instead. The lowercase subject lines and no-button final email reduce the "marketing email" signal that gets you ignored.


5. North Star Metric Interrogator

You are a growth strategy advisor. I will state our current North Star
metric. Interrogate it by answering these 5 questions:

1. Does this metric measure value RECEIVED, or value REQUESTED? (If a user
   can inflate it without getting value, it's wrong.)
2. Does this metric lead or lag revenue? (Leading metrics let you act
   before the quarter ends.)
3. Can a single user's usage pattern distort this metric? (If one whale
   can move it 20%, it's not a north star — it's a lottery ticket.)
4. Does this metric align all teams, or does it create cross-functional conflict?
5. What behavior would a rational team member optimize for if they ONLY
   cared about this number? Is that behavior good for users?

After answering, suggest ONE alternative metric that addresses the biggest
weakness you identified.

Constraints:
- Do not define "north star metric" — assume I know.
- Be direct about whether my current metric is wrong. Do not hedge.

Current North Star metric: [DESCRIBE]
Enter fullscreen mode Exit fullscreen mode

6. A/B Test Readout (Pre-Mortem Edition)

You are a growth experimenter. I will describe an A/B test result
(variant, metric, lift, significance, sample size).

Instead of just declaring a winner, produce a pre-mortem readout:

1. THE RESULT: [variant] beat [control] by [X]% on [metric] (significant: Y/N)
2. WHAT THIS PROVES: The narrow, specific claim this result supports.
3. WHAT THIS DOES NOT PROVE: Three things a stakeholder might incorrectly
   infer from this result.
4. SEGMENTATION RISK: Name one user segment where this change might HURT,
   even if the aggregate looks positive.
5. NEXT EXPERIMENT: The single follow-up experiment that would confirm
   the mechanism (not just replicate the result).

Constraints:
- Do not say "roll it out" or "ship it" — that's a product decision, not an analysis.
- "What this does not prove" must contain at least one counterintuitive item.
- The next experiment must test the mechanism, not just scale the sample.

Test result: [DESCRIBE]
Enter fullscreen mode Exit fullscreen mode

Why this matters: Most experiment readouts either declare victory or demand more data. This prompt forces you to identify what you can't conclude — which is where bad growth decisions originate.


7. Referral Program Architect

You are a growth marketer who has designed referral programs for 20+ SaaS
companies. I will describe our product, pricing, and typical user persona.

Design a referral program. Output:

1. THE DOUBLE-SIDED INCENTIVE: What the referrer gets AND what the
   referred user gets. Must be asymmetric if one side values the reward more.
2. THE TIMING: When in the user lifecycle do we ask for a referral?
   (Must reference a specific moment of value realization, not "after signup.")
3. THE FRICTION MAP: List every step between "user decides to refer" and
   "referred user converts." Identify the step most likely to cause drop-off.
4. THE ABUSE VECTOR: How could a rational bad actor game this program?
5. THE KILL SIGNAL: What metric, if below [threshold] after [timeframe],
   means we shut down or redesign the program?

Constraints:
- Do not default to "give both sides a month free" — justify the specific reward.
- The timing must be event-based, not time-based.
- The abuse vector must be realistic, not paranoid.

Product/pricing/persona: [DESCRIBE]
Enter fullscreen mode Exit fullscreen mode

8. Churn Autopsy (From Exit Survey Data)

You are a retention analyst. I will paste exit survey responses (free-text)
from churned users over the past 30 days.

Perform a churn autopsy:
1. GROUP the reasons into clusters (max 5 clusters). Name each cluster
   with a behavior, not a sentiment. (e.g., "Hit the usage limit during
   a critical workflow" not "frustrated with limits.")
2. For each cluster, estimate the % of churn it explains.
3. For the TOP cluster only: propose ONE product or pricing change that
   would address the root cause. Must be specific enough to spec as a ticket.
4. Identify any cluster that is actually UNFIXABLE (e.g., "company shut down,"
   "switched to an enterprise tool we can't compete with") — these are not
   actionable and should be separated from fixable churn.

Constraints:
- Do not say "improve the product" — that is not a change.
- Clusters must be behavior-based, not emotion-based.
- The proposed change must reference the specific complaint language from the responses.

Exit survey responses: [PASTE]
Enter fullscreen mode Exit fullscreen mode

9. Viral Coefficient Reality Check

You are a growth modeler. I will provide our viral coefficient inputs
(number of invites per user, conversion rate of invites).

Calculate the K-factor, then perform a reality check:

1. THE MATH: K = [invites] × [conversion]. Show the number.
2. THE ASSUMPTION AUDIT: For each input, state whether it's measured,
   estimated, or aspirational. Flag any input that is "aspirational."
3. THE ORGANIC VS. INDUCED SPLIT: Of the invites per user, what % are
   organic (user would invite anyway) vs. induced (only because of a
   program/prompt)? If you can't split this, the K-factor is inflated.
4. THE CEILING: At what user count does the viral loop saturate?
   (Every viral loop has a ceiling — name a plausible one.)
5. THE HONEST VERDICT: Is this product actually viral, or are we forcing it?

Constraints:
- Do not say "increase invites per user" as a recommendation.
- If any input is aspirational, say so explicitly.
- The ceiling must be a number or range, not "eventually."

Inputs: [PASTE]
Enter fullscreen mode Exit fullscreen mode

10. Growth Experiment Backlog Prioritizer

You are a growth lead. I will paste a list of 10+ experiment ideas,
each with a one-line description.

Prioritize them using a modified ICE framework, but with a critical twist:

For each experiment, score:
- IMPACT (1-10): If this works, how much does it move the primary metric?
- CONFIDENCE (1-10): How sure are we it will work? (Subtract 3 points if
  we've never run a similar experiment.)
- EASE (1-10): How fast can we ship and measure this?

Then add a fourth column:
- REGRET: If we DON'T run this and a competitor does, how bad is that?
  (1-10, where 10 = "they'd win the market.")

Sort by (IMPACT × CONFIDENCE × EASE) + (REGRET × 2).

Output the top 5 ranked experiments with their scores and ONE LINE on
why each is ranked where it is.

Constraints:
- Confidence must be penalized for novelty — if it's a "new channel" or
  "unproven tactic," cap confidence at 4 regardless of how exciting it sounds.
- Do not rank based on excitement or novelty.
- The "why" line must reference a score component, not a gut feeling.

Experiment ideas: [PASTE LIST]
Enter fullscreen mode Exit fullscreen mode

Why the Regret column: ICE alone optimizes for safe, incremental wins. Adding a Regret-weighted term forces you to consider the strategic cost of not running a hard experiment — which is how categories get disrupted.


How I Test These Prompts

Every prompt above went through the same 3-step validation before I trusted it:

  1. Run it 3 times with the same input. If the outputs vary wildly in structure, the prompt's constraints are too loose.
  2. Generalize it. Swap in a different product/context. If the output breaks, the prompt is overfit to one scenario.
  3. Read it aloud. If the output sounds like it could appear on any marketing blog without modification, it's too generic — tighten the constraints until the output is specific to the input.

The prompts that survived this filter are the ones where the constraints do the thinking, not the role assignment. "You are a growth marketer" adds almost nothing. "The kill criterion must be actionable within 7 days" changes the entire output.


Where to Get More

These 10 are a slice of a larger system. If growth marketing prompts are useful to you, the full libraries are here:

Or browse the full storefront.


What growth prompts have worked for you? I'm always testing new structures — drop a comment with a prompt that outperformed your expectations.

Top comments (0)