Personalization that feels personal, not creepy
The same recommendation can feel like a great shop assistant or like being followed home. The difference is where the data came from.
Using what a customer told you or bought feels like service. Leaking what they never shared feels like surveillance. AI is great at generating relevant picks; your job is to keep them on the right side of the creepy line.
Fastest path: one prompt, end to end
🤖 AI prompt — paste into ChatGPT / Claude
You are a personalization reviewer. Use only my details (invent nothing).
Data I have on customers: [list, e.g. stated preferences from a quiz, purchase history, email clicks, on-site browsing]
Where I want to personalize: [emails / product page / returning-visitor homepage]
My typical reorder cycle: [e.g. ~40 days]
Do this:
1. Split my data into three buckets: knowingly given (zero-party, purchase history), inferred/tracked (browsing, opens), and off-limits (anything they don't know I hold).
2. For each place I personalize, suggest moves that lean on the knowingly-given bucket, use inferred data only subtly, and never surface off-limits data.
3. Flag any idea that would reveal I'm tracking something they didn't share, and give a warmer alternative.
4. Recommend replenishment timing from my reorder cycle, framed as a helpful reminder, not I know you're running low.
If you're unsure which bucket a data point is in, ask; do not assume.
Output: the three buckets + safe personalization moves per surface + creepy-line flags.
Or do it in 4 steps
- Sort your data by how you got it. Knowingly given (quiz answers from zero-party-data, past orders) is safe to use openly. Inferred or tracked (browsing, opens) is fine to act on quietly. Anything they'd be surprised you know is off-limits.
- Recommend from what they chose or bought. Their stated skin type, their last order, their favorite category, use these out loud. It reads as a shop assistant who remembers them.
- Time replenishment from the usage cycle, gently. For consumables, a reminder just before they run out is the highest-value personalization there is. Frame it as helpful (time for a refill?), never as we've been counting your days.
- Personalize the return visit lightly. A returning visitor can see their recent category or a welcome back, but don't parade every page they viewed. Subtle beats spooky every time.
Worked example (labeled): A supplement store personalizes two ways. Good: it recommends a refill of the exact product you bought, timed to a 30-day supply, and greets return visitors with their chosen goal from the onboarding quiz. Creepy (cut): a popup saying you looked at this 4 times this week.
Same customer, same data warehouse, opposite feeling. One uses what she gave knowingly; the other exposes what she didn't know was tracked.
Use knowingly-given data out loud, inferred data quietly, and hidden data never; that's the whole line.