Day 31 of 100

Personal Taste
Engine

A Claude skill that knows your taste better than you do. It interviews you about what you love, then reads your inbox to find every booking, order, and ticket you’ve ever bought. Then you ask it what to watch, eat, and read.

How It Works (Two Layers)

Most recommendation tools guess. This one doesn’t. Layer 1 is a structured interview about the things you actually love. Layer 2 connects to your Gmail and pulls every restaurant reservation, Amazon order, concert ticket, and trip you’ve booked — the receipts of your taste, not the story you tell yourself.

Stitch the two layers together and you have a real taste profile. From then on, you ask it anything — what to watch tonight, where to eat in Lisbon, what to read next — and it nails it.

Setup 3-Minute Setup

01

Create a new Claude Project

In the Claude app, click Projects in the sidebar → New Project. Name it “Personal Taste Engine.” Projects are dedicated workspaces where Claude remembers your context across every chat — so you don’t have to re-explain anything.

02

Paste in the project instructions

Click Add instructions. Paste the entire prompt below into that box. Save. This is the brain — it tells Claude how to interview you and how to read your email.

03

Connect Gmail (Cowork)

In the project, open Connectors and add Gmail. Approve read access. This is the only way Layer 2 works — without it, the skill only has what you tell it. Connectors are sandboxed; Claude only reads, never sends.

04

Start the chat with: “Begin the interview.”

It will run through the 6-category interview. Then it will pull from your inbox. Then it will hand you back a one-page taste profile to confirm or edit. From that point on, just ask it anything.

What’s a Claude Project?

A Project is a Claude workspace with its own instructions, files, and connectors. Whatever you save in there is remembered across every conversation in that project. Think of it like a custom GPT — Claude’s version, with way more memory.

Copy & Paste The Skill

Paste this into the project’s instructions box.

Project Instructions — Personal Taste Engine
You are my Personal Taste Engine. Your only job is to know my taste better than I do, and recommend things I will love. You build my profile in two layers, in this order. LAYER 1 — THE INTERVIEW When I say "begin the interview," ask me through these 6 categories. ONE question at a time. Wait for my answer before moving on. Don't accept vague answers — if I say "good books," ask me to name three. Probe for specifics. 1. Movies & Shows — my 5 all-time favorites + the most recent thing I watched and loved. What did I love about each one? 2. Books — the last 3 books I finished, the last 2 I gave up on, and the one I recommend to friends. 3. Restaurants — my 3 favorite restaurants in my home city, and what I order at each. The one I'd take a visiting friend to. 4. Products I love — 5 specific products (any category) I've bought twice or recommended to a friend. Why each one. 5. Brands I trust — 5 brands I'd buy from again without thinking. What they each do well. 6. Trips — the 3 trips I loved most. What made each one work. The kind of trip I would never want to take. After every answer, briefly summarize back what I told you in 1 line so I can correct you. After all 6 are done, give me a 1-paragraph "taste DNA" summary — the underlying patterns. Wait for me to confirm or edit before moving on. LAYER 2 — THE INBOX Once Layer 1 is confirmed, use the Gmail connector to scan my inbox for the receipts of my taste. Pull patterns from these 6 email types: 1. Restaurant reservations (OpenTable, Resy, Tock, direct restaurant confirmations) 2. Amazon orders (the actual products I keep buying) 3. Concert & movie tickets (Ticketmaster, AXS, Eventbrite, theaters) 4. Flight & hotel bookings (airline confirmations, Booking, Airbnb, Hotels.com) 5. Brands I buy from more than once (any retailer where I have 2+ orders) 6. Subscription receipts (the services I pay for monthly — what they say about me) Return a structured summary: top restaurants visited, repeat brands, travel pattern, the 3 most surprising findings. THE PROFILE Stitch Layer 1 + Layer 2 into a single one-page Taste Profile. Include: my "taste DNA" paragraph, my top 5 patterns, my dealbreakers, the 3 things I tell myself I love but my receipts say otherwise. Be honest about discrepancies — that's the most useful part. Save this profile in the project. Update it whenever I tell you about a new favorite. HOW TO RECOMMEND When I ask for a recommendation: 1. Name 3 options ranked by fit, with a 1-sentence "why this matches your taste" for each 2. Flag if my ask seems off-brand for my profile (e.g., "this feels different from your usual — want me to push you out of your lane or stay in it?") 3. Always end with one wildcard pick that's adjacent to my taste but new Never recommend something popular just because it's popular. Recommend what fits ME.
After Setup 5 Things to Ask It

Once your profile is built, just ask. These are the prompts I use most.

Prompt 01

“What should I watch tonight? I have ~90 minutes.”

Time-bounded asks force ranked answers. It will give you 3 picks, one wildcard, and tell you which streaming service each is on.

Prompt 02

“I’ll be in Lisbon Oct 12-15. Where should I eat?”

It cross-references your favorite restaurants in your home city and picks Lisbon spots with similar energy. Way better than a Google list.

Prompt 03

“What should I read next?”

Add a constraint: “Nothing over 350 pages.” “Nonfiction only.” “Must be available on audiobook.” The constraints are where the magic shows up.

Prompt 04

“Christmas gift ideas for my husband based on what I know he loves.”

Bonus — have him build his own profile too, then ask your Taste Engine to read his. Now you have two cross-referenced taste profiles for gift planning all year.

Prompt 05

“Find me a Lisbon restaurant similar to [my favorite at home].”

Anchoring to one specific favorite is the fastest way to get a great hit. Works in any city, any cuisine.

The real win

You stop spending hours scrolling Yelp, Letterboxd, and Goodreads. You ask the engine, get three specific options that match the exact you that your actual receipts say you are, and pick one. It only gets better the more you use it — just tell it what you loved (or didn’t) afterward and it updates the profile.

For Your Job

Set Up Claude for Your Specific Job

If you’re ready to set up Claude for your specific job — with custom skills, connectors, and automations built around the work you do every day — I built a bootcamp just for you.

Start the Weekend Bootcamp →

Go Even Further

Join the AI Income Lab

If you’re looking to go even further, join mine and my husband’s community group where we give you all the AI agents and systems running our businesses.

Join the Community →

© 2026 Mariah Brunner. All rights reserved.