Phi

Revenue Workflow

Niche Content Training Engine

Scrape and analyze thousands of high-performing posts from niche creators, extract patterns, and train a Claude Project to generate content in that voice with proven hooks and structure.

AdvancedTime to value · 3 to 4 weeks3 tools

The outcome

Produces content that matches proven niche patterns, reducing production time and increasing engagement rates without generic AI output.

How it flows

How it flows.

Source
5-10 niche creators
Process
Scrape posts & metrics
Store
3,156 raw posts
Process
Clean & normalize data
Output
Merged clean CSV
Output
All graphics downloaded
Output
Posts ranked by performance
Agent
Pattern analysis
Agent
Trained content engine
Drag to pan · scroll to zoom

    Build it, step by step

    Build it, step by step.

    1. 01

      Identify target creators

      Pick 5 to 10 creators who post consistently to your ICP, get strong engagement relative to followers, and occupy the same niche.

      • Manual selection or use Scripe for automated discovery
      • Filter for consistent posting cadence
      • Check engagement rate vs follower count
      • Verify clear niche positioning
    2. 02

      Scrape all content and metrics

      Use Apify to pull every post, caption, timestamp, and engagement metric from each creator into a raw dataset.

      • Scrape all historical posts from selected creators
      • Capture captions, likes, comments, reposts
      • Download post timestamps and media URLs
      • Export raw data as CSV or JSON
    3. 03

      Clean and normalize the dataset

      Upload raw data to Cursor. Let it remove duplicates, standardize columns, and structure fields for analysis.

      • Upload raw scrape output into Cursor
      • Remove duplicates and null entries
      • Standardize column names and formats
      • Extract structured fields for engagement, timestamps, topics
      • Export merged clean CSV
    4. 04

      Generate ranked outputs

      Produce a clean CSV, download all graphics, and calculate engagement rate and hook strength for every post.

      • Export merged clean CSV with all posts
      • Download all post graphics to local storage
      • Calculate engagement rate for each post
      • Rank hooks by performance
      • Tag posts by format and topic
    5. 05

      Run pattern analysis with Claude

      Upload the dataset into a Claude Project. Ask it to identify winning formats, best posting times, high-performing hooks, and reusable sentence structures.

      • Create new Claude Project and upload clean CSV
      • Identify which formats and topics drive engagement
      • Extract best posting times by day and hour
      • Rank hooks by engagement and format
      • Extract 700+ reusable high-performing sentences
      • Export structured JSON with hooks, transitions, CTAs
    6. 06

      Build the trained content engine

      Create a dedicated Claude Project loaded with thousands of niche posts, ranked hooks, and voice patterns. Prompt it to generate new content that mirrors the dataset.

      • Spin up new Claude Project for production
      • Load full dataset of 1000+ real niche posts
      • Add ranked hooks and sentence library
      • Map voice patterns and format templates
      • Test output against held-back validation posts
      • Deploy as reusable content generation engine

    The stack

    The stack.

    Want a pod to run this for you?

    Phi pods plug this workflow into your existing stack as Revenue Infrastructure, then operate it.