STEP 2
Paste the prompt
Now you give Claude its job. Open a new Cowork chat and paste the full prompt below. It tells Claude exactly what data to pull through Windsor, how to analyze it, and what to build. Copy it word for word, paste it, hit enter. Claude takes it from there.
You are my Instagram content analyst. Your job is to analyze my last 25 to 50 posts and build me an interactive HTML artifact that shows exactly what is working, what is not, and what I should do next.
## Step 1 — Pull the data
Use the Windsor.ai connector to fetch my Instagram data. For each post you need:
Performance metrics:
- Views and reach
- Likes and comments
- Saves and shares
- Total interactions
- Average watch time (reels only)
Content data:
- Caption text
- Audio transcript of the video (use Windsor.ai to pull this — do not skip it, this is essential for hook analysis)
- Post type (reel, carousel, image)
- Date posted
Fetch as many posts as Windsor.ai allows, up to 50. Sort them newest to oldest before analyzing.
## Step 2 — Analyze the data
Before building anything, analyze every post. Look for:
1. Hook patterns — what words, phrases, or sentence structures appear in the highest-performing captions and transcripts? What do the top 10% share that the bottom 10% do not?
2. CTA patterns — which posts used a "comment [word]" CTA? How did comment volume compare to posts without one?
3. Format performance — average views and engagement for reels vs carousels vs images
4. Topic clusters — group posts by topic (AI tools, business advice, tutorials, hot takes, etc.) and rank the clusters by average performance
5. Hook type performance — categorize each hook as: contrarian take, industry death sentence, specific outcome, anti-paywall/free giveaway, curiosity gap, question, or other. Rank by average views
6. Trending topics — based on what is working plus what is happening in the space right now, what topics should be tested next?
7. Experiments to run — based on patterns in the data, what 3 format or content experiments would move the needle most?
## Step 3 — Build the artifact
Create a fully interactive HTML artifact. This is not a screen recording — it should be a real tool the user can interact with.
### Design specs
The artifact must look premium. Use these exact design guidelines:
Fonts (load from Google Fonts):
- Fira Code — for all numbers, stats, labels, tags, and monospace elements
- Fira Sans — for all body text, titles, and descriptions
Colors:
- Background: true black #000000 with a subtle orange radial glow at the top (rgba(249,115,22,0.06))
- Cards: background: rgba(255,255,255,0.05) with border: 1px solid rgba(255,255,255,0.08) and backdrop-filter: blur(20px)
- Primary text: #ffffff
- Secondary text: rgba(255,255,255,0.55)
- Muted text: rgba(255,255,255,0.25)
- Brand accent: #f97316 (orange) with glow text-shadow: 0 0 30px rgba(249,115,22,0.35)
- Win color: #22c55e (green) with glow text-shadow: 0 0 20px rgba(34,197,94,0.3)
- Problem color: #ef4444 (red)
- Amber accent: #f59e0b
Section headers:
- Font: Fira Sans, 18px, font-weight 700, color white
- Followed by a horizontal rule that fades from the accent color to transparent
Cards:
- All cards use glassmorphism: backdrop-filter: blur(20px), semi-transparent background, thin border
- cursor: pointer on all clickable/hoverable elements
- Hover: transform: translateY(-2px) with transition: 0.2s
- Top metric cards get a subtle colored glow blob in the corner
Winning insight cards (green glass):
- background: rgba(34,197,94,0.08), border: 1px solid rgba(34,197,94,0.25)
- Green top edge line: border-top: 2px solid rgba(34,197,94,0.5)
Problem cards (red glass):
- background: rgba(239,68,68,0.08), border: 1px solid rgba(239,68,68,0.25)
- Red top edge line: border-top: 2px solid rgba(239,68,68,0.5)
### Artifact sections (build all of these)
1. Sticky nav bar
- Account name/handle on the left with a pulsing orange dot
- Jump links to each section: Metrics, Posts, Insights, Experiments
- Small date badge on the right showing when the analysis was run
2. Key metrics row (4 cards)
- Top post views
- Top comment count
- Median reel views
- Content type split (% reels vs carousels)
- Large monospaced numbers with colored glow on the standout metrics
3. Top 10 posts list
- Rank number, caption snippet, tag pills (hook type, CTA type, format), view count on the right
- Performance tier dot: orange for viral (top 10%), green for strong (top 30%), amber for average, gray for underperformed
- Gold left border glow on rank 1 and 2
- Rows shift right 4px on hover
- Built with Chart.js 4.4.1 (load from https://cdnjs.cloudflare.com/ajax/libs/Chart.js/4.4.1/chart.umd.js)
4. Top 10 bar chart
- Bars colored by tier (orange, green, amber, gray)
- Custom dark tooltips showing post title and exact number
- No default Chart.js tooltip styling
5. Top 3 carousels
- Green-tinted glassmorphism cards with rank numbers (01, 02, 03)
- Comment count in large green monospace, caption excerpt below
6. What is working (6 green glass cards)
- The 6 biggest patterns driving your best results
- One insight per card with a title and 2 to 3 sentences of explanation
- Lead with the data, then the actionable implication
7. What is holding you back (4 red glass cards)
- The 4 biggest drags on performance
- Same format as above but red glass
8. 5 trending topics to test
- Numbered list (01 through 05) in a single glass panel
- Bold topic title, 2 sentence rationale per topic grounded in the data
9. 3 format experiments for the next 30 days
- Orange-bordered glass cards
- Each has a label (Experiment 01, 02, 03), a title, and 3 to 4 sentences on what to test and why
Footer:
- [X] posts analyzed · built with claude · [handle]
- Fira Code, muted color, right-aligned timestamp
### Interactivity requirements
- All section jump links must smooth-scroll
- Post rows must highlight on hover
- Chart bars must animate in on load (staggered)
- The artifact must be fully responsive (works on mobile at 375px and desktop at 1280px)
- No emojis in the UI. Use clean text labels only
## Tone for written analysis
Be direct. No corporate fluff. Call out what is actually working and what is not. If something is performing badly, say so clearly and explain why based on the data. The person reading this wants real talk, not a participation trophy.
Never use em dashes. Use commas, periods, or parentheses instead.
Start by connecting to Windsor.ai and pulling the data. Tell me how many posts you retrieved before building the artifact.
Tip: Claude will tell you how many posts it retrieved before it builds anything. If it pulled fewer than you expected, just tell it to grab more.