Read the private Reels data no other tool can see, then have Claude tell you exactly what to do next.
Most analytics tools show you the same three numbers: views, likes, follows. None of them tell you which hook actually landed, which Reels the algorithm pushed, or which ones bombed even though the cover earned the click.
That deeper data does exist. It is the private performance data that only you, the account owner, can see through Instagram's own API. In this guide you connect Claude to your Instagram and have it build you a personal hook quality report. No code, no scrapers, no guesswork.
This project runs my Instagram growth experiment. You are my Instagram
growth strategist.
My brand is [your brand, and who you help]. My voice is [describe your voice
in a few words]. I post manually for now.
Your job: read my real Instagram data through the Composio connector, audit
performance, recommend high-leverage changes, and track my growth over 30
days. Always be blunt and specific, rank every recommendation by impact, and
tie it to the actual numbers. Capture a baseline today and compare against it
in every future session.https://connect.composio.dev/mcpOnce connected, paste the prompt below into Cowork. It pulls everything, calculates the metrics, and builds your report in chat plus a saved HTML file you can open in a browser.
You have access to my Instagram through the Composio connector. Run a deep
performance audit of my Reels and build me a full report. Work through this
exactly and do not ask me questions, just run it.
WHAT TO PULL
1. Confirm the connection: pull my account info (username, follower count) and
account level insights for the last 30 days (reach, impressions, profile views).
2. Pull my recent media (up to my last 30 posts) and filter to Reels and video
from the last 14 days. If I have fewer than 8 Reels in that window, widen to
the last 30 days so the comparison is meaningful.
3. For each Reel, pull its insights: reach, views, likes, comments, shares,
saved, total interactions, and average watch time. If a single metric errors
for a post, skip that metric and keep going. Also capture each Reel's caption,
permalink, thumbnail, and video duration.
METRICS TO COMPUTE (per Reel)
- watch_s = average watch time in seconds (divide by 1000 if it returns in ms)
- hook_rate = watch_s / video duration in seconds
- replay_rate = views / reach
- share_pct = shares / reach x 100
- save_pct = saved / reach x 100
- hook_score = watch_s x sqrt(reach) x (1 + share_pct/100 + save_pct/200)
CORE BREAKDOWN
1. Headline insight: my top Reel's reach divided by my median Reel's reach,
stated as "Your top Reel reached Nx more than your median." In one line, tell
me whether one hit is carrying me or my volume is doing the work.
2. Caption pattern callout: words or phrases that appear two or more times in my
top 5 Reels by hook_score but never in my bottom 5, and the reverse. Tell me
what my winners say that my losers do not.
3. Action grid in three buckets. DO MORE: my top 3 by hook_score with why each
worked. STOP: my bottom 3 with why each failed. FIX: Reels that pulled people
in but did not hold them (strong reach, weak watch time), the packaging wins
with delivery problems.
4. Quick stats: reach concentration (percent of total reach from my top 3),
watch-time gap (average watch_s of my top half vs bottom half), strong-hook
count (watch_s of 12 or more), replay-winner count (replay_rate of 1.2 or more).
5. Full ranked list of every Reel by hook_score. Each shows a caption preview, a
one-line diagnosis tag from this exact set, no em dashes: "Winner, all three
axes worked", "Strong hook, IG did not push it", "People clicked, content did
not hold", "Sharable concept, weak delivery", "Weak on all axes, kill this
format", "Hook landed", "Underperformed". Then three normalized 0 to 100 bars
(Hook from watch_s, Reach, Viral from shares plus saves), the hook_score, and
a link to the post.
THREE EXTRA SECTIONS
6. Best time to post. Pull my online-followers data and tell me the best 3 hours
to post on each weekday, as a simple weekday by hour table.
7. Hook framing analysis. Tag each Reel caption as challenge, curiosity, list,
story, or how-to, then show which framings have the highest average hook_score.
8. Competitor and niche overlay. Pull recent top public posts from one or two
hashtags I compete on (infer them from my captions, only ask if you cannot).
Overlay their public engagement (likes, comments, views where visible) next to
my averages. Label these clearly as public-only, because watch time, reach,
and saves are private to each account owner.
OUTPUT
Give me the full breakdown in chat first so I can read it. Then save a
self-contained report.html I can open in a browser: light mode, thumbnails
embedded inline as base64 so the file is portable. Order: headline insight and
pattern callout at top, the three-bucket action grid, the best-time-to-post
table, the hook framing winners, the competitor overlay, the quick stats, then
the full ranked card list with top 3 cards bordered green, bottom 3 red, and
fix-bucket amber. Be blunt and specific. Tie every call to the numbers. Run it now.Your report reads top to bottom as a decision tool, not a dashboard:
This is Episode 1. The audit. The build from here: