Most of us make more of whatever got the most likes. The problem is that likes and follows are two completely different behaviours, and your most-liked post is very often not the one bringing new people in. This fixes that by reading your own numbers instead of generic advice.
When I ran it on my own grid it was humbling. My almost-400k-view explainer reel converted close to nobody, and the quiet posts I nearly stopped making turned out to be the two lanes doing all the real work. Views are not growth, and your data will tell you that faster than any guru will.
start hereYou are giving one model your whole posting history at once, which is something you cannot do inside Instagram itself. Instagram shows you one post at a time. Claude reads across all of them and finds the pattern that connects your winners, then names the pattern out loud so you can repeat it on purpose.
The value is that it reads YOUR numbers. It finds the two or three formats that convert strangers into followers on your specific account, tells you which posts only pull likes, and then you make more of the first kind.
Same analysis, two different ways of getting your numbers in front of Claude. Start with the free one. It takes about fifteen minutes and it works on a free Claude account.
You do not need the extension to get the interesting half of this. You just need about twenty rows of honest data.
Below are my last 90 days of Instagram posts. The first block is my ten best performers and the second block is my ten worst, and each line has the link, format, topic, views, saves, shares and follows where the number was available.
Read across all twenty and find the pattern, not the individual posts. First, name the two or three CONTENT LANES that actually convert strangers into followers on my account, and say what the posts in each lane have in common: the topic, the promise in the hook, the format, and the length. Second, name the lanes that are dead for me, meaning they take real effort and return nothing, and be blunt about it. Third, call out any post with high views and low follows and explain what most likely went wrong, because a post that reached a lot of people and grew nobody is a more useful lesson than a flop.
Then check for these specific things and tell me if any of them show up in my data: posts that only work in one format (for example carousels that only win when they are actual numbered lists), reposts or cross-posts that underperform my originals, and topics I have quietly stopped making even though they perform. Finish with the five things you would film next, each one tied to a specific winning post of mine and the reason it should work.
Where the data is too thin to support a confident answer, say so instead of guessing.These are the patterns that showed up when I ran this on my own account, and they are the ones worth checking for on yours. None of them are visible from inside Instagram one post at a time.
Claude in Chrome is a browser extension that lets Claude see and act inside your Chrome tabs: read what is on screen, click, scroll, switch tabs, and pull it all into one place. Instead of copying numbers into a list by hand, Claude reads your Instagram insights directly and does the analysis while you watch.
Open the Claude side panel while you are sitting on your Instagram professional dashboard, then paste this. Everything after this section works with either route, because from here on Claude is analysing a table it already has.
You can see my Instagram professional dashboard in this tab. Go through my posts and Reels from the last six months and pull, for each one: reach, saves, sends, and how many follows it drove (the post-watch behaviour or follows-from-this-post figure), not just likes. If a number is not visible for a post, write "not shown" rather than estimating it.
Then rank every post by follows gained, highest to lowest, and give me a clean table with these columns: post, format (Reel, carousel or photo), topic, reach, saves, sends, follows. Treat likes as a vanity metric and never use them as the ranking.
From that table, tell me the three content formats or topics that convert the most strangers into followers on MY account, and the ones that get likes but do not grow me. Explain the pattern in plain language so I know exactly what to make more of, and name the single post I should study hardest and why.Once the ranked table exists, run this in the same chat so the plan is built from your data rather than from scratch.
Based on that ranking, write me a two-week content plan. Take my top three converting formats or topics and give me four post ideas for each, staying inside the lane that already works but without repeating a post I have made.
For every idea, give me the format, the hook angle in one line, and which of my past winners it is modelled on, so I can see the logic. Keep the ideas realistic for one person to film and edit alone, no studio, no crew, no props I would have to buy.
Then flag any idea that is only a small variation of something I posted in the last eight weeks, because I would rather rework it than post something my audience has already seen.This is where most people stop, and it is the step that actually changes your week. The report is a diagnosis. Scripts are the treatment. Run this next.
Take my three highest-follow posts from the table above. For each one, work out what specifically made it convert: the promise in the hook, the structure of the middle, the payoff, and the call to action. Write that out as three short "why it worked" breakdowns before you write anything new.
Then write me ten short-form video scripts that reuse those working mechanics on new topics. Each script needs a hook line of under twelve words, three to five beats in the middle, and a closing line, all in spoken language I would actually say out loud, aiming for about 45 to 60 seconds when read at a natural pace.
Label each script with which winner it borrows its structure from. Do not copy the topic of the original post, only the mechanic, and avoid any hook that starts with a question I would not really ask a friend.This is what turns a one-off audit into a system you can watch trend over time.
Turn this into a simple dashboard I can update once a week. Give me a table with these columns: week, top post by follows, follows gained, total reach, save rate, send rate, and a one-line note on what worked.
Set it up so I can paste next week's numbers straight in as a new row and immediately see the trend across weeks, and tell me which two of those columns are the ones to actually watch so I do not drown in it.
Keep it clean and skimmable, and give it to me as plain text I can paste into Notion or a Google Doc without any formatting breaking.Save the output somewhere you will genuinely revisit, then re-run the audit prompt weekly so the table stays current. The compounding value is in the trend line across weeks, not in any single audit.
Take my highest-follow post and give me five fresh ways to remake the same core idea so it does not read as a repeat. Change the format, the angle or the opening scenario each time, and keep the underlying promise identical because that is the part that worked.
For each version, tell me what specifically changed from the original and who it is aimed at, for example someone who has seen the first one versus someone brand new to my account.
Then rank the five by how likely they are to reach people who do not already follow me, and explain the ranking in one line each.Look across my top-converting topics and find one adjacent topic I have never posted about that the same audience would very likely want, based on the pattern in my data rather than general advice about my niche.
Justify it by pointing at specific posts of mine and the pattern that connects them, and be honest if the data does not actually support a confident answer.
Then give me three post ideas inside that new topic, each one structured like whichever of my existing winners is the closest match.Find my post with the highest reach but the lowest follows, which means plenty of people saw it and almost nobody stayed. From what you can see on screen, tell me the most likely reasons it did not convert viewers into followers.
Go specific rather than general: look at the hook, whether the payoff arrived early enough, whether the topic was useful to a stranger or only to existing followers, and whether there was a reason to follow at the end.
Then rewrite the hook three different ways and tell me what you would change about the call to action, in the same voice as my posts that did convert.If your account is new or low volume, six months of data may be too thin for a real pattern and Claude will confidently describe noise. Give it enough posts first. Either route can only work with what Instagram actually shows, so anything it does not display for a given post is genuinely not available, and a good run is one where Claude writes "not shown" instead of filling the gap. If a big account makes the browser route feel slow or flaky, narrow the request to three months at a time and run it twice rather than asking for everything in one pass. And treat the output as a hypothesis you test on your next five posts, not a verdict. One re-audit two weeks later is worth more than the first report on its own.