Instagram competitor analysis with Claude: outliers and a content plan
Connect Claude to Manifold MCP, pull rival profiles and outlier posts, read comments and transcripts, then have Claude write a 2 week content plan.

How do I do Instagram competitor analysis?
Here is the whole job in Claude, from picking rivals to a 2 week content plan, run on Manifold MCP:
- Connect Manifold MCP to Claude (claude.ai, Claude Desktop or Claude Code), or to ChatGPT.
- Pick 3 to 5 rival accounts: ones you know, plus accounts that keep showing up under your niche hashtags.
- Pull each rival's profile: followers, bio, website and the accounts Instagram suggests beside it.
- List their recent posts and Reels, compute engagement rate, and flag posts at 2x or more the account's own median.
- Read the outliers' comments, captions and, for the top Reel, the spoken hook.
- Ask Claude for the content gap and your next 2 weeks of posts.
This page covers rival Instagram accounts, outlier posts and the plan. For a one-off look at what is posted under a hashtag, see research TikTok and Instagram. For sorting comments across many posts, see how to analyze TikTok and Instagram comments. For ads, pages, rankings and social accounts together, see how to track competitors. For the same job on YouTube channels, see YouTube competitor analysis. For the same job on TikTok accounts, see TikTok competitor analysis.
How to run Instagram competitor analysis in Claude
Each tool's parameters are on the Instagram tools page, and call costs on the credits page. To keep watching the same accounts afterwards, there is a monitor at the end of this guide (how monitors work).
What to compare
Instagram competitor research is not one follower number. Compare the same things on you and on each rival, and know which tool answers each. Stories, follower growth over time, reach, saves, shares, audience country split and a profile post count are not returned, so do not ask Claude to invent them.
| What | Tool | What you get |
|---|---|---|
| Followers and bio | instagram_get_profile | Followers, following, bio, verified, website, other bio links, a public email and phone where a business account publishes them, industry, and related accounts Instagram suggests beside it. posts_count is null: a signed-out visitor is shown 12 posts whatever the account has, so count posts from the listings. |
| Posting cadence | instagram_get_posts, instagram_get_reels | The exact created_at of each post, so Claude can count posts per weekday. Pinned posts sit at the top: sort by date and drop them from cadence. |
| Engagement rate per post | instagram_get_posts, instagram_get_reels | Likes and comments on each row. Rate is (likes + comments) / followers, an estimate. likes is null when the owner hides like counts: leave those rows out, they are not zero. Shares and saves come back empty. |
| Outlier posts | instagram_get_posts, instagram_get_reels | Claude works out the account's own median and flags 2x or more. Reel views come from instagram_get_reels and from search, not from the posts listing, where video rows leave views empty. |
| Format mix | instagram_get_posts, instagram_get_reels | media is video or image. A carousel reads as an image, so do not ask Claude to split carousels from single images. Use the Reels listing for video, the posts listing for image posts. |
| Captions and hashtags | instagram_get_posts, instagram_get_reels | Caption text, cut at 2,000 characters, and hashtags taken from the caption (without the #). |
| Comment themes | instagram_get_comments | Comments on one post or reel, newest first, with likes, reply count, and which comment a reply answers. Replies themselves come back only when you turn include_replies on. |
| Reel hooks | instagram_get_transcript, instagram_get_post | The full transcript, or NoData when the reel carries no speech. A reel over two minutes is refused as InvalidTarget; read its caption instead. The listing already has likes, comments and caption; instagram_get_post is for a URL you only have as a link. |
| Hashtag search | instagram_search_posts | Instagram's own ranking for one hashtag. A phrase is run together into one tag, so "gym wear" searches #gymwear. Dates can be filtered with since. There is no free-text search and no sort. |
Profile and listing tools take a handle, with or without the @. Comment, transcript and single-post tools take the full Instagram URL.
Step 1: connect Manifold MCP to Claude
Manifold is a hosted MCP server, listed in the official MCP registry as com.manifoldmcp/manifold. There is no Instagram login or Meta app of your own to set up. You pay per call, and every account gets 250 credits a month on the free plan. Full client steps are in the docs, and credit formulas on the credits page.
Claude.ai and Claude Desktop. From the Claude setup guide:
- Open Settings, then Connectors.
- Choose Add custom connector. Name it Manifold and paste the server URL from the setup guide.
- Choose Connect, sign in to Manifold, and approve Claude.
- Start a chat, check that Manifold is on in the chat's tools menu, and ask a question.
A connector you add on the web is also there in the desktop app and on your phone. The first time you sign in, Manifold creates your account on the free plan.
Claude Code. From the Claude Code setup guide: add the server with the URL from that guide. Start Claude Code, run /mcp, choose manifold, and finish sign-in in the browser. Add --scope user to have it in every project.
ChatGPT. Manifold MCP is an official ChatGPT app. From the ChatGPT setup guide: open Apps, search for Manifold, click Connect, sign in and click Allow. Then type @Manifold or pick it from the + menu under More. There is no URL to paste and no developer mode to turn on.
On Business, Enterprise and Edu, an admin decides which apps the workspace can use. The rest of this guide uses Claude, and the same Instagram tools work in ChatGPT.
Paste this to check the connection:
Which Manifold tools get an Instagram account, list an account's posts, list its Reels, get one post, get comments, get a Reel transcript, and search Instagram by hashtag? Say that instagram_get_profile returns followers but posts_count is null, that Reel views come from instagram_get_reels and search, that likes is null when the owner hides like counts, that media is video or image, that instagram_search_posts takes one hashtag, that instagram_get_transcript returns NoData when there is no speech, and that a Social account tracking monitor can watch those accounts on a schedule at 0 credits per check.
Claude should name instagram_get_profile, instagram_get_posts, instagram_get_reels, instagram_get_post, instagram_get_comments, instagram_get_transcript and instagram_search_posts.
Step 2: pick rivals from niche hashtags
Start with accounts you already lose to, then add anyone who keeps showing up under the hashtags your buyers use. instagram_search_posts is Instagram's own ranked sample for one hashtag, over the last month by default. Search a year back instead (since: "year"), so you see who stays in the results rather than who had one lucky week. There is no newest-first search and no free-text search: a phrase is run together into one tag.
Paste:
Search Instagram with instagram_search_posts for these hashtags, since year: "[niche]", "[niche]tips", "[niche]setup". One hashtag per call. Return a table: author (handle), url, likes, comments, views, media, created_at, text. Flag which handles appear more than once. Add these accounts I already know: [@handle1], [@handle2]. Keep 3 to 5 rivals. If likes is exactly 3, treat it as a hidden count and leave that row out of any rate. Do not invent handles or likes.
That is how to pick Instagram competitors you have not named yet: see who keeps showing up, and keep their handles for the next step. A one-off hashtag look without the rival teardown is the Instagram hashtag search tool.
instagram_get_profile also returns related accounts Instagram suggests beside a profile. After the first snapshot, add any related handle that is actually in your way.
Step 3: profile snapshot
This is the Instagram profile analyzer pass. instagram_get_profile returns one account's followers, following, bio, website, other bio links, verified, industry, related accounts, and a public email or phone where a business account publishes them. It has no post count and no follower history. The posts come in the next step.
Paste:
Call instagram_get_profile on these handles: [yours], [@rival1], [@rival2], [@rival3]. Return a comparison table: handle, name, followers, following, verified, industry, website, related. posts_count is null: do not invent a post count. If a field is null, say so. Save the table in this chat.
What to look at, not a score:
Followers vs recent engagement. Size the account by what its new posts get in the next step, not by followers alone. A large following with a low recent median means most followers no longer react.
Industry and bio. kind is company only when Instagram flags a business account; many brands read as a person. Judge a brand by industry and the bio.
Website and related. The website can be a link-in-bio page rather than the homepage. Related accounts are Instagram's own suggestions, a useful extra shortlist.
To analyze an Instagram account on its own, without rivals, the same profile call is the start of the Instagram engagement rate calculator.
Step 4: recent posts, Reels, engagement and outliers
instagram_get_posts lists one account's posts, newest first, with pinned posts on top. Each row has the date, caption, likes, comments, hashtags, media (video or image) and URL. Collab posts invited onto the account come back with another author: keep only rows whose author is the account before you take a median.
instagram_get_reels lists Reels only, with views and length. That is the Instagram Reels analytics pass: views are here, not on the posts listing.
Engagement rate per post is (likes + comments) / followers. It is an estimate, the same formula the Instagram engagement rate calculator uses. Do not mix it with a rate over views. For Reels, also keep median views from instagram_get_reels as a separate column.
An outlier here is a post at 2x or more the median engagement of that account's recent listings, or a Reel at 2x or more the median views. Use the median, not the mean: one hit drags an average away from what a new post will get. Leave out posts under 14 days old, whose numbers are still climbing, when you have enough older rows. Leave out rows where likes is null.
Paste:
Call instagram_get_posts and instagram_get_reels on [yours] and on [@rival1], [@rival2], [@rival3]. Page once if you need more rows. Sort by created_at, drop pinned rows from cadence and medians, and keep only rows whose author is the account. For each account: median engagement of posts 14 days or older as (likes + comments) / followers, median Reel views, posts per week, posts by weekday from created_at, format mix (video vs image), and every post at 2x or more the median, with url, likes, comments, views, media, created_at, and the caption's first line. Then a combined outlier table, sorted by how many times the median it did. Do not invent likes, views or dates. Leave out rows where likes is null.
Cadence is the weekday pattern in the dates, not a promise they will post that day next week. An account with no post in 90 days is paused: keep it as topic evidence, not as a posting model.
Step 5: hooks and captions
Captions and hashtags come from the listing. For the spoken open of a Reel you need instagram_get_transcript on the Reel URL. NoData means the reel carries no speech (music and text on screen), so take the hook from the caption and say so. Always call the transcript tool. A reel over two minutes is InvalidTarget at no charge: do not retry it. The Instagram transcript generator is the same call on one URL.
Paste:
Call instagram_get_transcript on the top 3 outlier Reel URLs from the last table. For each: url, duration_s, likes, comments, views, the first 30 seconds of the transcript (or NoData), the caption's first line, and the hashtags. Then a short list of hook and caption patterns the outliers share that our recent posts do not. Do not invent transcript lines. Do not claim a carousel when media is image.
Look at:
Caption formulas. Who leads with a question, a number, a vs, a how-to, a payoff in the first line.
Hashtags. Which tags the outliers actually use. Broad tags (#reels, #explore) are noise. Read real niche tags off the captions.
The first 30 seconds. A question, a result, a mistake, a promise. That is the hook you can copy the shape of, not the wording.
Step 6: comments on the outliers
instagram_get_comments returns the comments on one post or reel, newest first: the author, date, the text, likes, reply count, and which comment a reply answers. Replies themselves come back only with include_replies, which costs more (see credits). Leave it off unless the top comments show many replies and the thread matters.
Paste:
Call instagram_get_comments on the same outlier URLs. Page once per post. Leave include_replies off. Drop spam, giveaways, "comment X for the link", jokes, emoji-only comments and "love this". Keep is_creator comments apart. Return a themes table: requests, complaints, questions. For each theme: count, total likes, which posts, two verbatim quotes with the post URL. Do not invent comments.
That is enough to see what followers ask the rivals to post next. The deeper pass, with every comment tagged and themes weighted by likes, is how to analyze TikTok and Instagram comments with Claude.
Step 7: content gap and a 2 week plan
You now have four tables: the profile snapshot, recent posts with outliers, hook and caption patterns, and comment themes. Ask Claude to write the plan, citing the rows rather than inventing a like count.
Also search the hashtags you want to show up for, a year back, so the plan shows who already owns each tag rather than this week's spike.
Paste:
From the profile table, the outlier table, the hook and caption notes, and the comment themes, write a content gap and a 2 week content plan for [our account]. First call instagram_search_posts for these hashtags, since year: [tag 1], [tag 2], [tag 3]. Note who shows up (author, url, likes). Then: (1) topics rivals win that we have not posted; (2) comment requests nobody answered with a post; (3) 14 post ideas, each with a format (Reel or image post) from the outlier mix, a first-line caption that uses a pattern from the outliers, a weekday from the cadence table, the rival post or comment it answers, and the hashtag it should compete on. Cite the row each idea came from. Do not invent likes, comments or ranks. Do not claim stories, reach or follower growth.
Run this in your agent
Connect Manifold to Claude, ChatGPT or Cursor, then paste this prompt.
Run Instagram competitor analysis for [niche], rivals [@handle1], [@handle2], [@handle3] if I already know them. Search instagram_search_posts for "[niche]", "[niche]tips", "[niche]setup" with since: "year" (one hashtag per call; a phrase is run together into one tag). List accounts that keep showing up. Then call instagram_get_profile on each rival (handle with or without @). posts_count is null; do not invent a post count. Call instagram_get_posts and instagram_get_reels on each, page if needed. Sort by created_at, drop pinned rows from cadence and medians, keep only rows whose author is the account. Compute median engagement of posts 14 days or older as (likes + comments) / followers, median Reel views, flag posts at 2x or more the median, and cadence by weekday from created_at. Leave out rows where likes is null. media is video or image; a carousel reads as image. Call instagram_get_transcript on the top Reel outliers (NoData means no speech; a reel over two minutes is InvalidTarget). Call instagram_get_comments on those outliers. Call instagram_search_posts for our target hashtags and note who shows up. Return tables: profile snapshot (followers, following, verified, industry, website), recent posts and Reels with outliers, format mix, caption and hashtag patterns, comment themes, hashtag search. Then write a content gap and a 2 week content plan with post ideas, citing the rows. Do not invent followers, likes, views, dates or comments. Stories, follower growth, reach, saves and shares are not returned. Then preview a Social account tracking monitor for the rival Instagram handles, daily. Preview it first, then create it once I agree.
That is the Instagram competitor analysis tool in this setup: the same agent reads across accounts, comments and transcripts and writes the plan. It does not publish the posts. You do.
Worked example (illustrative)
The names, handles, followers, dates and comments below are an example, not measurements of a real account.
Acme Desk posts standing-desk setup Reels as @acmedesk. After connecting Claude, Acme pastes the search prompt for #standingdesk, #standingdesktips and #standingdesksetup. A year of hashtag results keeps showing @rivaldesk, @otherframe and @thirdtop, plus @acmedesk on a few rows. Acme keeps those three.
Acme then pastes the profile snapshot prompt. Example profile table (illustrative):
| Handle | Followers | Following | Verified | Industry |
|---|---|---|---|---|
| @rivaldesk | 186,000 | 420 | yes | Furniture |
| @otherframe | 74,000 | 190 | no | Home decor |
| @thirdtop | 31,000 | 880 | no | (null) |
| @acmedesk | 9,400 | 210 | no | Furniture |
On @rivaldesk, the median engagement of posts 14 days or older is 0.42%. Three outliers clear 2x: a Reel "Cable management in 12 minutes" (1.1% engagement, 94,000 views), a Reel "I tried 7 standing desk mats" (0.91%, 61,000 views) and an image post "Drawer desk tour" (0.88%). Rival posts Reels on Tuesdays and Fridays. @otherframe has a median of 0.61%, posts image posts on Mondays, and has one outlier, a Reel "Budget standing desk setup". @thirdtop has a median of 0.80%, no fixed cadence, and one outlier, a Reel "72-inch cable lengths".
The first 30 seconds of Rival's cable-management Reel ask "Why does every standing desk look messy after a week?", then promise a 12-minute layout. Its comments ask for a dual-monitor arm version, complain that the tray in the Reel is out of stock, and ask about cable length for a 72-inch top. Acme has no cable-management Reel, and a search for #standingdesktips ranks Rival often and Acme nowhere.
Example Claude table (illustrative):
| Idea | Format | Why | Evidence |
|---|---|---|---|
| Cable management for dual monitors | Reel, about 12 minutes of talk | Rival's outlier; comments ask for dual-monitor arms | Outlier table; comment theme |
| Budget standing desk mat | Reel, "I tried N" | Other's mat outlier; Acme has no mat post | Outlier table |
| Drawer desk cable routing | Image post, tour | Rival's image outlier; comment request on trays | Comment theme; Rival image post |
| 72-inch cable lengths | Reel, how-to | Third's outlier; comment questions on cable length | Comment quotes; Third Reel |
| Standing desk vs treadmill desk | Reel, vs | Rival caption pattern; Acme has neither side | Caption notes; gap list |
None of those rows are live measurements. They show the shape of a digest Claude can write from the four tables: a gap, a format, a first line that copies a pattern, a weekday, and the row it came from.
Keep it current: a social account monitor
The snapshot is on demand. To watch those rival Instagram accounts on a schedule, create a Social account tracking monitor. Product page: social media monitoring. The full setup, take-off rule and worked examples are in how to monitor competitor social media.
The monitor reads public accounts on LinkedIn company pages, Instagram, TikTok, YouTube and Facebook pages, 1 to 10 accounts per monitor. For Instagram, pass the handle.
Each check lists the latest posts. A new post is reported once, when it first shows. It is reported again if it takes off: at least 3 times the median views of that account's posts in the same check, and at least 1,000 views. The check needs 5 or more posts with views before it can take a median. That is stricter than the 2x you used above. Instagram post listings often leave views empty, so new posts are the reliable signal; for Reel views, rerun instagram_get_reels on demand.
Creating the monitor and its checks cost 0 credits on every plan, including Free. The free plan keeps 1 monitor. Monitors a workspace can keep, by plan: 1 on Free, 5 on Pro, 25 on Growth, 100 on Agency.
Checks run hourly by default. Daily or weekly is enough for rival Instagram accounts, unless they post in bursts you need the same day. Alerts start on: email on every plan (at most one a day per monitor), Slack and webhooks on Growth and Agency. The first check runs within five minutes and only records a starting point, so changes appear from the second check on. Checks skip the cache, so they read live data.
Paste, then let it preview:
Create a Social account tracking monitor named "[niche] rival Instagram" for these Instagram handles, daily: [@rival1], [@rival2], [@rival3]. Description: "New rival posts, so we add outliers to the comparison table and refresh the 2 week plan." Preview it first, then create it once I agree.
On Free, put the rival Instagram handles in the one monitor you keep. You can mix platforms in that same monitor, up to 10 accounts. After the second check, ask Claude to read monitoring_get_changes and to pull the transcript and comments on any take-off worth a full pass.
Instagram competitor analysis tools vs this setup
Instagram's Competitive Insights lives in the Professional Dashboard. It compares you with up to 10 public accounts on follower growth, posts, Reels, stories and frequency, and shows top content from those accounts. That is the right view when you already work in Instagram and want a side-by-side of posting volume and follower change.
Social suites such as Sprout Social and Hootsuite are Instagram competitor analysis tools in the usual sense: dashboards, historical charts, scheduled publishing and client reports, priced per seat. That is the right software when you need a year of follower graphs, a shared calendar, or a report you send to a client.
This setup is a different job. You already work in Claude or ChatGPT. You pass handles, the agent calls the Instagram tools on demand, and the same agent writes a 2 week plan from the tables. You pay per call, with no seat in a social suite. There is no story view, no follower history chart and no browser overlay on instagram.com. History is the table you saved, plus the monitor's new posts and take-offs.
Use this setup when you want the AI to read across accounts, comments and transcripts and write the plan, without a suite login. Use Competitive Insights when you want Instagram's own side-by-side of follower growth and posting frequency. Use a dedicated suite when you need historical charts, scheduling, or a client-ready report.
So the honest answer to "what is the best Instagram account analyzer" depends on that split. For in-app follower growth and posting frequency, Instagram's own dashboard is the tool. For three rivals, comment themes and a pasteable 2 week plan from inside Claude, it is this guide.
Run your first Instagram competitor analysis
Run your first Instagram competitor analysis on the free plan, with 250 credits a month. Connect Claude to the Instagram MCP tools and paste the prompt above with your niche and handles. For a one-off hashtag search without the rival teardown, see research TikTok and Instagram. For scheduled posts from those accounts, see social media monitoring. To turn the same accounts into a creator shortlist, see how to find Instagram influencers.



