Use Case
Competitor Analysis in Claude: Compare Tracked Accounts, Posts, Comments and Metrics
Run a social competitor analysis in Claude from accounts your Canvas tracks: saved posts, comments and metrics, with quoted runs for fresh data.
By Bell Chen, founder of Superdirector — how the analysis works.
Reference visuals
Frames and thumbnails behind this guide
A competitor analysis is only as good as its sample. When I review one by hand I want the same accounts, read over the same window, with the posts, the comments under them and the metrics laid side by side. When I ask an AI to do it, I want exactly the same thing, plus a citation for every claim so I can check the ones that surprise me.
Claude can do the comparison well once it can reach the evidence. Through the Model Context Protocol (modelcontextprotocol.io) it can read a Canvas where your competitors are tracked, instead of guessing at them. The method below is the one I would run, written so it holds up whether or not you use our product.
What a Claude competitor review can rest on
The raw material is the competitor accounts tracked on the Canvas, alongside your own account and any inspiration accounts. Each completed read saved that account’s posts with captions and metrics, its account-level metrics, and the comments under its posts when comment collection was part of the read. Every record is dated.
Claude reads those records through native tools and can put them side by side: what each competitor posted over the window, which posts drew comments, what the comments asked.
Two limits shape the method. Claude does not see posts that were never collected, so coverage gaps must be named, not filled in. And it does not watch videos from saved posts; a video analysis report, run per video, is the only way to compare edits and spoken content.
A competitor review in Claude, step by step
| Stage | When / duration | Tools | Deliverable |
|---|---|---|---|
| 1. Choose the set and the window | 15 minutes | the web Canvas or Claude | three to five tracked competitors and a stated time window |
| 2. Check coverage before analysis | one prompt | knowledge_inventory | each competitor with its last read date and retained post count |
| 3. Refresh only what is stale | minutes to approve | quote_module_run, run_module, get_module_run | committed reads inside the window |
| 4. Compare posts, metrics and comments | 20 to 30 minutes | read_source_evidence | a comparison table plus comment themes per competitor |
| 5. Compare covers on the Canvas | 10 minutes | open_canvas | a note on visual patterns by account |
| 6. Save the review and brief the gap | 15 minutes | save_artifact | a saved review and one brief in Canvas Files |
Choose the set and the window
Pick direct competitors, the accounts a customer would actually compare you with, and write the window down: the last 30 days, the last quarter. A review without a window drifts into general impressions.
Check coverage before analysis
Ask Claude to list every competitor with the date of its last completed read and how many posts it retained. Any account read before your window starts needs a fresh read before it can be compared fairly.
Refresh only what is stale
For each stale account, Claude asks for a quote. Look at the total before approving; then approve the runs one by one. Claude keeps each run ID and reads the result when it commits, reporting anything it could not collect.
Compare posts, metrics and comments
Ask Claude for a table: posts in the window, format mix, median metrics, and the posts that drew the most comments for each account. Then have it read those comment threads and list what viewers asked for, with quotes. The unanswered requests across competitors are usually the most useful line in the review.
Compare covers on the Canvas
Ask Claude for the Canvas link and look at the covers of each competitor’s most-commented posts there: framing, on-screen text and subject. The connection returns no images, so add what you see to the review yourself. Note where every competitor looks the same; that sameness is an opening.
Save the review and brief the gap
Ask Claude to save the review to the Canvas, then draft one brief for the clearest gap it found. Next month, a fresh read and the saved review make the comparison over time straightforward.
What a good review contains
Every finding names the account, the post and the read date. Coverage gaps are listed rather than smoothed over. Comment themes come with quoted examples. Metrics are compared within the stated window. And the review ends in one concrete brief, not a list of observations nobody acts on. If a claim depends on what happens inside a video, it is marked as needing a video analysis report.
Common mistakes
Comparing reads from different months. A competitor read in August and another read yesterday are not comparable on volume. Refresh the stale one or drop it from the comparison.
Ranking on likes alone. Comments carry the intent. A post with modest likes and a long thread of questions often tells you more about what to make than the most-liked post does.
Letting the AI fill gaps. If comments were not collected for a post, the honest answer is "not collected." Ask Claude to say so explicitly.
Where Superdirector fits
Superdirector is the Canvas Claude reads in this method. Connecting, account status and Canvas inventory are free. Paid plans start at $29 per month; reads and analysis are metered in credits with a quote and approval before each run. A verified account gets a one-time 20-credit MCP starter in the AI app, at most one account check and one X text draft. The MCP starter is not a trial of the plan; Starter has its own 7-day free trial. Your AI cannot post; posting is approved in the Canvas.
For the same review without an AI app, see (Turn Competitor Content Analysis into a Production Brief). For the connection itself on Instagram, see (Instagram MCP Server: Your Tracked Instagram Data Inside Claude, ChatGPT and Codex).
Disclosure: I build Superdirector. Tool names and limits are those the MCP server advertises today and can change; Claude reads the current capabilities before it promises an action. Claude is a product of Anthropic, which is not affiliated with this page.