LinkedIn Content Audit

Audit LinkedIn creator profiles by scraping posts (Apify) and classifying them across 6 dimensions: hook patterns, post types, media types, media content, CTAs, engagement. Diagnostic only — does not write posts or check algo compliance. Over-fetches ~100 posts/profile (6-month window), deep-analyzes the top 25 by engagement.

When to run

Run when: auditing a creator's LinkedIn, benchmarking 2+ profiles, extracting hook/CTA patterns, or sourcing competitive intel for content strategy. Skip when writing posts (linkedin-content), checking a single post's algo fit (linkedin-algo-audit), building an ICP guide (linkedin-content-guide), or running a multi-channel audit (content-audit). Full trigger/anti-trigger list → the premium reference.

Inputs

Required: 1-10 LinkedIn profile URLs (linkedin.com/in/... format).

Optional (defaults): scrape depth 100, time window 6months (alt: month/3months/year), top-N for analysis 25, include quote posts true, focus areas all 6, client context none. Validate URLs before scraping; confirm parameters with user. Full input table → the premium reference.

Steps

1. Validate inputs — All URLs are linkedin.com/in/...; 1-10 profiles; confirm depth + window with user.
2. Phase 1.1 — Apify scrape (single batch). Call mcp__apify__call-actor with actor harvestapi/linkedin-profile-posts, input: targetUrls: [all], maxPosts: 100, postedLimit: "6months", includeReposts: false, includeQuotePosts: true, scrapeReactions: false, scrapeComments: false. One run = one start fee + per-post cost (cheaper than N runs). Reposts excluded so patterns reflect creator's own voice.
3. Phase 1.2 — Retrieve + validate. Use mcp__apify__get-actor-output with returned datasetId. Confirm 50-100 posts/profile, text + engagement + timestamps present. Flag profiles with <10 posts as "insufficient data." Fallback if Apify fails: ask user for manual posts (copy-paste, Shield/Taplio/AuthoredUp CSV, or screenshots).
4. Phase 2.1 — Group by author (URL or name).
5. Phase 2.2-2.4 — Engagement filter. Compute total_engagement = likes + comments + shares per post. Sort each profile descending. Take top 25 for deep analysis. Retain full dataset for volume/cadence metrics (total posts, posts/week, consistency stdev).
6. Phase 3.1 — Hook classification. Read first line / first sentence. Map to 14-category taxonomy in the premium reference. Output: count, %, avg engagement per hook type.
7. Phase 3.2 — Post type (pillars). Map to Educational / Personal / Promotional / Organizational / Engagement. Compare mix to 40/25/25/10 target. Definitions → the premium reference.
8. Phase 3.3 — Media type. Classify from Apify attachment data: text-only, carousel/document, single image, multi-image, video, poll, article/newsletter, external link. Output: format mix + avg engagement per format.
9. Phase 3.4 — Media content. For posts with visuals, classify what media depicts using post-text context only (screenshots, charts, selfies, memes, infographics, text-on-image, BTS, professional photo, AI-generated, undetermined). NEVER guess from URL.
10. Phase 3.5 — CTA classification. Analyze final 1-3 lines. Bucket into 8 types: comment prompt, DM invite, link/resource, follow/connect, save, repost, no CTA, multiple CTAs. Output: distribution + avg engagement per type. Pattern examples → the premium reference.
11. Phase 3.6 — Engagement analysis. Per-post likes/comments/shares; aggregated avg/median/max; top 5 with excerpts; cross-tabs (engagement × hook, × media, × pillar, × CTA); volume metrics from full dataset.
12. Phase 3 checkpoint — All 6 dimensions classified; percentages sum to 100% within each category; cross-tabs computed; volume metrics from full dataset.
13. Phase 4 — Cross-profile comparison (only if 2+ profiles). Build profile overview matrix; format mix table; hook style heat map; engagement benchmarks; CTA distribution; top patterns to emulate (evidence-cited); anti-patterns to avoid (evidence-cited). Detail → the premium reference.
14. Self-evaluation. Completeness (all phases + profiles + comparison if applicable); accuracy (3 random hook spot-checks, engagement avg sanity, format consistency vs attachments); honesty (zero invented numbers, no guessed media content, gaps marked). Full protocol → the premium reference.
15. Write outputs. Per-profile: linkedin-audit-{username}.md. Cross-profile: linkedin-audit-comparison.md. Templates → the premium reference.

What good looks like

Examples: None on file — first run will seed examples/.

Evaluations: Quality gate passes when (a) data: every profile has 10+ posts, engagement + timestamps present; (b) analysis: all 6 dimensions classified, percentages sum to 100%, cross-tabs use consistent metrics; (c) output: per-profile reports stand alone, comparison includes actionable takeaways with evidence citations, all findings traceable to specific posts.

Anti-hallucination (load-bearing): Only analyze Apify-returned data. Never invent engagement counts, post text, or follower counts. Mark <10 posts as insufficient data. Source every finding (Apify, harvestapi/linkedin-profile-posts, YYYY-MM-DD). When hook/CTA classification is ambiguous, mark as "ambiguous" and note the two closest matches. Use precise language ("23 of 25 posts (92%)" not "almost all"). Full guardrails → the premium reference.