competitor-watch

Competitor Watch — Weekly Competitive Intelligence

A self-contained weekly competitive scan that any Motion customer can run. Tracks competitor ad strategies, messaging shifts, creative patterns, and new launches — then compares against last week's baseline to surface what actually changed.

No external context required. This skill uses only the Motion MCP and stores its own baselines. No repo, no strategy docs, no internal knowledge base needed.

Core principle: The value is in the delta, not the state. "Foreplay launched 6 new Lens-focused ads this week" is intelligence. "Foreplay has 41 active ads" is a fact sheet. Every finding answers: "What should I pay attention to this week?"

Baseline model: First run establishes a baseline. Every subsequent run compares current state to the previous week's baseline, reports what changed, and updates the baseline. Baselines are stored locally as markdown files.

---

Phase 1: Setup

1a. Parse Arguments

• --competitor: Optional. A domain (e.g., foreplay.co) to scan a single competitor. If omitted, scan all saved competitors.
• --baseline-only: Establish baselines without producing a delta report. Use for first run.

1b. Resolve Workspace

Call get_auth_context().

If multiple workspaces exist, ask the user which one to scan. Store the workspaceId for all subsequent calls.

1c. Load or Create Competitor Watchlist

Check for a saved watchlist at ~/.claude/competitor-watch/watchlist.md.

If watchlist exists: Read it. It contains brand names, domains, and resolved brandIds from a previous session. Use these directly — no need to ask again.

If no watchlist exists (first run): Ask the user to set up their competitor watchlist:

"This is your first competitor watch. Who are the 3-5 brands you want to track? Give me their names or website domains and I'll set everything up."

Use AskUserQuestion to collect this. Accept brand names, domains, or both.

Then resolve each brand:

1. Try get_brand_by_domain(brandUrl) if a domain was provided
2. If that fails or a name was given, try search_brands(query) with the brand name
3. If multiple results, show the matches and ask which one
4. If no match, tell the user the brand wasn't found in Motion's ad library and skip it

Once all brands are resolved, save the watchlist to ~/.claude/competitor-watch/watchlist.md:

---
created: [YYYY-MM-DD]
updated: [YYYY-MM-DD]
workspace_id: [workspaceId]
---

# Competitor Watchlist

| Brand | Domain | Brand ID | Added |
|-------|--------|----------|-------|
| [Name] | [domain] | [brandId] | [YYYY-MM-DD] |
| [Name] | [domain] | [brandId] | [YYYY-MM-DD] |

Updating the watchlist: If the user says "add [brand]" or "remove [brand]" or "change my competitors", update the watchlist file accordingly. The watchlist is the persistent config — it survives across sessions.

1d. Get Own Brand Context

get_workspace_brand(workspaceId)

Used for comparison framing — "how does this compare to our own ad strategy?"

1e. Load Previous Baselines

Look for baseline files at ~/.claude/competitor-watch/{competitor-slug}.md for each brand on the watchlist. If no baseline exists for a brand, this is a first run for that brand — establish the baseline, don't compute deltas.

---

Phase 2: Ad Intelligence — Full Active Portfolio + Recently Killed

For each competitor, dispatch in parallel:

2a. Pull Ad Data (Queries per Competitor)

CRITICAL: Always set limit=500 on every get_inspo_creatives call. The API defaults to ~20-50 results if you omit the limit or set it low. Setting limit=500 ensures you get the full portfolio in one call for most brands. This is non-negotiable — partial data produces wrong analysis.

MANDATORY: Two-pass pull for every brand. Even with limit=500, always pull NEWEST and OLDEST and merge. This catches edge cases where the API caps results.

For each competitor, run these queries in parallel:

Query 1a: Active Portfolio — NEWEST first

get_inspo_creatives(brandId, workspaceId, sort="NEWEST", status="ACTIVE", limit=500)

Query 1b: Active Portfolio — OLDEST first (same call, different sort)

get_inspo_creatives(brandId, workspaceId, sort="OLDEST", status="ACTIVE", limit=500)

Query 2: Recently Killed

get_inspo_creatives(brandId, workspaceId, sort="NEWEST", status="INACTIVE", limit=500)

Filter client-side to only those with pauseDate within the last 7 days.

Query 3: Brand Context

get_inspo_brand_context(brandId, workspaceId)

After pulling, merge and deduplicate Query 1a + 1b by creative _id. The merged set is the full active portfolio. Log the count.

Validation gate: Before proceeding to analysis, verify the ad count matches expectations. If a brand shows activeAdCount in its metadata (from workspace competitors or search_brands), compare your pulled count against it. If your count is significantly lower, you missed ads — paginate further or increase the limit.

NEVER set limit below 500. NEVER omit the limit parameter (it defaults too low). NEVER skip the OLDEST pass.

2b. Full Active Portfolio Analysis

Analyze the complete active portfolio as a body of work — this reveals the competitor's live-in-market strategy:

Format & Production:

• Video / static / carousel split
• Production styles (UGC, screen recording, studio, talking head, motion graphics)
• Production quality spectrum

Messaging Themes:

• What hooks appear most frequently? Group by type: problem-agitation, testimonial, question, bold claim, social proof, demo walkthrough
• What value propositions are being pushed? Rank by frequency (= conviction level)
• What emotional registers dominate? (fear, aspiration, frustration, curiosity, urgency)
• Quote actual hooks verbatim — don't summarize

Feature & Product Emphasis:

• What product features are mentioned across active ads? Rank by frequency.
• Any features being promoted that your brand doesn't have? (competitive gap signal)
• Any features you have that they're NOT promoting? (potential differentiation)

Landing Page Strategy:

• Where are ads driving traffic? Group by destination.
• Which products/features get the most ad-driven traffic?

2c. Survival Analysis — Weekly Cohorts

Cohort every ad by age and compute survival:

Long-runners (60+ days, still ACTIVE) These are evergreen winners. They've survived multiple creative refresh cycles. Pay attention to what makes them durable.

Survivors (launched 15-60 days ago, still ACTIVE) Proven creative. If it's been running 2+ weeks, they've seen enough data to know it works. This is the strongest signal of what's actually performing.

• What do survivors have in common? (format, hook, messaging)
• Survival rate for this cohort

Testing Zone (launched 7-14 days ago, still ACTIVE) Past the first cut but could still be in evaluation. Moderate signal.

• Which are iterations on survivors? Which are new directions?
• Any killed from this cohort? (fast kills = strong negative signal)

Fresh Tests (launched < 7 days ago) Newest bets. Reveal strategic intent — what they're thinking about right now.

• New messaging themes being tested?
• Events, promos, product launches?
• Iterations on existing winners or entirely new directions?

Important: "Inactive" does not mean "failed." An ad can stop running because:

• It underperformed (the obvious read, but not the only one)
• It was time-bound — a promo, event, or seasonal campaign with a planned end date
• Budget shifted to a hotter creative
• It was replaced by a new iteration
• It completed its audience/impression goal

Always reason over the likely reason. Look at content clues (mentions a date/event/sale?), runtime (2+ weeks before killing = probably adequate), and siblings (newer ad with similar messaging = iteration replacement). Quick kills (< 5 days, no promo content) are more likely genuine underperformance.

2d. Deep-Dive Prioritization

You can't pull summaries and transcripts for every ad. Use this priority:

Always deep-dive (transcript + full analysis) — 5-8 per competitor:

• Long-running survivors (15+ days active) — what messaging is actually working?
• Ads representing NEW messaging directions not seen in last week's baseline
• Ads mentioning or positioning against your brand or other competitors

Summary only (read copy + metadata):

• 7-14 day survivors, ads in clusters where another ad already got a deep-dive

Metadata only (format, CTA, landing page, launch date):

• Fresh tests (< 7 days), clear duplicates, minor copy variations

For deep-dives, dispatch Agent sub-agents in parallel:

For each ad: pull get_creative_transcript, extract opening hook (verbatim),
core claim, proof structure, emotional register, CTA, competitive positioning,
feature depth. Return a structured brief card.

2e. Strategic Clustering

Group ads into clusters that reveal the creative strategy:

• Test clusters: Ads with same messaging angle but different hooks, formats, or creators. Name the bet: "They're testing whether UGC outperforms screen recordings for their analytics feature."
• Promotional intel: Events, free trials, discounts, webinars, partnerships, product launches?
• Sub-product detection: If a competitor's ads drive to multiple domains or products (e.g., superside.com + superads.ai), analyze each product's messaging separately. A company running two businesses through one ad account is a different signal than a single-product company.

2f. Competitive Evaluation

For the full portfolio, answer: If a prospect evaluating your brand saw these ads, how would it make them think about you?

• What claims are they making that you can't match today?
• What claims are they making that you do better but aren't advertising?
• What audience overlap exists?
• What social proof might sway a prospect?
• Is their creative quality higher, lower, or comparable to yours?
• What would your sales team need to know to handle this in a competitive deal?

---

Phase 3: Own-Brand Self-Benchmark

Run the same analysis on your own brand's ads. This creates the comparison frame.

3a. Pull Own-Brand Data

Use the brandId from get_workspace_brand. Pull the EXACT same way as competitors — limit=500, two-pass NEWEST+OLDEST, merge and deduplicate:

get_inspo_creatives(brandId, workspaceId, sort="NEWEST", status="ACTIVE", limit=500)
get_inspo_creatives(brandId, workspaceId, sort="OLDEST", status="ACTIVE", limit=500)
get_inspo_creatives(brandId, workspaceId, sort="NEWEST", status="INACTIVE", limit=500)

Merge active pulls by _id. Filter inactive to pauseDate within last 7 days. Do NOT use a lower limit for your own brand than competitors.

3b. Analyze the Same Way

Apply the same framework: format distribution, messaging themes, survival analysis, test clusters.

3c. Comparison Output

For every competitor finding, provide the own-brand comparison:

• "They run 73% video vs your 45%"
• "Their oldest survivor is 27 days vs your 12"
• "They lead with AI agent messaging; you lead with analytics depth"

Don't just compare numbers — interpret what the gap means. A format gap might be intentional or a blind spot worth testing.

---

Phase 4: Delta Computation (vs Last Week's Baseline)

If a baseline exists from the previous scan, compute what changed:

Volume & Velocity:

• Total active ads now vs last week (net change)
• New launches since last scan
• Ads that went inactive since last scan
• Weekly launch cadence trend

Strategic Shifts:

• New messaging territories they weren't running last week
• Format mix changes
• Features promoted that weren't in last week's baseline
• Tone or positioning shifts in brand context

What They Stopped Doing:

• Messaging angles from last week's baseline that are gone
• Formats abandoned
• Claims stopped

What Survived:

• Ads that were active last week AND still active this week = confirmed winners
• Anything that was in the "Testing Zone" last week and survived to "Proven" this week

Signal Strength:

Signal	Strength	Why
A survivor from 15+ days with a new messaging angle	Very High	Proven message they're scaling
3+ fresh tests all exploring the same new theme	High	Coordinated strategic bet
A feature promotion for something you don't have	High	Competitive gap
Format mix shift (e.g., UGC ratio doubled)	Medium	Production strategy change
Volume spike or drop vs last week	Medium	Resource allocation signal
Single ad with an unusual approach	Low-Medium	Could be a one-off test
Minor copy variations on existing themes	Low	Skip unless pattern emerges

---

Phase 5: Output

Report Structure

# Competitor Watch — [Date]
**Workspace:** [name] | **Competitors:** [N scanned] | **Baseline:** [first run / delta from YYYY-MM-DD]

---

## What Changed This Week
[3-5 bullets MAX. Only findings that matter. Each: what shifted + why it matters + what to consider.
If nothing significant: "Stable week — no major competitive shifts detected."]

---

## Competitor Intel

### [Competitor Name] — [one-line: quiet / active / shifting / launching]

**Active Portfolio:** [N] total | Video [X%] / Static [Y%]
**30-Day Launches:** [N] launched | [N] survived | [N] killed | [X%] survival rate
**Weekly Cadence:** [accelerating / steady / slowing / paused]

**Live Strategy:**
- [Top messaging themes, ranked by frequency]
- [Dominant hooks and emotional registers — quote the actual hooks]
- [Format & production strategy]
- [Features being promoted]

**Survivors (15+ days, still running):**
- [What they have in common — the pattern that's working]

**Fresh Tests (this week):**
- [New launches — what they're testing, what it signals]

**Test Clusters:**
- [Cluster 1: name the bet, N ads, what they're varying]

**Competitive Eval:**
- [If a prospect saw these ads, how would it affect your deals?]

**Delta from Last Week:**
- [What's new, what's gone, what shifted. Skip if first run.]

[Repeat per competitor. Skip sections with no content.]

---

## Your Brand Benchmark

**Your Active Portfolio:** [N] total | Video [X%] / Static [Y%]
**30-Day Survival Rate:** [X%]

**How You Compare:**

| Dimension | You | [Comp 1] | [Comp 2] | [Comp 3] | Read |
|-----------|-----|----------|----------|----------|------|
| Active ads | [N] | [N] | [N] | [N] | [who's investing more?] |
| Oldest active ad | [days] | [days] | [days] | [days] | [portfolio maturity] |
| Video/static mix | [%] | [%] | [%] | [%] | [format strategy] |
| Visual format diversity | [N types] | [N] | [N] | [N] | [creative range] |
| Customer proof in ads | [Y/N] | [Y/N] | [Y/N] | [Y/N] | [credibility gap?] |
| Biggest bet | [what] | [what] | [what] | [what] | [strategic direction] |
| Key feature promoted | [what] | [what] | [what] | [what] | [positioning overlap] |
| Comment-gate CTAs | [Y/N] | [Y/N] | [Y/N] | [Y/N] | [engagement tactic] |

**Gaps worth testing:**
- [Things competitors do that you don't — worth a test?]
- [Things you do that nobody else does — double down?]

---

## Recommended Actions
[2-3 specific, grounded-in-findings recommendations.
Each connects to a specific finding above.
Not "diversify your creative mix" — instead: "Test a UGC creator ad using
the solopreneur burnout angle that's been running 27 days for [competitor]."]

Output Rules

• Lead with "What Changed." If someone reads only the first section, they get the important stuff.
• Delta-first on repeat runs. Don't report state — report change. "No changes" = skip or one line.
• Assume the reader knows the competitive landscape. Don't explain what Foreplay or Superads are. Don't summarize product categories. The team follows this space — they want the specific tactical insight they'd miss without pulling the data. Prioritize: survival signals (what's been running 60+ days and why), test cluster patterns (what they're A/B testing and what it implies), format/production shifts, and messaging angles that directly compete with Motion's positioning.
• Quote actual hooks. Don't summarize when you can quote.
• Include specific numbers everywhere. "27 days active", "6 variants launched same day", "72% image vs 28% video", "109 active ads". Specificity = credibility. Vague = skimmable.
• Carry the competitor data caveat. You're reading bets, not results. Frame as "they're testing this" not "this is working for them." The only proven patterns come from your own performance data.
• Recommendations must be specific. "Test a confessional UGC ad where a creator admits a frustration with analytics tools before showing your product" not "consider UGC."
• Length: 5-minute read max. If the market is quiet, make it 2 minutes.

---

Phase 6: Update Baseline

After producing the report, write the new baseline for each scanned competitor.

Baseline File Format

Write to ~/.claude/competitor-watch/{competitor-slug}.md:

---
competitor: [Name]
brand_id: [brandId]
last_scanned: [YYYY-MM-DD]
---

# [Competitor Name] — Weekly Baseline

## Portfolio Snapshot
**Active Ad Count:** [N]
**Format Mix:** [video X% / static Y%]
**Production Styles:** [UGC, screen recording, etc.]

## Active Ad Inventory
[For each active ad: ID, format, hook (verbatim), CTA, landing page, launch date, days active]

## Messaging Themes
- [theme 1 — frequency, example hook]
- [theme 2 — frequency, example hook]

## Survival Cohorts
**60+ day long-runners:** [list with IDs]
**15-60 day survivors:** [list with IDs]
**7-14 day zone:** [list]
**< 7 day fresh tests:** [list]

## Test Clusters
- [Cluster name: ad IDs, what they're varying]

## Features Promoted
- [ranked list]

## Brand Context Summary
[positioning, voice, audience from get_inspo_brand_context if available]

Baseline Rules

• Overwrite the previous baseline on each scan — this is a weekly snapshot, not an append log
• Store the full active ad inventory (IDs + hooks) so next week's scan can diff: "these 3 ads are new, these 2 are gone"
• If a section couldn't be populated (e.g., brand context unavailable), note the gap
• Baseline files live at ~/.claude/competitor-watch/ — outside any repo, purely local

---

Error Handling

• No watchlist and no --competitor flag: Trigger first-run onboarding — ask for 3-5 brands to track
• Brand not found: Suggest checking the domain or trying a different search term
• Brand context unavailable: Note it, proceed with ad data only
• Few/no ads returned: Note thin data, don't over-interpret. Suggest expanding to LAST_90_DAYS
• No previous baseline: Treat as first run, produce a full landscape report, label as "Baseline Established"
• Transcript quality poor: Some transcripts return garbage (single words, unrelated text). If a transcript doesn't make sense, skip it and note the gap. Don't quote bad transcripts.

---

Phase 7: Slack Distribution + Schedule Prompt

After producing the report and updating baselines, deliver it to the team and prompt for recurring scheduling.

7a. Build the Slack Canvas

Create a Slack canvas containing the full competitor watch report using slack_create_canvas:

• Title: Competitor Watch — [YYYY-MM-DD]
• Content: The entire Phase 5 report output, formatted as Canvas-flavored Markdown

7b. Write the Slack Message

Craft a short, high-signal Slack message that makes people want to open the canvas. This is NOT a summary — it's a teaser.

Structure:

🔍 *Competitor Watch — [Date]*

[One punchy headline: the single most surprising or actionable finding. Make it specific and provocative enough that someone stops scrolling.]

*Top signals this week:*
• [Insight 1 — the sharpest competitive move. Quote a specific hook or claim. Name the competitor.]
• [Insight 2 — something that directly affects your team's next creative decisions]

Full report → [canvas will auto-attach]

Rules for the message:

• Max 4-5 lines. This is a hook, not the report.
• Assume the team already knows what competitors exist and roughly what they do. "Superside launched Superads" is not news to this team. They follow the space. What IS news: a specific tactical move, a messaging shift, a survival signal, or a pattern nobody would notice without pulling the data.
• Lead with the deepest, most specific insight — the thing someone would only know if they'd pulled 100+ ads and analyzed survival cohorts. Examples of GOOD insights:
  • "Foreplay's vertical hook 'Supplement brands, if you run ads on Facebook...' just hit 27 days active — the only industry-specific ad surviving past the testing zone in this entire category"
  • "Atria launched 6 identical-copy Raya ads on the same day with different video — they're not testing messaging, they're testing which creator converts on a single script"
  • "Superside's UGC creators are ONLY used for Superads ads — their enterprise brand runs zero UGC across 94 ads, which means they see UGC as a SMB tactic"
  • "Motion's 3 oldest active ads (136 days) are all static image content marketing — the same format Superside's longest survivors use. Video dominates new launches but static survives longer."
• BAD insights (too obvious, skip these): "Competitor X launched a new product", "Everyone is using video", "AI is a trend in the category", "[Brand] is running ads"
• Quote actual hooks and include specific numbers — days active, ad counts, survival rates. "Foreplay's founder story has run 70 days across 3 variants" is 10x more compelling than "Foreplay uses founder stories."
• Create an information gap — hint at pattern-level findings in the canvas without fully resolving them. Make the reader think "wait, really?" and click through.
• No emojis beyond the 🔍 opener. Keep it clean and professional.
• No "here's what we found" preamble. Jump straight into the insight.

7c. Send to Slack

1. Read the Slack channel from settings (slack_channel_id). If not configured, ask which channel to post to.
2. Create the canvas with slack_create_canvas
3. Post the teaser message to the channel with the canvas link using slack_send_message. Ask the user for permission before sending.

7d. Prompt for Weekly Schedule

After the first successful run, ask the user if they want this automated:

"Want me to run this every Monday morning and post to [channel]? I'll scan all competitors, generate the report, and drop the highlights in Slack automatically."

If yes, create a scheduled task:

• Task ID: competitor-watch-weekly
• Schedule: Monday mornings (cron: 0 9 * * 1 — 9am local)
• Prompt: The full competitor-watch skill execution including Slack delivery
• Description: "Weekly competitor intelligence scan — posts highlights + full canvas to Slack"

Only prompt for scheduling on the first run or if no competitor-watch-weekly task exists. On subsequent runs, just deliver silently.

---

Connected Workflows

After the scan:

• Deep dive on one competitor's ads: /industry-trends --brand [domain]
• Generate concepts that counter competitor moves: /create-concepts with competitive context
• Analyze your own performance in detail: /performance-analysis
• Build a brief inspired by a competitor's approach: /build-brief