Claude for Marketing: The Complete Playbook

Three modes. One stack. Each has a different role in a marketing team's workflow.

Claude chat: fastest entry point, no setup

What it does: paste in anything (a brief, a post, a competitor URL, a CSV export, a transcript) and ask Claude to do something with it. No setup, no tools, no configuration.

What it is good for:

• Drafting and editing: write a first draft, improve an existing piece, change tone, cut length
• Quick research: summarise a competitor page, explain a concept, write a briefing on a market
• Ad hoc analysis: paste in a spreadsheet and ask what is notable
• Ideation: generate hooks, angles, post ideas, email subject lines
• Prompt refinement: iterate on copy in real time

Limitation: every session starts fresh. Outputs stay inside the chat unless you manually copy them out. For recurring tasks, this becomes friction. That is what Cowork and Claude Code solve.

Claude Cowork: recurring workflows that run on a schedule

What it is: a tab in the Claude Desktop app (download free at claude.ai/download). Not a chatbot. An agent that executes on your computer. You describe what you want done, it does it.

What it can access: your local files and folders, Slack, Gmail, Google Drive, Microsoft 365, HubSpot, and more via the connector marketplace. Each connector is a toggle you authorise once. No IT team, no API keys, no code.

What it does that chat cannot:

• Reads files directly from your computer without you uploading them
• Creates and saves files locally in the right folder without you downloading them
• Connects to your actual tools in real time (pulls this week's Slack messages, checks your live analytics dashboard, reads your inbox)
• Runs on a schedule: set up a task once, it runs every Monday at 8am
• Operates from your phone via Dispatch (mobile app paired with desktop): send one message from your phone, Claude runs the task on your desktop

Best for: competitive intelligence briefs, weekly performance reports, content repurposing, anything that needs to happen consistently on a schedule and pull from real connected data.

Setup: download Claude Desktop, open Cowork tab, enable connectors in settings (each is a simple toggle + authorise), run your first task manually and verify, then type /schedule to set the cadence.

Claude Code: bulk processing and API-based automation

What it is: Claude running in your terminal. It writes and executes code, calls APIs, scrapes websites, processes bulk data, and chains multi-step workflows from plain English prompts.

Install:

npm install -g @anthropic-ai/claude-code
claude

What it does that Cowork cannot:

• Processes thousands of rows of CRM data in one run
• Calls enrichment APIs (Apollo, Hunter, Clearbit) row by row and merges results
• Scrapes competitor pricing pages and diffs them against saved baselines
• Generates 50 personalised outreach emails from a CSV in under 10 minutes
• Builds and runs Python scripts without you writing a line of code

Best for: list building, data enrichment, CRM cleanup, outreach personalisation at scale, competitive page monitoring, signal processing.

For marketing engineers and technical GTM teams: this is the primary tool. For non-technical marketers: Cowork covers most recurring workflows without needing to touch the terminal.

Which tool for which job

Task	Tool
Draft a LinkedIn post	Chat or Cowork
Write a competitive brief (one-off)	Chat
Competitive brief every Monday, automated	Cowork
Repurpose a blog post across formats	Cowork (browses your LinkedIn for style context)
Build a prospect list from ICP criteria	Claude Code
Enrich 200 contacts with emails and company data	Claude Code
Clean and normalise job titles in a CRM export	Claude Code
Write 50 personalised cold emails from a CSV	Claude Code
Weekly performance report, automated	Cowork
Monitor competitor pricing page for changes	Claude Code
Generate a carousel post and schedule to LinkedIn	Cowork + Claude Code (GPT Image 2 + Buffer)
Audit landing page copy and rewrite weak sections	@landing-page-optimizer agent
Build a marketing attribution dashboard	@analytics-engineer agent

Sources: Toni (Actionable AI for Work) + Claude Code Workflows for GTM lead mag + 60 Claude Code Agents lead mag

Three levels of competitive intelligence with Claude: the automated weekly brief (Cowork), real-time pricing page monitoring (Claude Code), and on-demand competitive teardowns (@market-analyst agent).

The Monday brief: Cowork scheduled task

The problem: competitive intelligence is either three months old or lives inside one person's head. Someone bookmarks competitor pages, screenshots pricing, saves a launch announcement in a Slack thread. None of it becomes anything usable because synthesising 15 scattered sources into a brief takes 2-3 hours nobody has.

The solution: a scheduled Cowork task that runs every Monday at 8am. You edit for 5 minutes. You forward it to the team. The brief exists every week.

Setup: enable Slack, Gmail, and Google Drive connectors in Cowork settings. Run the task manually first and check the output. Type /schedule, pick Monday 8am, done.

Critical step most people skip: define signal vs noise in your instructions. Without it, Claude treats a competitor's intern blog post with the same weight as a pricing restructure. Add a section defining what counts as a signal and what to skip. Week two will be significantly better.

Estimated time saving: 8-10 hours per month of assembly work eliminated. Your 5-minute editing pass provides the judgment that AI cannot: the so-what context behind each signal.

You are running my weekly competitive intelligence brief.

Sources to check:
1. Visit these competitor pages:
   - [COMPETITOR 1] pricing page: [URL]
   - [COMPETITOR 1] changelog or product updates: [URL]
   - [COMPETITOR 1] blog: [URL]
   - [COMPETITOR 2] pricing page: [URL]
   - [COMPETITOR 2] blog: [URL]

2. Pull this week's messages from the #[SLACK CHANNEL] Slack channel
   Look for: pricing mentions, feature announcements, customer complaints, comparison threads

3. Check Gmail for newsletters from: [LIST COMPETITOR DOMAINS]

What counts as a SIGNAL (include):
- Pricing changes or new tiers
- New feature announcements or product updates
- Messaging shifts on their homepage
- Funding announcements or major hires
- Customer complaints or praise revealing gaps

What counts as NOISE (skip):
- Blog posts about culture, team, intern programmes
- Generic industry thought leadership with no product angle
- Anything published more than 7 days ago

Output format: one-page brief, three sections only:
1. What changed this week
2. Pricing moves
3. Messaging shifts

For each item: one sentence on what happened, one sentence on the so-what for our positioning.
Save as a Word doc to Google Drive at: [FOLDER PATH]
If nothing significant changed, say so clearly. Do not pad the brief.

Pricing page diff: Claude Code

Runs on demand or via a cron trigger. Scrapes competitor pricing pages, compares against a saved baseline, and writes a plain English summary of what changed. Takes a screenshot for visual reference.

# First run: create the baseline
"Visit [COMPETITOR URL]/pricing.
Extract all pricing tiers, prices, features per tier, and any promotional language.
Save the structured output as reference/[competitor]-pricing-baseline.md"

# Weekly run: diff against baseline
"Visit [COMPETITOR URL]/pricing.
Compare today's pricing against the saved version in reference/[competitor]-pricing-baseline.md.

If anything changed:
- Write a plain English summary of every change (tier changed, old price, new price, what features moved)
- Save the diff to competitive-intel/pricing-changes-[DATE].md
- Take a full-page screenshot and save it

If nothing changed: write 'No changes detected on [DATE]' to competitive-intel/pricing-log.md.
Save today's scraped version as the new baseline."

Add hiring signal tracking alongside it:

"Scrape the jobs page at [COMPETITOR URL]/careers.
For any new job postings since [LAST CHECK DATE]:
- Extract: job title, department, location, date posted
- Flag if 3+ new postings in the same department (hiring surge signal)
Write findings to competitive-intel/hiring-signals-[DATE].md"

On-demand competitive teardown: @market-analyst agent

From the 60 Claude Code Agents lead mag. Install via the agents directory. Use when you need a full competitive brief on a specific competitor for a deal, a launch, or a positioning review.

@market-analyst Compare us against [COMPETITOR 1], [COMPETITOR 2], [COMPETITOR 3].
We sell [PRODUCT] to [ICP].

Sources: their homepage, G2 reviews from last 90 days, their recent LinkedIn content,
their job postings, their changelog.

Return:
1. Their current positioning (in their own words from their homepage)
2. Top 5 customer complaints from G2 reviews
3. Recent product signals from changelog or job postings
4. Messaging shifts visible in recent LinkedIn content
5. Where we win clearly and why
6. Where they win and what we would need to match it
7. The one claim we can make that they cannot make back
8. What to say when their name comes up on a call

Save to competitive-intel/[competitor]-teardown-[DATE].md

Connect this to the Monday brief:

@market-analyst Take this week's competitive brief: [PASTE BRIEF]
Build a battlecard update:
1. Positioning vs ours (updated for any messaging shifts this week)
2. 3 discovery questions that surface their current weaknesses
3. What to say when their name comes up on a call

Source: Toni (Actionable AI for Work) + @analytics-engineer agent

The Friday report that keeps not getting done, and why the consistency is worth more than the time saving.

Why the report never gets done

Good marketing teams have been weeks or months behind on their weekly report. It is not a motivation problem. It is a friction problem. Building the report means opening three dashboards, exporting data, formatting it, writing the takeaways, and by then it is 5pm Friday and everyone has gone.

The real cost is not the 30-45 minutes per week. It is the decisions that do not get made because the data is not ready. Those cost more than the time math suggests.

Cowork makes the report exist every Friday. You spend 10 minutes verifying the numbers. You forward it. That consistency is the value.

The honest limitation

Computer use (Cowork navigating your screen to pull data) is the most fragile part of this workflow.

Clean dashboards with standard layouts: fine.

Complex UIs with custom date pickers or heavy JavaScript: struggles.

Toni ran this on HubSpot and Cowork clicked the wrong date range. The report looked great. The numbers were wrong. He caught it because he checked.

The actual flow: Cowork builds the report, you spend 10 minutes verifying key numbers against the actual dashboards. That is the quality filter.

Start with one data source. Get Google Analytics working first. Then add email metrics. Setting up 5 dashboards on day one costs more debugging time than it saves.

The exact prompt: Cowork Friday report

You are building my weekly marketing performance report.

Data sources:
1. Open Chrome, navigate to [GOOGLE ANALYTICS URL]
   Pull: traffic by channel for last 7 days vs prior 7 days
   Pull: top 5 pages by sessions
   Pull: traffic by device (desktop vs mobile)

2. Open Chrome, navigate to [EMAIL PLATFORM URL]
   Pull: campaign metrics for emails sent this week
   For each: subject line, open rate, click rate, unsubscribes

3. [ADD ANY OTHER DASHBOARD AND WHAT TO PULL]

Report structure:

Weekly Marketing Report [DATE RANGE]

Section 1: Traffic
- Total sessions and % change vs last week
- Top 3 channels and trend (up/down/flat)
- One sentence on the most notable traffic pattern

Section 2: Email performance
- Best performing email (subject line, open rate, click rate)
- Worst performing email (what may have caused low performance)
- One recommended action based on this week's email data

Section 3: Recommended actions
- Three specific actions for next week
- Each: what to do, why, what metric it targets

Save as a Google Doc to: [DRIVE FOLDER PATH]
File name: Weekly Report [DATE]

IMPORTANT: If you cannot clearly read a number from a dashboard (wrong date range, page did not load, unclear filter), do not guess. Flag it as NEEDS VERIFICATION and note what you attempted.

10-minute verification checklist

Run this before forwarding the report to anyone:

1. Open Google Analytics and check: does the sessions number in the report match what you see for last 7 days?
2. Open your email platform and check: do the open rates for each campaign match?
3. Read the three recommended actions: are they specific enough to act on, or are they generic advice?
4. Check the date range is correct throughout the report.
5. If anything is flagged NEEDS VERIFICATION, check it manually and fill it in before sending.

Deeper attribution: @analytics-engineer agent

For marketing teams who need to go beyond surface metrics and connect campaigns to pipeline:

@analytics-engineer Audit our email campaign tracking.
We send from [SEQUENCER] but have no visibility into which sequences produce pipeline.
Fix the tracking so the weekly report can include pipeline attributed to each campaign.

@analytics-engineer Build a weekly marketing dashboard that shows:
- LinkedIn post reach by ICP-matching accounts
- Email reply rate by sequence
- Meetings booked attributed to each channel
Output as an HTML dashboard saved locally.

@analytics-engineer Our Google Analytics shows 2,000 visitors per month but we book 3 calls.
Find where they are dropping off and what to do about it.

Sources: Toni (Actionable AI for Work) + LinkedIn in 2026 lead mag + LinkedIn Hook and Post Template Library

Three layers: repurposing existing content (Cowork), automating carousel production (Claude Code + GPT Image 2), and the 2026 LinkedIn strategy for content that actually performs.

What changed on LinkedIn in 2026

Two forces hit at once: AI tools lowered the barrier to create content to near zero, flooding the timeline with average posts. More executives than ever started posting because founder-led content became mainstream B2B marketing strategy. LinkedIn also started suppressing organic reach in favour of paid.

The founders who are still growing have one thing in common: their content is impossible for AI to copy. Three moats create that:

Narrative mode: pair every take with a story or specific anecdote that only you could have written. A hiring post that names the two specific hires, mentions the first three years the founders were the de-facto head of client services, and references the DMs asking about freelance writers outperforms a generic hiring post. That specificity is the moat.

Data mode: share proprietary data only you have access to. Aggregate client performance data, response rate benchmarks, funnel conversion numbers. Put it in a chart. LinkedIn audiences engage heavily with unique data visualisations because AI cannot replicate data it does not have.

Physical mode: IRL photos and real footage. New hire photos, office behind-the-scenes, conference and event content. Real images are the clearest signal to both LinkedIn's algorithm and to readers that your content is not AI-generated.

Content filter to run before every post:

1. Is there a narrative, data, or physical moat in this content?
2. Could this exact post have been written by someone who entered the topic into Claude?
3. Is there at least one specific detail (a name, a number, a quote) that only you could have included?

If the answer to question 2 is yes, the post needs more work.

Content repurposing: Cowork

Two things make Cowork repurposing better than pasting into Claude chat:

1. Cowork browses your actual LinkedIn page before drafting. It checks what your recent posts look like, what formats are getting engagement, and matches the draft to your current style.
2. The Brand Voice Skill applies automatically to everything. Write your brand rules once in a plain text file. Cowork loads it on every task. Every output follows your voice without you re-briefing it.

Build the Brand Voice Skill first (30 minutes, compounds across everything):

# Save as brand-voice.md in your Cowork folder

# Brand Voice

## Who we are
[Company name]. [What you do and who you do it for in one sentence].

## Tone
WE ARE: [3-5 adjectives: e.g. direct, peer-level, pragmatic, specific]
WE ARE NOT: [3-5 anti-adjectives: e.g. corporate, salesy, preachy, vague]

## Our audience
[Title]. [Their pain]. [What they are trying to achieve].
They are [time-poor / sceptical of hype]. Write to that.

## What we never say
- No em dashes
- No hashtags in post bodies
- No: leverage, synergy, bandwidth, streamline, foster
- No filler openers: I hope this finds you well, I wanted to reach out
- No generic CTAs: Let me know if you're interested, Feel free to reach out

## What we always do
- One idea per line. Three lines maximum per paragraph.
- Lead with specifics: a number, a name, a real example
- British English

## LinkedIn specifically
- Hook earns See More in 2 lines
- No hashtags in the body
- One CTA per post

Repurposing prompt (trigger from phone via Dispatch, or run manually):

"Read the brand-voice.md file in my Cowork folder and apply it to all outputs.
Open Chrome and go to my LinkedIn: [YOUR URL]
Check my last 10 posts. What formats are getting the most engagement? Match the drafts to that style.

Source article: [PASTE URL OR TEXT]

Produce:
1. LinkedIn post (under 200 words, hook earns See More in 2 lines, one CTA, no em dashes)
2. Email snippet: prospect angle (under 100 words, opens with the problem not the product)
3. Email snippet: customer angle (under 100 words, opens with an insight or expansion opportunity)
4. Three social captions for [X / other platform] (under 60 words each, different angles)
5. Slack summary (under 50 words, internal sharing)
6. Lead magnet hook post (LinkedIn format, Comment [KEYWORD] and I'll DM you [RESOURCE])

Save all outputs to Google Drive at: [FOLDER PATH]"

Quality by format: LinkedIn draft is strongest (minor tweaks). Email snippets need personality added. Social captions often too safe and need rewriting. Slack summary usually ready as-is.

Carousel automation: Claude Code + GPT Image 2 + Buffer

Full pipeline: topic in, on-brand carousel PDF out, scheduled to LinkedIn via Buffer. Built in one Claude Code session.

Stack:

• Claude Code: orchestrates everything
• GPT Image 2 (OpenAI): generates the slide images. Best image model available as of 2026.
• Cloudinary: stores images and PDFs with public URLs (required for Buffer API)
• Buffer: LinkedIn post scheduling tool (free plan works)

Setup sequence:

# Step 1: Set up Cloudinary
"Set up Cloudinary media storage in this project folder.
Whenever images are generated, also upload them to Cloudinary automatically.
Ask me for my Cloudinary account details."

# Step 2: Set up the carousel generator
"I need to create LinkedIn carousel posts.
I've uploaded style reference images to /style-guide/ for you to take design style from.
Use the GPT Image 2 model for image generation.
Store all generated images in Cloudinary.
Create 4-5 slides that flow together as a cohesive carousel.
Include a slide number in the corner of each slide (1 of 5, etc).
The final slide should include a comment CTA.
Add the LinkedIn post writer skill from my Skills folder before writing the copy.
Topic: [PASTE YOUR TOPIC OR VIDEO TITLE]"

# Step 3: Schedule via Buffer
"I've set up a Buffer account connected to my LinkedIn.
Schedule this carousel as a LinkedIn post for [DAY] at [TIME].
Connect to the Buffer API, upload the PDF to Cloudinary to get a public URL, then schedule."

API key security: never paste API keys in chat. Ask Claude for instructions on how to store them as environment variables instead.

Note: LinkedIn carousel posts must be uploaded as a PDF, not individual PNG files. Claude stitches the images into a PDF automatically. Enable PDF delivery in your Cloudinary settings when prompted.

10 LinkedIn hook formulas for B2B founders

# 1. Mistake with specific loss
I made X huge mistakes during my first Y months of [endeavour].
(these cost me [£155k+ and 417 hours])

# 2. Hard truth reveal
Here's the no.1 mistake [specific group] make on LinkedIn
(this may be hard to hear):

# 3. ICP direct address
If you're a [specific audience], you're likely making this mistake:
[common mistake].

# 4. The reframe
Most people know exactly [common knowledge].
The problem isn't [common limiting belief].

# 5. Proven strategy with two outcomes
I [achieved desirable outcome] in just [short time frame].
I also [additional related outcome].
But there's a strategy behind this.

# 6. Remove the prerequisite
I [achieved goal Y] without [common pre-requisite 1].
I [achieved goal Y] without [common pre-requisite 2].
To [achieve goal Y], all you really need to do is:

# 7. Contrarian
[thing everyone does] is dumb.
Here is what I do instead.

# 8. Trend observation
There's a rising trend in [industry/platform] to [behaviour],
[get sub-par outcome] then [misguided conclusion].

# 9. The progressive milestone
My first [time period] doing [activity], I [achieved very little].
The next [time period]: [small milestone].
By the end of [larger timeframe]: [massive milestone].
But I could have done it quicker.

# 10. Stakes escalation
[Target audience]: You need [critical element].
(And not just because [superficial reason]).

Source: Claude Code Workflows for GTM lead mag (6 workflows) + LinkedIn in 2026 lead mag (algorithm and thought leader ads)

Claude Code handles the volume work (list building, enrichment, personalisation at scale). Claude chat and Cowork handle the judgment work (qualifying, sequencing, DM strategy).

List building: Claude Code

You describe your ICP in plain English. Claude writes a scraping or API-based script, pulls from relevant sources, deduplicates, and outputs a structured CSV.

Typical result: 50-200 qualified prospects in under 15 minutes vs 4-8 hours of manual SDR research.

# Basic ICP list build
"Build me a list of Series A through Series C fintech companies that use Salesforce.
Include: company name, funding amount, employee count, headquarters, CEO LinkedIn URL.
Output as leads.csv."

# Signal-enriched list
"Find [TARGET TITLES] at [COMPANY TYPE] companies that have [SIGNAL: raised funding / posted SDR roles / published content about outbound] in the last 30 days.
Filter: 50-500 employees, US/UK/EU.
For each: name, LinkedIn URL, company, signal source, signal date.
Output as leads.csv."

# Job posting trigger list
"Find companies posting [JOB TITLE] roles on LinkedIn and Indeed in the last 14 days.
Filter: [INDUSTRY], [GEO], [COMPANY SIZE].
For each: company name, posting date, company website, employee count.
Save as hiring-signals.csv."

Data enrichment: Claude Code

Takes your input CSV, calls enrichment APIs row by row, handles rate limits and retries, and merges all results into one output file.

# Multi-field enrichment from Apollo
"Enrich every row in leads.csv using Apollo API.
For each contact, return: verified work email, direct phone, current title, current company, LinkedIn URL, company size, company industry, funding stage.
Add as new columns. Save to leads-enriched.csv.
Handle rate limits: max 10 requests per second. Retry failed rows up to 3 times. Log failures to failures.csv."

# Email waterfall (3 providers)
"Enrich emails for every row in leads.csv:
Step 1: Try Apollo. If returned, mark source = Apollo, move to next row.
Step 2: If Apollo returns null, try Hunter. If returned, mark source = Hunter.
Step 3: If Hunter returns null, try Findymail. If returned, mark source = Findymail.
Step 4: If all three return null, mark email = null, source = none.
Output: leads-enriched.csv with columns: name, company, email, email_source."

Outreach personalisation at scale: Claude Code

Not merge tags. Actual personalisation based on company context per row: tech stack, funding stage, job posting signals, recent content. 50 unique personalised emails in under 10 minutes.

Quality control: review 10-15% of the output before sending. Common failure modes are hallucinated company details and generic openers despite specific data being available.

# Signal-led personalisation
"Read prospects.csv. For each row, write a 3-sentence cold email referencing their specific tech stack and a pain point common to companies at their funding stage.
Tone: direct, peer-to-peer.
Rules:
- Opening line references their specific tech stack (use tech_stack column)
- Second sentence names the pain we solve for companies at their stage (use funding_stage column)
- Third sentence: one question CTA
- Under 90 words. No em dashes. No filler phrases.
Output: outreach.csv with columns: name, email, subject_line, body."

# Full 3-touch sequence per prospect
"Read prospects.csv.
For each prospect, write a 3-touch sequence:
Touch 1 (Day 1): Signal-led email referencing tech_stack or recent_news column. Under 80 words.
Touch 2 (Day 5): Different angle. Relevant insight or stat. No repeat of Touch 1. Under 70 words.
Touch 3 (Day 14): Breakup email. Honest, no guilt trip, under 50 words.
Each touch needs its own subject line. No em dashes.
Output: sequences.csv with: name, email, touch_1_subject, touch_1_body, touch_2_subject, touch_2_body, touch_3_subject, touch_3_body."

LinkedIn algorithm: the daily 15-minute system

LinkedIn is suppressing organic reach. You cannot rely on the algorithm to serve your content to the right people. Two manual actions bend it in your favour.

20 outbound connection requests per day (Monday to Friday): send to your ICP only, no message attached (research shows blank connection requests have higher acceptance rates than those with a message). When they accept, they are likely to see your next 1-2 posts. 100 per week seeds 100 ICP-relevant people into your audience.

5-10 comments per day on ICP-adjacent content: comments are getting more impressions than posts themselves on LinkedIn right now. Commenting on content your ICP follows puts your name in front of them even if they never see your posts. Do not say great point. Add something the post did not say.

Thought leader ads: take your best-performing organic posts (those that attracted ICP engagement) and put $30/day behind them as thought leader ads. They do not look like ads: the only indicator is small promoted text under the author name. Use Apollo to build a company list matching your ICP and upload it directly to LinkedIn Ads for targeting. The PS line URL trick: run posts organically without a link for a week, then edit the post to add a URL as a PS line before boosting. That URL stays visible in the ad and serves as your CTA.

Source: 60 Claude Code Agents lead mag + all existing lead mags

The Claude Code agents and Cowork skills that execute marketing work. Install once, invoke with @agent-name or naturally in any Claude Code session.

@analytics-engineer

Tracks what actually matters. Audits your analytics setup, identifies what is being tracked vs what should be, and builds dashboards that connect marketing activity to pipeline.

@analytics-engineer Audit our email campaign tracking.
We send from [SEQUENCER] but have no visibility into which sequences produce pipeline.
Fix the tracking so we can attribute meetings to specific campaigns.

@analytics-engineer Build a weekly marketing dashboard that shows:
- LinkedIn post reach by ICP-matching accounts
- Email reply rate by sequence
- Meetings booked attributed to each channel

@market-analyst

Maps the competitive landscape, identifies positioning gaps, and surfaces unfair advantages.

@market-analyst Compare us against [COMPETITOR 1] and [COMPETITOR 2].
We sell [PRODUCT] to [ICP]. Sources: their homepage, G2 last 90 days, job postings, changelog.
Return: positioning summary, top customer complaints, product signals, where we win, where they win,
the one claim we can make that they cannot make back, what to say when their name comes up on a call.

@market-analyst Take this week's competitive brief: [PASTE BRIEF]
Build a battlecard update: 3 discovery questions that surface their weaknesses,
what to say when their name comes up on a call.

@email-automator

Designs and builds email sequences for every stage of the customer lifecycle.

@email-automator Build a 5-touch cold email sequence for [TITLE] at [COMPANY TYPE].
Signal: they just posted a [ROLE] role.
Offer: [ONE SENTENCE VALUE PROP].
Different angle per touch. Subject line per email. No em dashes.

@email-automator Build a 7-day trial onboarding sequence.
Activation event: connecting their CRM.
Goal: get them to connect within day 3.

@email-automator Our re-engagement campaign has a 0.8% click rate.
Audit it and rebuild it. Target: churned users from 30-90 days ago.

@landing-page-optimizer

Audits and rewrites landing pages for conversion. Tests headline formulas, restructures page flow, rewrites CTAs.

@landing-page-optimizer Audit our homepage. Conversion rate: 1.2%. ICP: [DESCRIBE].
Rewrite the hero section. Give me 3 headline variants using:
- Outcome: [Specific result] in [Timeframe] for [ICP]
- Pain: Stop [painful thing]. Start [desired outcome].
- Proof: [Customer] got [specific result] using [product]

@landing-page-optimizer Our demo request page has a 40% form abandonment rate.
Find out why and fix it.

@content-writer

Audits all UI copy and marketing microcopy: button labels, empty states, error messages, tooltips, CTAs.

@content-writer Audit all copy on our [landing page / product onboarding / email footer].
Apply these brand voice rules: [PASTE YOUR RULES]
Flag: vague labels, jargon, passive voice, generic CTAs.
Rewrite every flagged item. Output: original vs rewritten per item.

Quick reference: all key prompts

# COMPETITIVE INTELLIGENCE

# Cowork Monday brief (scheduled)
"Visit [5 COMPETITOR PAGES], pull #[SLACK CHANNEL] messages, check Gmail for competitor newsletters.
Signal vs noise rules: [DEFINE]. Output: 3-section brief. What changed / pricing moves / messaging shifts.
Save as Word doc to Google Drive. One sentence per item on what happened, one on the so-what."

# Claude Code pricing diff
"Visit [COMPETITOR URL]/pricing. Compare against reference/[competitor]-pricing-baseline.md.
If changed: write summary, save diff, screenshot. If unchanged: log. Save today's as new baseline."

# CONTENT PRODUCTION

# Cowork repurposing (from phone via Dispatch)
"Read brand-voice.md. Browse my LinkedIn [URL] for recent style.
Source: [URL or text]. Produce: LinkedIn post (200 words), 2 email snippets, 3 social captions, Slack summary, lead magnet hook post.
Save to Google Drive [FOLDER]."

# Claude Code carousel (schedule to LinkedIn)
"Create LinkedIn carousel for topic: [TOPIC]. Reference images in /style-guide/.
GPT Image 2 model. 4-5 slides, numbered, comment CTA on last slide.
Store in Cloudinary. Combine into PDF. Write post copy using LinkedIn post writer skill.
Schedule via Buffer API for [DAY] at [TIME]."

# PERFORMANCE REPORTING

# Cowork Friday report (scheduled)
"Open Google Analytics [URL]: pull traffic by channel last 7 days vs prior 7 days, top 5 pages.
Open email platform [URL]: pull campaign metrics for emails sent this week.
Write report: traffic section, email performance, 3 recommended actions.
Flag NEEDS VERIFICATION for any number you cannot clearly confirm.
Save to Google Drive [FOLDER]."

# OUTREACH AND LEAD GENERATION

# Claude Code list build
"Build a list of [ICP CRITERIA: funding stage, company type, tech stack, geo, size].
Include: [FIELDS]. Deduplicate. Output as leads.csv."

# Claude Code enrichment
"Enrich leads.csv via Apollo API. Return: verified email, phone, title, company size, funding stage.
Add as new columns. Log misses to failures.csv. Max 10 requests/second."

# Claude Code personalised email batch
"Read prospects.csv. For each row, write a 3-sentence cold email referencing tech_stack and funding_stage columns.
Under 90 words. No em dashes. No filler phrases. One question CTA.
Output: outreach.csv with name, email, subject_line, body."

Failure modes to know

Cowork computer use is fragile: works on clean, standard websites. Struggles with complex JavaScript-heavy dashboards and custom date pickers. Always verify numbers Claude pulls from a screen against the actual dashboard.

Your laptop must stay awake: Cowork scheduled tasks only run while Claude Desktop is open. Not cloud-based yet.

Token consumption: complex Cowork tasks use significantly more tokens than regular chat. Schedule for early morning to take advantage of doubled off-peak limits.

Claude Code hallucinations on enrichment: review 10-15% of any bulk-generated copy or enriched data before using it. Common failure modes are hallucinated company details and outdated information in API responses.

Signal vs noise without rules: Cowork treats all content equally without explicit definitions. Define what counts as a signal and what to skip in every task instruction.