Content & SEO
Primary tools: Claude (via client Projects), ChatGPT (for spreadsheet formulas).
Blog post drafting workflow
- Build a CRISP prompt with the client's Claude Project open (see Module 3)
- Generate a first draft — treat it as an outline with good bones, not a final piece
- Run the output through the banned words check before touching anything else
- Rewrite at least 30% by hand. Add specific examples, data, and your own opinions.
- Read it out loud. Fix what sounds wrong.
- If a client has flagged AI content before, do not use AI for their work
Meta descriptions and SEO copy
- Use the Claude Chrome extension for quick meta description generation directly in your browser
- Include the target keyword in the prompt explicitly. Claude won't always infer it.
- Specify the character limit (150 to 160 chars for meta descriptions)
- Use the SEO+LLM Content Optimization system prompt for content you want to rank in both traditional search and AI search
Reddit content
Follow the Reddit AI Rules document exactly. No shortcuts. Key points:
- Give Claude the subreddit, the specific question, the account persona, and examples from that community
- Never start a prompt with "Write a Reddit post about..."
- Answer the question first. Brand second.
- No em dashes, no emojis, no hashtags, no sign-offs
- Short sentences. Direct. Opinionated.
AEO (Answer Engine Optimization)
AEO is about writing content that AI search engines (like Perplexity, ChatGPT, and Google AI Overviews) will cite and quote. Here's how to approach it:
- Use the SEO+LLM Content Optimization system prompt when writing content for clients who want AEO coverage
- Structure content with clear Q&A sections (AI search engines love explicit answers)
- Use Otterly.AI to monitor where your clients appear in AI search results
- Think of AEO and traditional SEO as overlapping, not competing
Design
Primary tools: Photoshop AI (Firefly), Ideogram, Claude (for copy needs).
AI image rules
- No AI-generated human faces. Photorealistic faces require written client permission and careful review for bias.
- Naturalistic lighting is required. AI-generated images with obvious artificial lighting are a quality tell.
- Use "photograph of" in Ideogram and Firefly prompts to get more realistic results
- Never generate the likeness of a real person without written permission
- Follow the AI Style Guide for when and how to label AI-generated visuals in deliverables
Generative fill boundaries
- Acceptable: Backgrounds, textures, extending existing images, removing objects
- Not acceptable without client approval: Using AI generation as the primary subject of a deliverable
- When in doubt, check the AI Style Guide for guidance on disclosure requirements
Digital Ads
Primary tools: Claude (for ad copy), ChatGPT (for formulas and spreadsheets).
⚠️ Data warning: Never trust AI with actual campaign numbers without verifying independently. AI hallucinates numbers. This is especially relevant for performance reporting and bid calculations.
- Ad copy variations: Claude is good at generating multiple headline and description variations. Give it the campaign objective, audience, character limits, and examples of approved copy.
- Audience research: Use Claude to build out audience segments based on demographics and psychographics. Verify with actual platform data.
- Budget calculations: Use ChatGPT or a spreadsheet formula for math. Don't ask Claude to do arithmetic on real campaign budgets and take the answer at face value.
- Performance summaries: Claude can help write client-facing summaries of campaign performance if you paste in the actual numbers. Do not ask it to generate or calculate the numbers itself.
Web Development
Primary tools: Claude, Cursor.
Claude vs. Cursor
- Claude: Use for reasoning about code architecture, understanding what a block of code does, debugging logic, and writing code from scratch when you need to explain the problem in plain language
- Cursor: Use for in-editor completion, refactoring within a file, and working within an existing codebase where the AI needs context from multiple files
Regex and formulas: Claude is excellent at generating regex patterns and HubSpot validation formulas. A real example: someone used Claude to fix a complex HubSpot zip code validation regex. That's a good model for this kind of task. Give Claude the exact format you need to match, the specific platform, and what should and shouldn't pass.
Non-negotiable rule
AI-generated code must be tested. Period. Claude doesn't run the code. It doesn't know your specific environment. It doesn't know what version of WordPress you're running. Test every generated snippet before it touches production.
Account Executives
Primary tools: Claude (via client Projects), the AI AE Assistant, Zapier.
Your AE AI workflow
- Meeting prep: Before a client call, open their Claude Project and ask for a summary of recent work, key talking points, and any outstanding feedback threads
- Follow-up emails: Use the AI AE Assistant output as a starting point, then edit. Never send the AI draft unreviewed.
- Proposal sections: Claude is good at drafting proposal language if you give it the scope, the client's goals, and examples of past proposals that were accepted
- Client reporting summaries: Paste in data and ask Claude to write the narrative. Don't ask Claude to generate the data.
- Codex maintenance: This is your job. When the client gives feedback that changes how we should write for them, update the Codex. Update the project. Post in Slack.
💡 AE reminder: You own the Claude Project for your clients. When it's well-maintained, your whole team gets better output. When it's neglected, everyone struggles. The five minutes it takes to update the project after a client call pays off every time someone uses it.
Try it yourself
Find the section above for your department. Pick one workflow you haven't tried yet. This week, use AI for that specific task using the approach described. Then:
- Did it save you time?
- What would you do differently next time?
- Is there a step missing from the workflow above?
Share your experience in #ai-and-automations. Your real-world feedback makes this guide better for everyone.
Across all departments: common mistakes
- Working outside the Claude Project for client work. Always use the project. That's where the context lives.
- Trusting AI-generated numbers without checking. Applies to everyone: Ads for campaign data, Dev for outputs from calculations, Content for statistics in blog posts.
- Not updating the Claude Project when the client changes direction. Stale projects produce stale output.
- Skipping the human review step. AI handles first drafts. You handle judgment and delivery.
Social Media
Primary tools: Claude (via client Projects), Ideogram (for image generation).
Caption generation workflow
Content calendar and brainstorming