“3 mistakes marketers make with AI” — @Mitch (Ship 30 for 30)

Why this is in the vault

Sales-funnel-adjacent newsletter, but the three-mistake taxonomy (fancy-Google / outsourcing-taste / staying-in-your-lane) is genuinely usable and maps directly onto frames Ray Data Co already operates from — file the framework, ignore the CTA.

Source

• Email subject: “3 mistakes marketers make with AI”
• From: hello@ship30for30.com
• Received: 2026-05-06T12:26Z
• Gmail thread: 19dfd409a804b54b
• Tail of email is a sales pitch for the Claude Code Marketing Bootcamp (May 18) — body content above the pitch is the substance.

Summary

Three mistakes marketers make when using AI, plus the framing for what “good” looks like in 2026. Newsletter is sales-funnel-adjacent (waitlist push at the bottom) but the three-mistake taxonomy itself is genuinely useful and matches frames Ray Data Co already operates from.

The three mistakes

Mistake 1 — The “fancy Google” problem

Marketers still treat AI like a smarter search box. They ask it questions (“what’s a good headline for X?”) instead of giving it tasks (“draft 10 headlines, score them against my voice doc, return the top 3 with rationale”).

Mitch’s frame: tools like Claude Code, Claude Cowork, Cursor, Codex now move AI out of the chatbot and into the wider world via:

• Documenting best-practices for copywriting, analysis, and design (skills / playbooks)
• Chaining prompts into an “assembly line” process
• Talking to software services that report, format, design, or even code

Implication: asking AI for trivia is “way below its current pay grade.” Start handing it tasks with defined outputs.

Mistake 2 — Outsourcing taste instead of tasks

Taste is the part of marketing nobody can teach in a 2-week bootcamp. If you delegate creative judgment to AI, you get timid, average “AI purple” creativity (the bland, hedged, middle-of-distribution voice).

The fix: put words to your specific likes and dislikes — in music, movies, websites, writing — and feed those preferences in. Then AI can either:

• Make more things you like (taste-aligned generation)
• Surface options you wouldn’t have thought up yourself (taste-extending exploration)

Tools called out: Claude Design, custom skills plugged into Claude Code for refining writing and visual styles even if you’re not a designer.

Mistake 3 — Staying in your lane

Good writers say AI sucks at writing. Good designers mock AI websites. Good engineers (until early 2026) said AI couldn’t code. Meanwhile creators, marketers, and business owners outside those guilds used AI to ship real work.

Case study cited: Zac O’Hara, a contractor running North Shore Masonry, used Claude Code to automate estimates, schedules, reports, and follow-ups — credited with helping generate $200K+ for the shop. (Source: jobtread.com case study.)

Implication: don’t be a craft-defender. Be a tool-user. The people winning aren’t the ones with the most native skill in a domain — they’re the ones willing to use AI to compress the parts of their workflow that aren’t their core craft.

Mapping against Ray Data Co

Strength: medium. All three mistakes map cleanly to things Ray Data Co already does or believes:

• Mistake 1 (tasks not questions) ↔ the move from chatbot Claude to skill-driven Claude Code COO. RDCO’s whole architecture is “give the agent tasks with structured I/O” — not “ask the agent questions.” This is what ~/.claude/skills/ is for. Mitch’s framing is downstream of what Ray operates on daily.
• Mistake 2 (don’t outsource taste) ↔ the founder’s whole “no slop cannon” rule, the design-critic skill, the IC-mode vs production-mode distinction. RDCO already explicitly pre-loads taste into the agent (SOUL.md, design-taste memories, founder voice match) before delegating creative work. This is the codified version of what Mitch is preaching to a less-built-out audience.
• Mistake 3 (don’t stay in your lane) ↔ Ray-as-COO doing finance, design, ops, deployment, and writing. The whole bet is that RDCO compounds because the agent is a generalist, not because the founder is.

The Zac O’Hara masonry case is a useful citation for the “Claude Code for non-tech SMBs” data dot. File it for Sanity Check / discovery purposes.

Hooks worth remembering

• “AI purple” — the bland middle-of-distribution voice that comes out when you outsource creative judgment. Useful pejorative.
• “Asking AI for your horoscope is below its current pay grade.” — concise way to articulate the “tasks not questions” frame.
• “Taste is the part of marketing nobody can teach you in a 2-week bootcamp.” — useful sentence for any anti-bootcamp / pro-portfolio essay.

Sales-funnel context (for transparency)

The email tail is a waitlist push for the Claude Code Marketing Bootcamp opening enrollment May 11 / starting May 18. Plus a free lead-magnet skill teaser. This newsletter run from ship30for30 has been heavy on the funnel, light on craft, for the past several weeks — this one earns its keep because the body content stands alone before the pitch.

Cross-references

• 2026-05-05-ship30for30-claude-code-marketing — yesterday’s email in the same Bootcamp drip sequence
• 2026-04-10-ship30for30-perfect-newsletter-frameworks
• feedback_design_taste_high_personality — RDCO’s own taste-not-template stance
• feedback_ic_vs_production_mode — the operational version of “don’t outsource taste”
• feedback_no_derivative_sanity_check_pieces — Sanity Check’s own rule against AI-purple takes

Decision

Status-only. File for vault, no founder action required. The “AI purple” phrase and the Zac O’Hara masonry case study are the two reusable bits.