Claude prompts for marketing

Which Claude Prompts Should Marketers Start With?

Start with the four that shape everything downstream. Positioning, audience, the competitive picture, and the campaign brief are the inputs every other asset inherits, so writing them down once pays into every ad, email, and page you produce afterward.

• Positioning statement: feed it your product page, the three closest alternatives, and five customer quotes. Ask for a positioning statement in the classic frame (for whom, in what category, the key benefit, the reason to believe), plus two alternative framings and what each one gives up. The tradeoffs are the output you want.
• ICP sketch: give it your closed-won list with company size, industry, and use case, alongside the closed-lost list. Ask for the pattern that separates the two, written as a one-page profile: firmographics, the trigger event, the title that signs, and the objection that kills the deal.
• Competitor teardown from public pages: point it at three competitor homepages, pricing pages, and docs. Ask for their claimed positioning in their own words, the audience their copy is written for, what the pricing page implies about their sales motion, and the gap they leave open for you.
• Campaign brief: hand it the goal, the audience, the offer, the channels, and the one constraint that binds (budget, date, or legal review). Ask for a one-page brief with the message hierarchy, the assets required per channel, and the measurement plan, then argue with it about the message.

The pattern in all four: you supply the raw material, and the prompt asks for structure plus alternatives. A prompt that returns one confident answer is less useful than one that returns two options and the tradeoff between them, because choosing is the part of the job you are paid for.

Run each of these once, then keep the outputs as files rather than as chat history. Positioning, the ICP, the competitive picture, and the campaign brief are the context every later prompt needs, so saving them turns four good afternoons into permanent inputs for everything you write afterward.

Which Prompts Carry the Weekly Execution Work?

The second set covers the work that repeats. These are the prompts that eventually deserve to become saved skills, because you run them on a rhythm rather than once.

• Ad variant grid: give it the offer, the audience, the platform character limits, and ten past ads with their click-through rates. Ask for a grid of variants across two axes you name, such as hook type by benefit, with the platform constraints enforced in every cell. Fifty on-spec variants take minutes; picking the three worth testing stays your call.
• Email sequence: supply the trigger, the state the reader is in, the offer, and your two best-performing past emails as voice examples. Ask for a five-email sequence with the job of each email named, three subject-line options apiece, and the exit condition that stops the sequence.
• SEO content brief: give it the target query, the current top results, your product angle, and your existing internal links. Ask for the search intent, the sections a strong answer needs, the questions to answer directly, and the internal links to place.
• Repurposing a long piece: feed it the transcript or the article plus your channel rules. Ask first for the five ideas worth extracting, then for the assets: a LinkedIn post per idea, an email, and a short-form script, each written to its channel rather than trimmed to fit.
• Landing page teardown: point it at the page and the ad that drives traffic to it. Ask where the message breaks between the click and the headline, which section carries the main objection, and three specific rewrites with the reason each one might win.
• Weekly performance readout: hand it last week readout plus this week exports from Google Analytics 4, the ad platforms, and the CRM. Ask for what moved, what it likely means, what to verify before believing it, and the one decision the numbers support this week.

Each of these has the same shape. Inputs from your own business, constraints written down, examples of the output you want, and a named format for what comes back. Swap in your material and they hold up across products and categories.

One upgrade applies to the whole execution set: run them against files rather than pasted text. A prompt that reads your ads export directly can rank the ten worst performers before it writes a single variant, and a repurposing prompt that reads the transcript keeps the speaker phrasing instead of your summary of it. Assembling the inputs by hand is most of the friction that leaves a good prompt unused after the second week.

How Do You Make Any Prompt Better?

Four upgrades work on every prompt above, and they matter more than the wording. A plain request with strong context beats a beautifully crafted prompt with none.

• Context files: put your positioning, brand voice rules, ICP, and product facts into files the agent reads. In Claude Code, a CLAUDE.md file at the project root loads at the start of every session, so that context arrives without you pasting it again.
• Examples of good and bad: three pieces of copy that landed and two that missed teach voice faster than a paragraph of adjectives about your brand ever will.
• Constraints stated up front: character limits, banned words, claims that need legal review, reading level, channel rules. Constraints written down once stop being feedback you give on every draft.
• A named output format: say what comes back and where it goes. A table with these columns. A markdown file at this path. A CSV shaped for the ads editor. Formatting is the difference between output you use and output you reformat.

Then iterate on the brief rather than the draft. When output misses, the instinct is to fix the paragraph in front of you. The move that compounds is to fix the instructions so the next twenty drafts start closer to right, which is also the moment a prompt starts turning into a system.

Why Do Prompt Libraries Decay?

A prompt library decays because it lives outside the work. The doc drifts from the product, the version in your notes is two revisions behind the version you keep pasting, and nobody else on the team knows which one is current. Six months in, you have a folder of good ideas and no working process.

Saving the prompt as a skill puts the instructions where the work happens. A skill is a folder with a SKILL.md file holding the instructions for one job, plus any reference files or scripts that job needs. Claude Code loads it when the work matches, or you call it by name with a slash command, and project skills live in a .claude/skills folder so they can be committed to version control and shared with everyone on the team (Agent Skills).

The practical difference shows up in the details you stop retyping.

• The context is attached: the skill carries your brand rules and product facts, so the twentieth run starts from the same ground as the first.
• The constraints stay enforced: character limits and banned claims live in the file rather than in your memory of what you asked for last time.
• Improvements accumulate: when a run misses, you edit the skill, and every future run inherits the fix.
• The team runs the same thing: a new hire clones the repo and produces work that matches your standard in week one.

That is the turn worth making with any prompt you have used more than three times. Claude Code marketing skills covers how to write one, and how to use Claude Code for marketing shows the workflow that surrounds it. If you are weighing the browser version against the terminal agent, Claude Code vs Claude chat lays out where each one fits.

What Do These Prompts Get Wrong?

Prompts produce drafts and options fast enough that skipping the check feels reasonable. The failure modes are consistent, which makes them easy to guard against.

• Confident wrong facts: a model will state a competitor price or a product capability that was true last year, or never was. Verify anything you would be embarrassed to publish.
• Statistics with no source: asking for supporting data invites invented numbers. Ask for sources you can open, then open them before the figure travels.
• Voice that averages out: without examples, output drifts toward the middle of the internet. Your voice comes from the examples you attach.
• Strategy dressed as analysis: a well-structured positioning doc can be confidently wrong about the market. Customer conversations settle that question. A prompt cannot.
• Brand and legal exposure: claims, competitive comparisons, and regulated language need a named human signing off every time.

One habit covers most of this: ask for the sources and the uncertainty in the same request. Have the prompt mark how confident it is in each claim and name what would change the answer. That single instruction turns a draft you have to audit line by line into a draft that tells you where to look first.

The working pattern is the same one every experienced operator lands on. The agent produces the volume and holds the constraints; you keep the judgment, the taste, and the accountability. Speed on drafting is worth having only when the review still happens.

Where Does the Workshop Fit?

You can take the ten prompts above, run them today, and get value out of the afternoon. Turning them into systems your team runs without you is a different exercise, and it goes faster with someone who has already built the version that works.

Claude Code for Marketers (CC4M) is a live workshop where marketers build go-to-market systems in their own repo and keep them. It is taught by Hank Taylor and Mitchell Wright, operators who have built GTM at GitLab, Vercel, Laravel, Neo4j, and ClickHouse. It runs as one live five-hour session, followed by two office-hours sessions. In it you build four working systems: a UTM Builder, a Content Engine, an AI SDR and outbound flow, and Automated Ad Generation. No coding background is assumed, and you leave with the repo. Private and team cohorts are available on request. For the wider picture of what the tool does, start with Claude Code for marketers.

Questions

Frequently asked