# Claude for marketing: 7 workflows founders can run

> Use Claude for repeated marketing work with clear inputs: customer language, positioning drafts, page audits, channel variants, comparison checks and report notes. Review everything before it publishes.

Author: Deb Pradhan
Published: 2026-11-07
URL: https://100xlab.com/blog/claude-for-marketing
Tags: AI agents, Marketing engineering

Claude is useful for marketing when the job repeats and the input is clear: sorting customer language, drafting from a brief, auditing a page against a checklist, adapting one message to several channels. A founder should hand it that work, then read, edit and approve everything before it goes out.

This post gives seven workflows you can run this week. Each has the input to supply, a prompt you can adapt and the check a person makes at the end.

## What is Claude good at in marketing?

Work with a defined input and a result you can check. Claude reads long material quickly, follows a checklist, rewrites in a set voice and keeps the same structure across many pieces.

It is weaker at the parts that need judgement and knowledge of your market: choosing a position, deciding which claim you can stand behind, and noticing that a competitor changed their price last week. It will also produce confident text from thin inputs. That is the main risk, and it explains the pattern behind all seven workflows: give it your material, tell it to stay inside that material, and check the output.

| Good fit | Poor fit |
|---|---|
| Sorting call notes into themes | Deciding your positioning |
| Drafting from a written brief | Writing claims with no source |
| Auditing a page against a checklist | Publishing without a read-through |
| Adapting one message to several channels | Judging whether a number is real |
| Summarising an analytics export | Choosing what to do about it |

Our wider view of where the line sits is in [AI agents in marketing: where a person signs off](/blog/ai-agents-in-marketing-with-human-approval).

## What do you need before you start?

Four things, kept in one place so every workflow starts from the same facts:

1. **A one-page brief.** Who you sell to, what you sell, what it costs, what it does not do.
2. **Voice rules.** Words you use, words you avoid, sentence length, spelling variant. If you have no brand guidelines yet, [brand as a system](/blog/brand-as-a-system) explains what to write down.
3. **Approved claims.** A short list of statements you can defend, each with its source.
4. **Real customer words.** Quotes from calls, emails, reviews and support tickets.

Keep these as plain text. Paste them in, or attach them in a Claude project if you use one, so you do not retype them. Updating the file once then improves every later run.

## Workflow 1: How do you turn customer calls into a language bank?

**Input:** transcripts or notes from five to ten calls, with names removed.

**Ask for:** the exact phrases buyers use for the problem, the alternatives they tried, their objections and the words they use for the result they want.

> Here are notes from customer calls. Extract (a) phrases the buyer used to describe the problem, quoted exactly, (b) tools or methods they tried before, (c) objections, (d) words they used for the outcome. Quote only what is in the notes. Group by theme and mark any theme that appears in only one call.

**Check:** open three quotes at random and find them in the source. Buyers now explain their budget, their current tool and what went wrong last time in full sentences, so these phrases belong in your headlines, ads and landing pages.

## Workflow 2: How do you draft positioning and test it?

**Input:** your brief, the language bank from workflow 1 and a list of two or three alternatives buyers consider.

**Ask for:** three positioning statements in a fixed format, then the three hardest objections a sceptical buyer would raise against each.

> Using only the brief and the language bank, write three positioning statements: for [buyer], who [problem], [product] is [category] that [outcome], unlike [alternative], because [reason to believe]. List the reason to believe from the material only. Then give the three strongest objections to each statement.

**Check:** you choose. Claude can propose and attack, but the choice is the decision that matters most in your marketing, so keep it yours. Where logic cannot pick a winner, test the options with real buyers.

## Workflow 3: How do you draft content without it sounding generic?

**Input:** the brief, voice rules, approved claims and one piece you wrote that sounds like you.

**Ask for:** an outline first, then a draft. Reviewing an outline takes two minutes and saves a rewrite.

> Write an outline for a post answering "[query]" for [reader]. Put the direct answer in the first two sentences. Use only the approved claims for any fact about our product. Where a fact is missing, write [NEEDED] instead of guessing. Follow the voice rules.

**Check:** search the draft for [NEEDED], numbers and named studies. Every number needs a source you opened yourself. Remove anything you cannot source.

## Workflow 4: How do you audit a page for clarity?

**Input:** the page text and a checklist.

| Checklist item | What to look for |
|---|---|
| Answer up front | Does the first paragraph say what this is and who it is for? |
| One job | Does the page ask for one action? |
| Proof | Is each claim backed by something a reader can see? |
| Objections | Is the likely objection answered on the page? |
| Quotable | Could an assistant lift one clear sentence as an answer? |

The last row matters because every page now has two readers: the person and the AI assistant they ask. Pages get quoted for their shape as much as their subject. Clear structure, an answer in the first 30 words and a comparison as a table all help. If you want the search side in detail, read [SEO vs GEO](/blog/seo-vs-geo).

> Score this page against the checklist, pass or fail per row. For each fail, quote the sentence and suggest one rewrite. Do not add claims that are not already on the page.

**Check:** a person reads the suggested rewrites against the approved claims. Reject any rewrite that adds a promise.

## Workflow 5: How do you adapt one launch story to every channel?

**Input:** the launch story in one paragraph, the audience for each channel and the format limits.

**Ask for:** a LinkedIn post, an email, three short social posts and a headline set, all from the same story and the same claims.

> Adapt the launch story below for LinkedIn, a customer email and a short social post. Keep every claim identical to the story. Change tone and length only. Return a table: channel, text, word count.

**Check:** compare the claims across channels. The common failure is drift, where each version quietly changes the promise. Because a system repeats a bad decision as faithfully as a good one, check the first run closely and reuse the prompt only after it is right.

## Workflow 6: How do you check a comparison page against its sources?

Comparison pages go stale. One forgotten page can keep a retired price or feature alive in assistant answers for months.

**Input:** the page text and the current pages of each product you compare.

> Compare the claims on our comparison page with the competitor pages pasted below. List each claim, whether the source supports it, and the sentence in the source. Mark anything you cannot find as UNVERIFIED. Do not use outside knowledge.

**Check:** open every source yourself. Claude only compares what you paste, so a stale source gives you a confident wrong answer. Date-stamp the page after you check it.

## Workflow 7: How do you turn an analytics export into a weekly note?

**Input:** a CSV export from your analytics tool for two periods, plus your list of key events. If tracking is not set up yet, do that first: [analytics before ads](/blog/analytics-before-ads) covers the event plan.

**Ask for:** what changed, what did not, and what to check next. Not what to do.

> Compare the two periods. List the five largest changes in sessions, key events and conversion rate by source. Say which changes are too small to trust given the volumes. Suggest three things to investigate. Do not recommend budget changes.

**Check:** spot-check three figures against the tool. Then you decide what happens next. Reading the numbers and choosing the change is judgement, and it stays with a person.

## How do you turn these into repeatable systems?

Run each workflow by hand twice. When the output is right, save the prompt, the inputs and the checklist together. That bundle is a skill: a written instruction file an AI agent can load for a repeatable job, so the steps stay the same each run. You can browse examples in our [agent skills directory](/resources/agent-skills).

Then add the approval step to the routine rather than leaving it to memory:

- **A queue.** Drafts land in one place, and a person approves before anything publishes.
- **A source file.** Claims live in one list with their sources, updated in one place.
- **A dated review.** Put a date on the brief and the claims, and re-check them when it passes.

The model behind this is covered in [what marketing engineering is](/blog/what-is-marketing-engineering): repeated work becomes a system, measured and improved, with people on the decisions.

## Where does 100xLab fit?

If you would rather have these workflows built and wired into your site, tracking and content than run them yourself, that is what we do. 100xLab builds marketing as systems at a fixed price, with AI agents inside each area and a person approving the output. See [what we build](/services) and how [Sprint, Launch and Partner](/pricing) work. If you can run the seven above on your own, you may not need us yet.

## FAQ

### Can Claude replace a marketing hire?

No. It takes drafting, checking and sorting work. A person still decides who you are for, what you claim and what goes live.

### What should I give Claude before asking for marketing copy?

Your brief, your brand voice rules, a list of approved claims and a few real customer quotes. Without them it writes fluent, generic copy.

### How do I stop Claude inventing facts or numbers?

Tell it to use only the material you supplied, to mark anything missing as unknown, and to cite where each claim came from. Then check the claims yourself.

### Should I publish Claude's drafts as they are?

No. Edit every draft, check every claim and approve it yourself. A weak rule or a wrong fact gets repeated every time you reuse the prompt.

### Is a saved prompt the same as an agent skill?

They are close. A skill is a written instruction file an AI agent can load for a repeatable job, so the steps stay the same each run.
