How I use Claude Code as a Marketer

More and more people are waking up to the potential of AI. We’ve moved from a period of whether to adopt, into a period of how to adopt. You can clearly see this when you look at search volume for Claude Code, Anthropic’s terminal-based AI coding tool, which grew by roughly 20x in 12 months (DataForSEO via Google Ads ). Claude Code itself passed $2.5 billion in run-rate revenue in early 2026 (Yahoo Finance, citing Anthropic ). Many marketers like me (and you?) are part of that growth. This blog is my marketing perspective as part of a wider set of content that shows how Fifty Five and Five as an AI marketing agency uses Claude products across multiple roles.

I use Claude Code as a marketer by distilling what I already know in my brain into skills and context files. The skills run the deep work (research, drafting, and monitoring) at speeds I could never match manually. I focus on the strategic decisions and creative judgement that still demand a human.

A quick aside before we go further. If you’re new to any of this, three definitions will save you time.

There’s a more user-friendly alternative in Claude Cowork, though I’ve found it lagging behind on the more technical work that revolves around APIs.

The Claude Code skills I actually use as a marketer

This may sound strange coming from me, a marketer, but the truth is a large part of my career work (things like keyword research, account research, content writing, campaign performance, and so on) can ultimately be boiled down to repeatable steps. With the guidance and wisdom of my experience, those steps can be turned into knowledge files readable and executable by an AI tool. Manual processes can be automated through APIs.

Keyword research, for example. I used to log into Ahrefs to find keywords with promising data. Then take those keywords into a content brief. Then a draft. Then a CMS. Then an SEO checker. Each step a copy-paste, a context switch, a different login. AI can do the keyword research without me clicking anything; it hits the Ahrefs API directly. And because skills can be given access to multiple APIs at once, the same skill (or a chain of skills) can take a topic, find the keywords, draft the piece, and publish it without me leaving Claude Code. No more copying and pasting. No more dragging files between tools.

I can’t speak for everyone, but I’m not losing any sleep over the hours I used to spend clicking through the same UIs trying to stitch tools together. What I do care about is the review and scrutiny of those keywords to make sure they’re fit for purpose. There is still a role for both human and AI in almost every way, but the time should go to value-adding work and quality, not hours spent on busy work.

That’s where Claude Code skills come in. A skill is a markdown file (basically a simple, easily readable text file) that captures how I approach a task: the steps, the observations, and the tips and tricks. Pair a skill with context files (about your offer, your brand, and your clients) and you get an AI that approaches your work the way you would, just faster. I distil the process. AI executes. I then review the quality.

Anthropic published a blog post on how their own marketing teams use Claude . Their Influencer Marketing team uses scripts to free up over 100 hours a month. Product Marketing saves 5 to 10 hours per launch brief by using skills. And across marketing teams using agentic tools like Claude Code, repetitive strategic analysis (SEO audits, PPC reviews, the stuff that used to eat a Monday morning) drops by roughly 75%.

That’s enough about what skills are. Let me tell you about my favourite skills, the ones I’ve made and rely on every day:

• ABM account research skill: finds insights and signals across many accounts in hours, not weeks per account.
• Blog writing and editing skill: drafts posts following our SEO and GEO checklist (GEO meaning content tuned for AI engines like ChatGPT to cite, not just for Google to rank). This post is the output of that one.
• Author profiling skill: turns research and 1:1 interviews into a profile that makes AI content sound like the actual author, not a content mill.
• Social listening skill: scans LinkedIn and Reddit every morning for relevant conversations and drafts responses in the author’s voice.
• Content publishing skill: runs content through our SEO and GEO checks before it goes live, handling the technical metadata that used to take an editor a full afternoon.

Some of these run inside Claude Code. Others now run as scheduled tasks in Cowork. The rest of this piece goes deeper on three of them.

Claude Code ABM: from weeks of account research to hours

Seven or eight years ago (feels like a distant memory now) I worked at an agency on a print company client. They wanted to sell colour printers into retail for in-store signage. We had a list of target accounts and the maths worked on paper. So we went on foot, store to store, doing the research that would inform our ABM. Most of the work was figuring out which accounts actually deserved the time and resources. Tesco was one we expected to make the cut. At a Tesco we asked the right person, and they told us Tesco had just signed a five-year deal for black and white printers the previous year. They were never going to buy. That single piece of research was exactly what I needed to make a tough call: cut what would have been a dream customer. That’s the very essence of ABM.

On-foot research still has its value today, especially in retail. But AI has only enhanced my ability to understand accounts through research at a scale and speed I couldn’t match manually. The keen eye for connecting the signals and insights to what I’m trying to sell, and the story I want to weave with the account, that part is still mine.

I know what you’re thinking: ABM is just personalised email at scale and AI hype. But I’ve worked with ABM since 2019, and the thing people get backwards is this. The old way wasn’t slow; it was hard. Meaningful, insightful, contextual account research requires three things working together:

1. A strong understanding of your offer (what you actually sell, where it fits, and what its real value is).
2. A strong understanding of the account (industry, leadership, strategy, and the actual pain points, not the surface-level ones).
3. The intelligence to find insights and signals that bring the two together.

That third bit is what separates real ABM from personalised email. It’s where most efforts fall down. ABM has always been 80% setup and 20% execution, and the setup is where most people skip the work.

A Claude Code ABM skill unlocks the first two layers. You encode your offer (what you sell, the value, and the fit) and your research methodology (what to look for, where to look, and what good looks like) into context files. The skill runs the deep work across many accounts in hours, not weeks. The bridging intelligence (spotting which insight actually matters and how to use it) still belongs to a human. Same brain, richer raw material.

A concrete example I lean on: investor documents and annual reports. Find the leadership, what they care about, and the multi-year strategic plans. If you can align what you sell with something on their three-year horizon, that isn’t a cold pitch any more, it’s a relevant offer. That kind of insight-mining is baked into my account research skill, and it’s baked in because it’s baked into my soul. I’ve spent hours and hours reading those often verbose annual reports, and let me tell you, they’re not a light read. The skill now runs across a list of accounts overnight and gives me back a brief I can read with a coffee.

The proof of what’s possible when ABM research is actually done: at Northern Data Group we ran an AI-driven ABM campaign across 110 enterprise accounts. The result was a 39% email open rate (roughly 2x industry average), 200% ROI from one opportunity alone, and qualified meetings at seven targeted accounts (Northern Data Group case study ). That was already strong with a focused, human-led approach. With the research bottleneck broken, the same quality is reachable across many more accounts.

Worth being clear about what humans still own. The strategic frame (defining the offer, the ICP, and the target accounts) is best served by a workshop with the revenue team. The creative concept and outreach plan still benefits from human craft. And spotting which research output is actually valuable is its own skill. The AI gives me richer raw material; the judgement is still mine.

Claude Code blog writing: from topic to publish in an hour

This section is the one I’m writing about while you’re reading the output. Bit recursive, sorry. The skill that produced this post follows ten steps, all running inside a Claude Code conversation with my input where it matters:

1. Pick a topic. Usually from a recent client conversation or a persona we’re building for.
2. Develop a customer question someone might actually search.
3. Generate 5 to 10 question variants (how the same question gets asked different ways).
4. Run query fan-out across OpenAI and Gemini APIs to capture the actual sub-queries AI search engines fire when answering the question.
5. Run keyword research via DataForSEO or Ahrefs API, factoring in the sub-queries from step 4.
6. Find synergy keywords. The ones that appear in both the keyword data and the AI search sub-queries.
7. Pick the primary keyword (becomes the H1, this post’s H1) and five secondary keywords. Each secondary keyword becomes an H2, which is to say a section heading in the post. Look up at the H2s on this page; each one is anchored to a keyword.
8. The skill takes the H1 + H2 skeleton and runs external research for stats, examples, and supporting points.
9. The skill drafts a formal synopsis.
10. Synopsis review. I’ve been involved the whole way through, but this is where I really dig in. I interrogate the synopsis section by section. Is the angle right, does it sound like me, what would I add from lived experience, and what’s missing? This step alone often takes longer than the previous nine combined, and it’s the most important.

Anthropic’s Customer Marketing team has reported drafting case studies in 30 minutes with a similar workflow, down from 2.5 hours by hand (Anthropic blog ). Having worked with content writing teams for many years, I’d say 2.5 hours is generous.

The skill doesn’t stop at drafting. After the human review, the writing pass uses the synopsis, the author profile (more on those in a second), and our internal SEO and GEO checklist. The result enters the publishing flow, which is itself a Claude Code skill that runs the optimisation checklist over every blog before it goes live. Headings, meta descriptions, structured data, internal linking, image alt text, and the lot. SEO veterans already know this is checklist-shaped work. Tools like Yoast literally tell you what they’re looking for. The skill just runs the checklist consistently.

Claude Code marketing automation in practice: the publishing skill

Once a blog draft is finished, the publishing skill takes over. It runs the draft through the SEO and GEO checklist (the one our team built up over years of doing this work manually). It generates the technical metadata: meta descriptions, structured data, image alt text, and schema markup. It places relevant internal links to other content already on the site. Then it pushes a Hugo build to staging for review. The whole pipeline from “topic” to “published draft on staging” runs inside Claude Code with human oversight throughout, and the deepest sit-downs at the synopsis and the final draft. That’s what Claude Code marketing automation looks like in our content workflow. What used to be a content team’s full week now runs in an afternoon.

Now for the elephant in the room, and the bit I care about most. You may be wondering whether all this is exactly what the term “AI slop” was invented for, and whether running every blog through a Claude Code pipeline doesn’t risk producing exactly that. Fair question, and you’re right. People do use AI lazily, and it shows. You can spot it instantly. Generic openings, filler phrases, no specific examples or real opinions. That’s exactly why so much anti-AI sentiment exists, and the sentiment isn’t entirely wrong. AI is, separately, a staggeringly cool piece of technology. The problem is over-reliance, not the tool. Paired with clever people, AI can do great things. Just because something is coherent doesn’t mean it’s good. Giving a language model your brand strategy is like handing a parrot a TED Talk; it’ll repeat the words back to you with confidence, but the meaning’s gone.

My answer is two editorial mechanisms baked into the skill. The first is author profiles. A separate skill that uses research, internal documents, and 1:1 interviews to capture how an author actually speaks, their career, the problems they solve, and the things they would never say. The author profile is what makes AI content sound like a person rather than a model. The second is editorial review steps at two stages: at synopsis and at final review (both of which just happened to this post). The job at both stages is to find wooden, AI-style explanations and replace them with personable, lived experience. AI can write words. The hard bit is can AI write the right words. Iteration and feedback loops are how that happens.

Even Google has been explicit on this. They’ve stated they don’t care if you use AI, as long as what you create brings real value to searchers (Google Search Central ). The rule is quality, not provenance.

Claude Code social listening on LinkedIn and Reddit

Offsite engagement matters more now than it did two years ago because AI search engines lean heavily on places like LinkedIn and Reddit when they answer questions. Showing up in those conversations has compound value. Three benefits stacking together rather than three separate goals: part lead generation (rare but real intent opportunities), part brand awareness (your audience already lives there), and part LLM citations (the answers AI gives tomorrow are partially shaped by the conversations you join today).

I think about offsite the same way I think about ABM. At enterprise companies it’s typically a full-time role for one person: monitoring platforms, finding relevant conversations, and crafting responses. A Claude Code skill (then a scheduled Cowork task) turns the finding and drafting into a trivial morning job. The human spends their time on what actually matters: adding value to the conversation, not trawling for it.

What used to take days of research now takes hours. Subject matter experts can spend a morning a week, every fortnight, or once a month catching up on the conversation happening on platforms like LinkedIn and Reddit. They show up where they’re useful, not chained to a feed.

The set-up looks like this. A skill encodes what I’d do manually if I had infinite time: the topic areas to watch, the personas to target, what good engagement looks like, and what to ignore. The skill runs every morning. Two reports waiting before I open my laptop, one for LinkedIn and one for Reddit. Each report lists relevant conversations and drafts responses using the author profile so the voice is mine. I review, edit, post. Or kill the response if it wouldn’t actually add value.

The thing that makes this possible, and the bit I think most people are missing, is browser control. Claude Code can drive a real Chrome browser. There is no proper LinkedIn API for monitoring like this, and Reddit’s is restrictive. Browser control sidesteps both. The skill opens Chrome, logs in, scrolls the feeds, clicks into threads, reads the comments, and pulls back what matters. The same trick works for almost any web tool: if a human can use it in a browser, the skill can use it too. That’s what unlocks offsite engagement at scale, and it’s what unlocks a lot more besides — any platform without an API is suddenly back on the table.

What does Claude Code on Reddit look like in practice?

Reddit behaves differently from LinkedIn. Anonymous, topic-led communities, sceptical readers who can spot a salesperson instantly. The skill knows that. It scans a list of subreddits relevant to the personas we serve, identifies threads where someone is actually asking a relevant question, and drafts a response that reads as a genuine contribution and not a pitch. The response includes specifics and names tools (sometimes including ours, sometimes including competitors when they’re genuinely a better fit). Most subreddits don’t allow external links anyway, so what matters is whether the contribution is genuinely useful, not whether it drives traffic. I review every response before it posts, because Reddit will downvote even the best-intentioned brand voice into oblivion if the contribution isn’t real.

We built a version of this process for Avalara, scanning Reddit, LinkedIn, Quora, and Medium for relevant conversations and generating responses from key people in their business (Avalara ). The skill finds the right topics for the right author, finds the conversation, and drafts the response. The author then reviews and posts.

Occasionally a conversation comes through that could turn into real business if you bring genuine insight. I’ve seen threads asking for advice on intent data tools, given my lived experience of how those tools actually perform, and pointed people toward a different way of getting intent signals for ABM altogether. That kind of contribution opens doors.

The compound value layers in on top of those moments. Brand awareness in the right communities, and the LLM citation flywheel that closes when those conversations get indexed and cited later. Offsite engagement breeds citations breeds inbound visibility.

The author profile shows up here again, this time driving response generation. It’s the connective tissue across content and offsite, the same skill in different applications. Whether I’m writing a blog or a Reddit comment, the voice comes from the same source.

This is still very much a focus of mine and something I want to crack properly. The hard bit isn’t whether AI can write the words. It’s whether AI can write the right words. That requires iteration and feedback loops on the author profile until the responses sound credibly like the person they belong to. We’re not far off being able to automate the posting itself, once the profile’s good enough that I trust the response without rewriting it. Not there yet.

Building your own: a Claude Code skills tutorial

The best Claude Code skills marketers are building share a pattern. They start with a process that’s already painful, repeated, and well-understood. Then they get encoded. Not all at once, not perfectly, but as a working draft that gets sharper with use. Anthropic’s Product Marketing team reported saving 5 to 10 hours per launch brief by encoding the process into a skill (Anthropic blog ).

Here’s the short version of a Claude Code skills tutorial that’s worked for me.

The build flow I’d recommend:

1. Pick one repeated, painful task you do regularly. Content briefing, account research, social monitoring, or whatever’s eating your week.
2. Describe the process you’d run manually, step by step, in plain English.
3. Capture the inputs (where do you get the source material) and the outputs (what does a finished version look like).
4. Write the steps as a skill file. It’s a plain text file, nothing fancy.
5. Run it on a real example. Watch where it gets things wrong.
6. Iterate. Add examples, clarify the steps, and define the outputs more precisely. Use the skill in production.

Three things I’d tell anyone starting:

• Don’t be scared to embrace the terminal. It’s just like any other chat interface, and you get to pretend to be coding like in the movies.
• Surround yourself with at least one technical person who can help bring your skills to life through APIs. The marketing-side work is encoded in the skill; the API integration is where the technical layer matters.
• Get in the habit of meticulously writing down the steps of how you work, so you can skillify them. The skill file is only as good as the documented process behind it.

That third point comes from a lesson I learned the hard way. I built a tool in Claude Code that I thought was working. The outputs looked reasonable, the responses were coherent. Turned out the tool was hallucinating because it didn’t actually have API access to do the tasks I’d assumed it could. Someone had to point it out to me. Crashed and burned. That was a proper wake-up call. I’m a marketer, not a developer, and I’d made assumptions about what the technology could do without really understanding the technical layer underneath. Since then I’ve gone deeper on the technical side and lean heavily on the technical people around me for what’s still over my head.

The lesson is broader than that one tool. If you’re building on AI, you need to understand the layer underneath, or surround yourself with someone who does. That’s why the second piece of advice above matters more than it sounds.

The skills I run today all started as manual processes. The author profile skill began as a list of interview questions I’d email SMEs. The offsite engagement skill began as a spreadsheet of subreddits and LinkedIn searches. The blog skill (the one writing this) began as a Google Doc template. Each one was something I’d worked out by doing it dozens of times. Once I’d written the steps down properly, turning them into a skill was the easy part. That’s the order: manual first, written down second, skill last.

Where this leaves us

I use Claude Code as a marketer by distilling the parts of my expertise that can be written down (research methodologies, writing checklists, and social listening processes) into skills and context files. The skills run that work across ABM, content, and offsite engagement at speeds a small team couldn’t match manually. The judgement, creative, and strategic decisions stay with me, where they belong.

The structure of this post is itself proof of the loop. The H1 came from a primary keyword a skill picked. Each H2 above is anchored to a secondary keyword. The stats came from external research a skill ran. The synopsis got reviewed by a human, twice. That’s not theory, it’s the workflow that produced what you’ve just read. The next person who searches an AI engine for “how to use Claude Code as a marketer” might get an answer extracted from this conclusion. That’s the loop closing.

The lane is still wide open right now. Early movers compound. The limitations that used to hold this back are being lifted, and I can see it happening in the work I do every day.

My colleague Chris talks about using Claude Code for partner marketing, and Fergus for design work. Both in the wider piece on running a marketing agency on Claude Code . If you’re trying to use Claude Code as a marketer and want a hand building something like this, drop me a line on LinkedIn .