Your first content engine: write one useful thing a week, do not automate the factory yet

You have a handful of paying customers. Word of mouth is trickling in, and you have one acquisition channel you can just about repeat. Then you read the advice: spin up an AI content pipeline, batch-generate two or three SEO articles a week from a keyword list, auto-publish, and syndicate everywhere. In a weekend you can have a content factory.

Don't. Not yet.

At 3-30 customers, the content factory is the wrong first move. Auto-spun articles rank for almost nothing when your domain is new, they build no trust with the few readers you get, and worst of all they hide the one content asset you actually have: the conversations you are already having with real customers. Your first content engine is not a pipeline. It is you, writing one genuinely useful thing a week, using Claude Code for the boring scaffolding and nothing else.

Here is how to build it.

Why "automate the factory" is the wrong first move

The pitch is seductive because it sounds like leverage. Configure a pipeline once, feed it competitor URLs and keywords, and it prints articles while you sleep. The problem is that content has two jobs, and automation only pretends to do the easy one.

The first job is to exist and be findable. A brand-new domain has almost no authority, so even perfect articles take weeks to surface. Look at the pattern on this very blog: a UK long-tail post routinely takes 14 to 30 days just to draw its first impression in Search Console. Volume does not shortcut that. Publishing forty thin articles in month one does not get you indexed forty times faster. It gets you forty pages Google crawls, shrugs at, and deprioritises.

The second job is the one that actually converts: the article has to be worth reading, from someone worth trusting, about a problem the reader genuinely has. Auto-generated content fails this quietly. It reads like every other auto-generated article because it was trained to. It has no specific number, no real customer story, no opinion it would defend. At 3-30 customers your bottleneck was never how many words you could produce. It was having something worth saying. Automating production before you have that just industrialises the emptiness.

So the honest version of the advice is: automation is a scaling tool, and you have nothing to scale yet.

Your conversations are the brief

The good news is you are already generating the raw material for great content. You just are not writing it down as content.

Every support ticket is a post. If a customer had to ask, the answer was not obvious, and if it was not obvious to them it is not obvious to the next hundred people searching for the same thing. Your support workflow (the one you keep manual on purpose) is a content backlog in disguise. Tag your tickets by theme for two weeks and the top three themes are your first three posts.

Every "why did you almost cancel" answer is a post. When you built your churn-save flow you started collecting the real reasons people hesitate. "I could not tell if it did X." "I did not understand the pricing." Those are not just product fixes. Each one is an article that meets a doubting reader exactly where they are.

Every question your first ten customers asked before they paid is a post. You answered those questions on calls and in DMs. Write the answer down once, publicly, and it works for you for years.

This is the difference between a factory and an engine. A factory needs fuel you do not have. An engine runs on the by-product you are already producing. Mine the conversations, and you will never stare at a blank page wondering what to write.

One channel, one format, committed for six weeks

You already learned this lesson with acquisition: run one channel at a time, committed for weeks, rather than six channels badly. Content is the same discipline.

Pick the single channel where your buyers already gather, and match it to who you sell to. If you sell to UK developers and technical founders, written search-first content and the occasional technical community post is your lane, because those people search for answers and read docs. If you sell to UK trades, freelancers, or non-technical operators, a niche community, forum, or a focused newsletter usually beats a blog nobody has found yet, because that audience gathers in specific rooms rather than in Google.

Pick the format you can sustain. One long, genuinely useful post a week beats five thin ones. Consistency over six months is the single biggest unlock in early content, and consistency dies the moment your target is unrealistic. Do not promise yourself a daily cadence you will abandon in a fortnight. Promise one good post a week and actually ship it.

And resist the urge to syndicate to Medium, a dev community, LinkedIn, and three others on day one. Cross-posting the same piece to six platforms is not distribution, it is dilution, and it splits your ranking signal. Publish on your own domain first, let it be the canonical home, and share it deliberately in the one place your buyers actually are.

Where Claude Code actually helps: scaffolding, not authorship

Here is the line that matters. Use Claude Code to remove the friction around writing, never to do the writing. The moment the prose is not yours, the trust and the specifics evaporate, and you are back to the factory problem.

Three genuinely useful, human-in-the-loop helpers you can build in an afternoon:

A draft-from-your-notes assistant. You take a real ticket or call note and dictate the argument out loud: here is the question, here is what I actually told them, here is the nuance. Claude Code turns your rambling notes into a structured outline with headings and a logical flow. Then you write the prose in your own voice against that skeleton. The AI does the structuring; you do the saying. Keep the customer detail, the specific number, the opinion. That is the part that ranks and converts.

An internal-link checker. As your corpus grows, every new post should link to your relevant existing posts and to the product. Point Claude Code at your content directory and have it suggest, for a given draft, which existing posts to link and where, plus a natural spot to mention the product. Internal linking is one of the highest-leverage on-page moves for a small site, and it is exactly the tedious work you will skip if it is manual.

A one-command publish pipeline. Wire a single command that takes your finished markdown, posts it to your blog API, and then resubmits your sitemap to Search Console. The sitemap resubmission is the real discovery lever for article content. Do not waste time on the Indexing API for blog posts: it officially only supports job postings and broadcast events, it returns a cheerful 200 for everything else, and Google ignores it for ordinary articles. A resubmitted sitemap is what actually tells Google to come and look.

Notice what none of these do: none of them write the article, and none of them auto-publish without you. Human-in-the-loop is the whole point. Claude Code is your production assistant, not your ghostwriter.

For the record, the blog you are reading is exactly this engine running. One useful post, mined from a real build arc, drafted against notes, checked for internal links, and shipped through a one-command Claude Code pipeline. No factory. Just the engine, running most days.

Measure the right thing (and be patient)

Content punishes vanity metrics harder than almost any channel, because the feedback loop is long. If you measure the wrong thing you will either quit too early or celebrate noise.

Do not measure posts shipped. Output is not outcome. Do not measure traffic spikes from a single share, which fade in 48 hours and tell you nothing about the compounding asset.

Measure signups and paying customers you can attribute to content, over an 8 to 12 week window. Instrument it the same way you instrumented acquisition: capture the source on signup into a simple field in your database, so you can group signups by where they came from. Then judge content on paying customers it eventually brings, not on the applause of any single week.

And set your expectations to the real timeline. SEO compounds slowly and quietly. A UK long-tail post can sit at zero impressions for three or four weeks before it surfaces at all, then climb for months. That is not failure, that is the shape of the channel. The founders who win at content are not the ones who wrote the most in month one. They are the ones still publishing in month seven.

When to stop, and when to graduate

The whole point of doing this manually is that manual reps teach you what a factory never could: which topics actually pull, which formats you can sustain, and what your readers really ask. That knowledge is the thing you are building, more than any single post.

So at 3-30 customers, deliberately do NOT: hire a content writer, sign with an agency, or switch on a full auto-generation pipeline. Every one of those is premature. They scale a process you have not proven, and they cost money and voice you cannot spare.

Graduate around 100 customers, and only with a documented, proven topic engine in hand: a list of themes that reliably draw your buyers, a format that converts, and a repeatable production process you could hand to someone else on Monday and expect the same result. That is the exact test from the hire-or-tool decision: buy capacity only for a recurring job you can specify in a sentence. "Write more of the posts that already work, in the format that already works" is that sentence. Until you can write it truthfully, keep doing the reps yourself.

Write one useful thing a week. Mine your own conversations for what it should be. Let Claude Code carry the boring parts. And give it the months it needs to compound.