Weekend project: set up a self-running company blog

First, the honest definition, because "self-running blog" gets sold two ways and one of them is a trap. The dishonest version publishes machine output on a timer with nobody reading it before the internet does. That experiment has been run at scale, and its results are on the public record: Google's spam rules explicitly target mass-produced pages "no matter whether content is produced through automation, human efforts, or some combination", and the 2024 enforcement wave erased sites built exactly that way.

The honest version automates everything that isn't judgment, drafting, formatting, publishing, distribution, measurement plumbing, and concentrates the judgment into a small standing loop, about two hours a week. The machinery self-runs; the editorial brain stays yours, at one-tenth the old time cost. That system is genuinely buildable in a weekend, and this is the build plan.

What you're assembling

Six parts, most of which you may partly have: a website you can publish to; measurement (Search Console plus GA4, connected, with your conversion defined); an AI drafting workflow with a standing brief; a prioritized topic backlog with your raw material attached; publishing and distribution automation; and a written quality gate. Saturday builds the plumbing and the content system. Sunday builds the backlog and the automation, and installs the loop. Clear the actual weekend: each half-day block is two to three focused hours.

Saturday morning: the plumbing

Start with measurement, because a blog that can't tell you what's working isn't self-running, it's self-hiding.

Verify your site in Search Console, submit the sitemap, and confirm your existing pages are indexed. Set up GA4 if it isn't, then define your conversion as a key event, Google's term for "an event that measures an action that's particularly important to the success of your business": the form submit, the call click, the signup. The conversion setup is the morning's core task and the single highest-value hour of the weekend, because every future decision leans on it. Link Search Console to GA4 so query data and behavior data land in one place. Finally, adopt one three-line rule for links you'll share in newsletters and social posts, lowercase UTM tags, standard media, so campaign traffic stays legible forever; GA4's own docs warn that tagging is case sensitive, and inconsistency here is the classic way small businesses shred their own data.

By lunch: the blog can measure itself. Most "self-running" setups skip this and automate blindness.

Saturday afternoon: the content system

Now build the thing that makes machine drafting produce *your* content instead of the internet's average.

Write the identity sheet. One page, stored where you'll draft: what the business does, for whom, where; your services and honest price ranges; the five opinions you hold that competitors don't; three short paragraphs you wrote yourself as voice samples. This sheet is the difference between a model that channels you and a model that channels everyone, and it feeds the brief that makes AI drafts worth editing.

Build the standing brief template. A reusable prompt scaffold: audience, search intent, the raw material for this post (facts, story, numbers), voice sample attached, plus your banned list, the phrases and structures you never want ("in today's fast-paced world", bullet-spam, summary conclusions). Test it end to end this afternoon: pick one topic you know cold, brief it, generate, and do the full edit-and-verify pass yourself. Producing post one today matters, both because the template always needs a first-contact revision, and because the weekend should end with something published.

Write the quality gate on an actual card. The Friday-gate rules the future you will follow when busy: every number, name, and claim checked against a source you'd link (non-negotiable, models fabricate at documented rates, one study caught GPT-4 inventing 28.6% of requested references); at least one thing in every post that only your business could say; sounds-like-me read-aloud pass; internal links in and out. Four lines. The gate is what the 7%-publish-unrevised statistic is really about: practitioners overwhelmingly keep a human check, because the blends are what rank while barely-edited AI pages measurably underperform.

Sunday morning: the backlog

A self-running system's scarcest input is decided topics, so spend the morning deciding a quarter's worth at once.

Inventory the questions your buyers actually ask: every question from sales conversations, every "how much / how long / what's the difference" you answer weekly, every comparison customers weigh, the searches made in the week before money changes hands. Sanity-check phrasing and volume against real query data, then rank the survivors buyer-first: pricing and cost questions, comparisons, service-specific explainers before general interest. Aim for a backlog of about 25 topics, and, the step that pays all quarter, jot each one's raw material now, while it's cheap: the price range, the anecdote, the contrarian take. Two lines per topic. Future drafting sessions start at 60% done because Sunday-you did the remembering.

Sunday afternoon: automation and installation

Wire the machinery around the system. Scheduling, so posts publish on their own cadence. Sitemap auto-updates and index pings on publish. Auto-reposting of published pieces to your social profiles, tagged per the morning's UTM rule. If your tooling supports it, automated draft generation from queued briefs, the deeper end-to-end pipeline can grow later; the weekend version just needs publish-day mechanics to be zero-touch.

Then install the human loop as an actual calendar object, because a system without a standing appointment is a good intention: one recurring two-hour block weekly. Its agenda, forever: ten minutes of data (what moved in Search Console, any conversions touching content), pull the next topic or two from the backlog, brief-generate-edit, run the card's gate, schedule. One post a week at high quality is the sustainable default; two in a good week. Add a monthly half-hour to the block for the owner's review: which posts produced inquiries, what to refresh, what the backlog needs next.

Before dinner: post one is live, posts two and three are briefed, the quarter is decided, and everything mechanical happens without you.

The three ways the weekend goes wrong

Forewarned is finished, so here are the derailers that claim most attempts at this project.

The replatforming rabbit hole. Saturday morning, you decide the real first step is a better website. It isn't. The system builds fine on whatever you have, and CMS migration is a separate multi-weekend project wearing this one's clothes. Rule: this weekend touches settings and content, never platforms.

The perfect-tool detour. Two hours comparing drafting tools is two hours the backlog didn't get. Pick something reasonable Saturday afternoon, if it disappoints in a month, switching costs an afternoon, because everything durable you're building this weekend (the identity sheet, the brief template, the backlog, the measurement) travels with you. The system is tool-agnostic by design; act like it.

The zero-post weekend. It's possible to build all the machinery and publish nothing, and machinery without a shipped post has a way of staying machinery. The Saturday-afternoon test post is non-negotiable for exactly this reason: the weekend's definition of done includes one live URL.

And one derailer for the following month rather than the weekend itself: the quiet death of the calendar block. The first time the weekly two hours gets sacrificed to something urgent, nothing visibly breaks, which is what makes the second sacrifice easier. Treat the block like a customer appointment, moveable, never cancelable, because the whole architecture assumes the judgment loop actually runs. Everything else recovers from a skipped week; the habit doesn't recover from a skipped month.

What Monday-you must remember

Set expectations against the real clock. The system starts producing *pages* immediately and *rankings* on the ordinary schedule: indexing and impressions first, meaningful clicks typically in months four to eight, compounding after. The weekly loop's data minutes exist so you can watch the leading indicators move while the revenue lags, the difference between early and failing.

And guard the one boundary that keeps the whole thing on the right side of history: the system's throughput is capped by your judgment hours, on purpose. The temptation, three months in, will be to queue twenty auto-drafted posts and skip the gate, converting your asset into exactly the volume-without-value pattern the spam policies name. The two-hour loop isn't the system's residue. It's the system's moat, and it's also just two hours.

Quick answers before you start

Do I need to buy anything before the weekend? Only a drafting tool if you don't have one, and even a general AI assistant works for version one. Everything else in the build, Search Console, GA4, your CMS's scheduling, is free or already yours.

What if I genuinely can't write, even edits? The system still works with the judgment redistributed: you supply the raw material and the fact-check (nobody else can), and a freelance editor takes the polish pass at an hour or two a week. The expensive version of this arrangement used to be called a content agency; the cheap version is possible now because the drafting is no longer the labor.

Can I run it for two businesses? The machinery duplicates fine; the judgment loop doesn't compress. Two blogs means two weekly blocks, and the honest failure mode is running both at half attention. Sequence them: get the first to cruising speed, roughly month three, before building the second.

What does "working" look like at day 30? Posts publishing on schedule without heroics, the backlog still ahead of the calendar, and Search Console showing the first impressions on the new pages. Nothing else yet, and that's on schedule.

The one-line version

One weekend builds the honest version of a self-running blog: Saturday wires measurement and a briefed, voice-fed drafting system with a written quality gate, Sunday banks a buyer-first topic backlog and automates publishing, and what remains forever is a two-hour weekly judgment loop that feeds the machine and checks its work. The machinery runs itself; the reasons anyone reads it stay yours.