We Tried to Automate Our Clients’ Blogs With AI. Here’s the Honest Version.
I’ll be honest — when we first started seriously experimenting with AI-generated blog content, part of me felt a bit weird about it.
We run a content agency. Writing is something we sell. So there’s an obvious question sitting right there: if we help clients automate their blog, are we handing them a way to stop paying for our copywriting service?
The answer, after actually building and running these workflows for a handful of clients, is more nuanced than yes or no. But we’ll get there.
This is the write-up of what we’ve learned — the real version, not the LinkedIn post version.
What we actually built
First, let’s be specific about what “blog automation” means here, because the term covers a wide range of things from genuinely useful to completely useless.
On the completely useless end: you click a button, get a 1,200-word article, publish it immediately. We tried versions of this early on. The content was technically coherent and utterly forgettable — the kind of thing that exists in abundance on the internet and that nobody actually reads.
What we ended up building is a workflow, not a content machine. The distinction matters.
| The workflow: → a scheduled trigger pulls keyword opportunities from competitor research and trend data → a shortlist lands in the client’s approval queue → they pick what gets written → Claude drafts a structured article calibrated to their brand voice → the draft appears in WordPress → the client reviews and publishes. Or, if they’ve opted in, it auto-publishes on a schedule. |
The human doesn’t disappear. They just stop doing the parts that don’t need them.
That framing — removing the unnecessary manual work without removing the judgment — is what made this feel right to build. It’s also, if we’re honest, what took the longest to figure out.
Why this became a thing we build for clients
It came from a recurring frustration.
Client comes to us for SEO help. We audit the site, find solid keyword opportunities, build a content plan. We recommend publishing two to four articles a month — not aggressive, pretty standard for organic growth.
Three months later, they’ve published one article. Maybe two.
Not because the strategy was wrong. Because nobody had time. Content is almost always the thing that gets bumped when a business gets busy, which is most of the time.
We’d tried the obvious fixes. Freelance writers — variable quality, slow, expensive. Our own content writing service — good output, not scalable at the volume and pace that SEO actually needs. In-house — great in theory, falls apart in practice for most teams under a certain size.
AI automation kept coming up as the answer we’d circle back to. Not because it was the fashionable thing (though it definitely is that too), but because it was genuinely the right tool for a specific, real problem: consistent execution at a volume that manual processes couldn’t sustain.
What no one tells you — the parts that are actually hard
Every writeup of AI content automation focuses on the technology. The tools, the integrations, the clever prompts. That’s the easy part, relatively speaking.
Here’s what actually takes time.
The brand voice prompt is a project, not a setting
The first version of any client’s prompt produces content that is accurate and completely indistinct. It sounds like a blog post. It doesn’t sound like their blog post.
Closing that gap requires a proper brief — what tone do they write in, what do they never say, who exactly is reading this, what’s the one thing they’d want a reader to take away. It requires example articles, rounds of iteration, refinement based on actual feedback.
By version three or four, the output is usually at a point where clients stop saying “can you tell it’s AI?” and start saying “can we just publish this?” That’s the threshold. Getting there takes longer than the technical setup.
Most clients underestimate this step. We’ve stopped letting them.
The approval step is load-bearing
We built a topic approval step into every workflow — the client reviews a shortlist of keyword opportunities before anything is written, and only the ones they greenlight move forward.
Honestly, we included it partly as a safeguard. What surprised us was how much clients actually value it. Not as a burden — as a sense of control.
They don’t have to think about content strategy. They don’t have to come up with ideas. But they do get to decide what represents their brand and what doesn’t. That line felt important to keep in place.
Auto-publish is great — once you’ve earned it
We offer fully automated publishing as an option. Some clients love it. The workflow runs, the article appears on their site, they find out when they check their analytics.
But we always recommend starting in review mode, and we’re pretty firm about this.
AI outputs are consistent most of the time. Not all of the time. Occasionally a draft will miss the tone, go off-brief on a detail, or make a factual claim that needs a second look. If you’re auto-publishing without a human pass, those go live before anyone sees them.
The clients who are happy with auto-publish are the ones who spent a month in review mode first, calibrated the prompt properly, built genuine confidence in the output quality, and then switched. That sequence matters a lot.
Your WordPress needs to actually be clean
The workflow pushes drafts via WordPress REST API. This requires a reasonably well-configured WordPress install — REST API enabled, correct user permissions, SEO metadata fields properly set up.
For clients on our headless stack, this is completely seamless. For clients with a few years of plugin accumulation and no one really looking after the backend, there’s sometimes a cleanup pass before everything connects properly.
Solvable every time. But worth knowing before you start, not during.
What’s actually working
We’ve been running this with clients long enough to see real patterns, not just early excitement.
The research layer catches things manual processes miss
The automated keyword and competitor research — running on a set schedule, every week, without anyone having to remember to do it — surfaces opportunities that a monthly manual check would either miss entirely or deprioritise.
One client told us they’d published more targeted SEO content in two months with the automation than in the previous two years of trying to maintain a manual content calendar. The strategy hadn’t changed. The execution finally matched it.
Consistency is the whole point
SEO content doesn’t work in bursts. One great article won’t do much. A year of consistent, relevant publishing on the right keyword territory starts to compound into something real.
Manual content production at that pace is genuinely hard to maintain. The automation makes consistency the default. That’s not a small thing.
It creates space for the writing that actually needs a human
This was the part we didn’t fully see coming.
When the routine informational content — the how-tos, the explainers, the comparison articles — is handled by the workflow, the human writing time that remains gets used for the content that genuinely benefits from a person behind it. The opinion pieces. The client stories. The things that have a perspective that’s actually yours.
That’s the right division. The automation isn’t replacing good content. It’s creating the space for it.
Who this is for — and who it isn’t
We’re not going to tell you this is the right move for everyone. It’s not.
If your writing voice is itself part of what makes your brand worth following — if people come back specifically for how you think and how you express it — full automation on your blog is probably the wrong call. Some things need to sound like you wrote them because you did.
But if your blog is a marketing channel, not a personal platform — if the goal is to rank for the questions your customers are already asking, drive organic traffic, and be findable — then the workflow does that job well. Better than most small teams manage manually, because it doesn’t get tired or distracted or deprioritised.
Good fit: WordPress-based business blog, a clear sense of the keyword territory you want to own, a small team that knows they should be publishing more than they are. Not a fit: blogs where the voice is the whole point, or businesses with no clear idea of what topics they’re trying to own.
The thing I didn’t expect
I said at the start that I felt a bit weird about building this. Like we were doing something slightly against our own interests.
I don’t feel that way anymore.
What I didn’t anticipate is that building the automation forces you to answer questions about content and brand voice that most businesses have never made explicit. What exactly is your tone? What topics do you actually own? What makes a good article for your audience specifically?
Those questions should have clear answers before any content gets written — automated or not. The workflow just makes vagueness impossible, because a machine can’t approximate its way past a fuzzy brief the way a human writer sometimes quietly does.
Building this has made us better at the manual version of content work too. That’s the part nobody mentions.
If you’re thinking about this for your business
We build the workflow once and hand it over completely. Your n8n instance, your API keys, your WordPress. Full documentation, a proper handover session, and then you run it independently — no retainer, no dependency on us.
If you want to know whether it makes sense for your specific situation before committing to anything, that’s exactly the conversation we’re happy to have first. Contact us now.
FAQs
Will it actually sound like us, or will everyone know it’s AI?
With proper prompt calibration, the output reads as on-brand rather than generic. The bigger risk isn’t sounding like AI — it’s sounding like every other blog in your industry. We spend real time on the prompt engineering to avoid that. The first drafts usually need refinement. By iteration three or four, most clients are comfortable publishing with light edits.
What if we decide we want to change something after handover?
You own the workflow files. Adjusting the prompt, changing the publishing frequency, switching the approval tool — all of that is documented and yours to modify. If you want us to make bigger structural changes, we’ll quote it as a separate session. But the point of the handover is that you shouldn’t need us for the day-to-day.
Can it handle Bahasa Indonesia, or just English?
Claude handles Bahasa Indonesia well — we’ve built bilingual workflows for clients targeting both local and international audiences. It does take more care in the prompt engineering phase, but the architecture is the same. If you need multilingual output, flag it at the start of the project.
Does this only work with WordPress?
The current build is designed around WordPress via REST API, which is where we have the most experience and the cleanest integration. Other platforms with open APIs — Ghost, Webflow, some others — are possible but would need assessing case by case. If you’re not on WordPress, bring it up during discovery and we’ll be straight with you about what’s feasible.
How much does it cost to actually run once you’ve set it up for us?
The main ongoing costs are n8n Cloud (from around $20/month), Claude API usage (roughly $0.50–$2 per article depending on length), and optionally a keyword research tool like SEMrush if you don’t already have one. Most clients running 8–12 articles a month land somewhere between $40–$80/month in tool costs. That’s the full bill once our setup fee is paid.
Ratri Jawanes is the CTO of Jupitr Agency, a digital agency based in Bali specialising in headless WordPress, AI automation, and digital strategy.
