How I Built a Content System That Doesn't Sound Like AI Slop

When I launched work.flowers, I knew that being a first-time founder would mean having to learn some new skills, including marketing. I’d always been active on social media in a personal capacity, but building a professional social media and blog presence would require developing some new muscles.

Naturally, I wanted to use AI and automation to make my content workflows as efficient as possible, but with one important caveat: I wasn't willing to trade efficiency for sounding like everyone else. The internet already has enough AI slop. I wanted to automate the mechanics, not the thinking.

Here's the system I've built, and the guardrails I built in to leverage automation without veering into "ChatGPT voice" territory.

Everything lives in Notion. Not just drafts — the entire content operation. Blog posts, social ideas, analytics, publishing status, writing guidelines, even the instructions that tell my AI agents how to write in my voice. If it's related to content, it's in Notion.

This matters because it means every tool in the stack has a single source of truth to read from and write to. There's no reconciling a Google Doc with a Trello card with a note in your phone. The context is always in one place.

I like to read. A lot. The first problem with most content systems is the blank page. I've solved this by wiring my reading life directly into my content pipeline.

I use an app called Readwise to capture notes and highlights from everything I read: books on my Kindle, email newsletters, articles I read online, and PDFs. Those highlights sync automatically to a Notion database, which means my AI agents can search them when drafting new content. The raw material is always there.

When I tag an article in Readwise Reader as something I want to write about, a Zapier automation creates a new entry in my social content database. When I finish a book on my Kindle, it gets added to my "Read" shelf in Goodreads, which triggers another Zap via the Goodreads RSS feed, which in turn creates a post prompt in Notion.

The result: I never sit down to write without a backlog of ideas that are already connected to things I've actually read and found interesting.

Once there's an idea in the database, a Notion Custom Agent handles the initial draft.

I've written detailed instructions and a style guide that reflect how I actually write — the topics I care about, the phrasing I avoid, the structure I tend to follow. Before starting any new piece, the Custom Agent is instructed to search my past posts and Readwise highlights for related themes. This does two things: it prevents me from repeating myself without acknowledging it, and it surfaces connections between ideas I might not have spotted on my own.

The Custom Agent drops the draft back into Notion, tagged and ready for my review. I edit from there: tightening the argument, adjusting the tone, cutting anything that feels like it was written by a machine trying to sound like a human.

The instructions themselves don't stay static either. I have a companion Custom Agent that runs once a month, reviews my recent posts, and suggests updates to the style guide. Most writing guidelines go stale over time; my voice shifts and my interests are continuously evolving, so I wanted my instructions to evolve as well. This keeps the system calibrated to where I actually am and what I’m currently thinking about. (I do worry that could create a self-reinforcing feedback loop in my writing, so that’s something I’ll be watching out for with this workflow.)

For social scheduling, I use a fantastic app called Scheduled.so that was built around Notion from the ground up.

Before Scheduled, I was connecting Notion to Buffer via Zapier. It technically worked, but it was clunky, stripped out a lot of formatting, lacked core functionality, required frequent maintenance, and got out of sync any time I wanted to edit a post after scheduling it (which was basically always). I was spending as much time maintaining the integration as I was actually publishing content.

Scheduled is different because Notion remains centred as the source of truth for your content . You write your posts directly in a Notion database, and it handles embedded media and links (for example, to tag people or LinkedIn profiles), natively. When a post is ready, you update a Status property to "Ready to schedule" and set the publish date and time in another property, and you’re ready to go.

Scheduled always reads the latest available content from your Notion page at the moment of publishing—not at the moment of scheduling. This solved a major pain point for me: my previous Notion → Zapier → Buffer workflow copied the content into Buffer at the time the post was scheduled, which meant any subsequent edits had to be made in Buffer rather than in Notion. The two would frequently fall out of sync, and Notion stopped being the source of truth the moment you hit the schedule button. With Scheduled, Notion remains the source of truth right up until the post goes live.

Scheduled works with LinkedIn, Facebook, Instagram Reels, TikTok, YouTube Shorts and X, and once a post goes live, it syncs performance stats — reach, impressions, engagement — back into the same Notion database daily. Everything is in one place, and the content calendar and analytics live right alongside the posts themselves.

Once Notion's upcoming Dashboard feature—which will enable you to consolidate multiple database views in a panelled layout— goes live, all of that data will be visible in a single view. One pane for the whole operation.

The work.flowers website and blog also run on Notion as the CMS, with Bullet.so converting the content into a proper website. Bullet was built specifically for this use case: it takes your Notion pages and renders them as a fast, professionally designed, SEO-optimised site, so we never have to leave Notion to manage content.

The setup is straightforward. You choose a theme on Bullet (which you can later customise), install a database template, and write and manage your content there. Bullet handles everything else: hosting, metadata, lightweight analytics, and RSS feeds for the blog section. The result looks nothing like a raw Notion page; it just looks like a properly built website.

For the blog specifically, Bullet reads properties directly from the Notion database — tags, author info, metadata, publish dates, slugs — to populate everything that matters for a blog post. SEO is managed entirely from Notion without touching any settings in Bullet itself, and you can use Notion AI to auto-populate metadata based on the post content.

When a post is ready, I just tick a "Publish" checkbox property. Bullet picks this up and makes the post live on the next sync.

The system isn't clever because of any single tool. It works because every piece connects back to Notion, and because the automation handles the mechanical parts — idea capture, draft creation, publishing, distribution, analytics — while leaving the judgment calls to me.

I still decide what's worth writing about. I still edit every draft. I'm still the one who reads the books, follows the threads, and decides what's actually interesting.

The automation just means I spend less time on the mechanics and more time on the thinking. Which, if you're trying to not sound like AI slop, is the whole point.