This end-to-end workflow will be a beast — from SERP research and intent validation to briefing to creation. I’ll walk you through how I did it and report on how well it works, plus learnings.
I’ve built simple workflows in Claude Code before, but when it comes to large-scale content creation workflows, I’ve been using tools like AirOps. After all, they make it really easy to do so!
That being said, these tools aren’t cheap, and especially as I use them for my personal content creation workflows as a bootstrapped small business, I can’t help but wonder: could I just create this myself in Anthropic‘s Claude Code?
Well folks, that’s what we’re here today to find out. I’m NOT a developer.
I’ll share what I’m building (and why it’s different from most other workflows out there right now). Then, in part two, I’ll test it, share the results, and trial and error as I refine my output.
Quick note on what this is and isn’t. The goal is not to fully automate the content creation process; I’m still a firm believer that humans create the most interesting content, and should be included at the writing stage.
It’s to automate the repetitive, time-sucking parts of content creation and to test which inputs lead to quality outcomes. I will be testing an article-creation step (backed by a tone-and-voice skill and a ton of input from an SME), but I have a strong feeling we’ll need significant human intervention at the writing step.
Well, here goes!
What I’m focusing on that’s different
Plenty of smart teams have built versions of this. The ones I’ve seen tend to skip the parts I think matter most, so those are the parts I’m spending the most time on.
A few of them:
-
The expert POV gate. This is the one I’m most excited about. At a specific point in the workflow, it stops. It won’t keep going until a real person (the SME, or me standing in for them) answers a set of questions. This gives concrete questions to ask an SME that fit in with the narrative and intent you’re trying to build with the article. Then, the AI works the brief AROUND this unique POV while still addressing the piece’s intent.
-
My brief template. The workflow doesn’t spit out a generic brief. It builds against my brief template, in the incredibly detailed format we’ve refined across every client for years. It goes deep into whom we’re speaking to, what we want them to come away with and what makes our piece better than everything else out there. By the time it’s done, a writer shouldn’t have to make a single strategic decision. Just execute. It’s my hope that by providing this level of detail to the LLM, it may output better content during the automation process, but we’ll see about that!
-
A voice profile. Instead of writing out a list of style rules, I fed it a handful of examples from my own writing and asked it to identify the patterns to create a writer’s guide. It came back with a voice guide based on what it actually observed, not what I told it about myself. That’s what the draft step leans on, and again, we’ll see how it turns out. For client work, their examples would get loaded instead of mine.
-
An accuracy and claims validation. I’m going to try building in a step, before it gets to the edit phase, where accuracy, claims, data, stats, and framing are validated. I’m asking it to double-check itself, testing each claim by verifying it vs. running from memory. We’ll see if that cuts down on errors and hallucinations when I check during the edit phase.
The plan: nine steps to start
I wanted this workflow to run from the very beginning, so I need live keyword data, SERP analysis, competitor rankings, and Brand Radar AI citation data. I wired it through Ahrefs so I can get these straight from the source without me ever opening the dashboard.
As a note: I’m keeping the deep, under-the-hood stuff light, but I also want to articulate how much is happening at each step.
Step 1: Research.
Calls Ahrefs via the MCP connection for live keyword data (volume, difficulty, parent topic, related terms, search intent), then fetches the top-ranking competitor pages and dissects their themes, headers, and proof points.
It sorts what everyone covers (table stakes), what few cover well (the opening), and what nobody’s touched yet (the ownable gap). When a term comes back thin, it runs a web-fetch fallback for competitor analysis instead of stalling. It ends with questions grouped into themes rather than a flat list.
Step 2: Existing content reference.
Uses Ahrefs Site Explorer to map what my site (or the client) already owns on the topic. It finds the existing pages, identifies which ones rank for related terms, and extracts owned proof points already published (proprietary data, product use cases, named expert quotes).
Then it flags internal linking opportunities and warns me if a new piece would cannibalize something already live.
Step 3: Expert POV.
This additional context is SO important. This is where it develops questions based on the gaps it found in step 1, pauses to wait for human input, and blocks every downstream step until a point of view is confirmed.
Step 4: Brief creation.
Synthesizes the first three steps into my full, proprietary brief template. The unique point of view and SME quotes, SERP analysis with a specific answer to “how will ours be better,” internal and external links mapped, and a full outline with two to three sentences of editorial direction per section.
Step 5: Product and service mentions.
Reads the finished brief and pre-loaded offerings/products from the context file, then adds signpost notes to the sections where a mention genuinely helps the reader. This helps place CTAs and gives direction on natural ways to weave the product or offering into the story.
Step 6: Draft.
This is the experimental phase, which I might take out later. Claude writes the article based on the brief and the preloaded voice profile.
It opens on a bold claim or a counterintuitive data point, stays answer-first in every section, loads the heaviest thinking into the first third of the piece (where the bulk of AI citations come from), skips the tidy wrap-ups, and AI tells, then it pauses for my review.
Step 7: AEO audit.
Runs the draft against my AEO framework (answer-first structure, named expert attribution, original data present, Q&A format where it fits, citations to credible external sources) and hands back a punch-list of fixes, not a rewrite.
I may actually move this before the drafting phase, now that I think about it, so it doesn’t have to redo any work.
Step 8: Links and claims validation.
I’ve given Claude tight parameters so that it validates every statistic against its actual source (vs. working from memory, as it likes to do), confirms that internal links are live, and catches linking opportunities I missed.
Step 9: Format for publish.
Outputs everything CMS-ready: meta fields, FAQ schema markup, TL;DR box, table of contents, CTA formatting. Paste and go.
Nine steps, each one its own skill file, telling Claude Code exactly what to read, what to do, and what to save before it’s allowed to move on.
What I’m hoping to end up with
When it’s finished, the workflow for a new piece should look like this: fill out the context file, drop in the keyword and any notes, run it. The research and existing content steps run automatically. The pipeline stops and asks for expert input. I answer (typed or dictated). The brief builds, the mentions get added, the draft comes back for review, and the final steps run after I approve it.
What comes out the other end is a fully formatted, CMS-ready article with a complete brief on file, every claim validated, internal links confirmed, AEO structure in place, and a real expert perspective running through it.
After this is refined, I’d love to see if I can get Claude to help with image creation (charts, graphs, etc), as that’s a super time-consuming part of the process. I’d also like to see if I can build a user-friendly dash to go with it.
Now, mine might not do everything a tool like AirOps does (I have a feeling I’ll miss the grid feature). I might fiddle around with this workflow for a week and decide it’s too much hassle, that using a pre-built tool is the move. That’s a real possibility, and if it happens, I’ll tell you. That’s sort of the point of building this in public.
What’s next
I’m aiming to finish the build this week and start running real keywords through it. Part 2 will be next week — so stay tuned! It will include the first outputs: what worked, what broke, and the trial and error of getting it to actually produce something good.
If you’re thinking you might want a system like this living inside your own business, that’s the direction this is headed. Reach out at any time, and I can build something like this custom or in your tool of choice.
Check out where I ended up with the workflow here!
Karli is content marketing consultant behind Wild Idea, a content marketing and SEO collective focused on driving big results. With over 12 years in the marketing industry, she’s worked with brands large and small across many industries to grow organic traffic and reach new audiences. She writes on everything from marketing, social, and SEO to travel and real estate. On the weekends, she loves to explore new places, enjoy the outdoors and have a glass or two of vino!
