5 Ways to Automate Content Creation with AI Workflows 

Automating content production is one of the main use cases for AI workflows in B2B SaaS marketing. 

The challenge? Identifying the most optimal ways to use them to increase content creation speed without sacrificing quality.

Hence the article. In the next 10 minutes, I will show you examples of AI workflows I’ve built to automate different content creation stages, from research through editing and beyond. 

I will also share a few best practices to help you get the most out of your workflows.

How to automate content creation with AI workflows

You can use AI workflows at every content creation step. Research, briefing, drafting, editing, repurposing, you name it. 

In the next few sections, I will show you how to do it.

Mind you, though, ‘can’ doesn’t mean ‘should.’ I believe some steps are best left to humans, while others need close human supervision. I will explain why. 

A quick note: I will use examples of workflows I’ve built in AirOps because that’s what I’ve been using recently. But you can easily replicate them in other tools, like Claude Code,  Hunch, or n8n. 

And in case you’re wondering, I’m not affiliated with AirOps in any way.

1. How to automate research with AI workflows

When I research content for my freelance clients, I use dozens of sources. 

Searching all of them one by one — even with AI tools like Consensus — takes forever. So research was the first thing I’ve automated with a sequence of workflows, one for each source type I might use.

Here’s an example of one for SERP analysis, with a quick overview of how it works:

1. It runs a Google search for the target keyword and extracts all URLs. 

2. It scrapes all the pages and combines their content.

3. It summarizes all the articles and extracts all the unique questions they answer, implicitly and explicitly. And analyzes it for patterns, angles, and gaps.

The workflow is chained to a similar one for Reddit and produces a detailed analysis report and a basic article outline, which you can pass to your content writer. 

Or enrich the outline with insights from Deep Research, YouTube, Google Scholar, newsletters saved in Notion, case studies and industry reports saved in the Knowledge Base, Google Books (if the book is available), or LinkedIn posts. 

Whichever is relevant for the particular topic.

All these workflows can run in parallel for every keyword, and that’s where the time savings happen.

I’ve built a similar sequence of workflows for software research. 

Here, the foundation isn’t SERP analysis but the product website research workflow that scrapes the official website, extracts information about all features, pricing, use cases, user personas, positioning, and everything else you need to write a decent listicle.

Of course, AI can’t test all the tools for you, and that’s the most accurate way to evaluate software. And I haven’t found a reliable way to scrape reviews from third-party sites, yet, so in-depth review analysis is another thing the writer must do themselves.

The biggest limitation of using AI for research is that it follows a certain pre-programmed path from which it cannot diverge to explore unexpected topics or angles that emerge during research. That’s where human researchers win hands down. 

2. How to automate content outlining with AI

I’ll be honest with you: Outlining is one area that I would be reluctant to automate with AI.

The reason? AI isn’t good at original thinking. 

It can spot patterns and themes across the competing content and sources, but it isn’t great at finding unique angles. Not ones that would resonate with the audience, anyway.

So the outline is likely to resemble everything else that’s already been written. Which won’t work if you want your content stand out.

But it will work for SEO/AIO content. 

If you choose to go this way, here’s an overview of the outline workflow that I’ve built:

1. Creates a detailed outline based on research and SERP analysis.
2. Checks how structurally sound the outline is (idea flow, H2/H3/H4 hierarchy, MECE, BLUF, etc.), and recommends edits.
3. Implements the edits suggested in step 2. 

3. How to automate editing with AI workflows

Chloe West, the Content Marketing Manager at Social Vista, has recently shared this post about editing “on vibes.” 

Lots of experienced editors work like that. We often can’t rationally explain why we choose to tweak the copy. 

AI can’t act on vibes. It can’t feel if something sounds right or wrong. 

But it’s good at following checklists, so it can take some of the mechanical editing off our plates. 

Here’s how my (optimization and) editing workflow works:

1. SEO Optimization: Adds relevant keywords and semantic terms across the copy. Naturally, without stuffing them where they don’t belong.

2. High-level edit: Inspects alignment with the outline/brief, the flow of ideas, looks for fallacies, factual inaccuracies, and sections that a sceptical reader might question. The follow-up prompt implements the recommended changes.

3. Line-edit: Removes cliches, repetitions, tautologies, changes passive voice into active, cuts to fluff.

4. AI text humanizer: Whether AI wrote the draft or not, it may contain language overused by AI, like participle clauses. This step gets rid of them.

5. SEO check: It checks if the keywords and semantic terms are still in the text after the previous rounds of edits and adds more if it can be done naturally.

I still edit manually after running the workflow, but it usually gets me 90% of the way. 

4. How to automate content refreshes with AI workflows

Content refreshes seem to be the hot topic at the moment, and rightly so. 

Companies sit on dozens — if not hundreds or thousands — of blog posts that gradually decay into oblivion. Updating them often delivers more value than creating new content. 

AirOps has an out-of-the-box refresh workflow grid that works well, so I haven’t built any bespoke refresh workflow yet (it is coming in the next week or so).

How does the AirOps workflow work?

1. Choose your objectives, like improving organic and AI search performance or adding extra internal links. 

2. Decide how to populate the grid with blog URLs. For example, by importing it from a csv file or directly from Webflow or WordPress. You can also copy/paste additional URLs once the grid is ready.

3. AirOps extracts the article title, meta description, article body, and the last modification date from the URL.

The grid consists of a sequence of workflows that you can run at the same time. 

The two I’m interested in are AEO visibility and organic search optimization. For each of them, AirOps assesses the article against a set of criteria and recommends improvements.

Here’s what they look like for AI search optimization:


And here for organic search:

At this point, I can manually edit the suggestions and assign the update to a writer. 

Or run another workflow that implements the suggested changes.

Prioritize pages to refresh

When building the refresh grid in AirOps, you can choose to filter the pages in your blog based on their last update date, click drop over the past 30 days, or rank drop in the last 30 days.

If you have a large library, you’re still potentially left with dozens of blog posts to update. If you’re using AI to do the complete updates and have enough credits in your wallet, that’s no problem.

But it may not be if you’re doing the actual updates manually. Also, I felt like there are more relevant criteria to consider when prioritizing blog refreshes, like their business value.

 A high-converting BOFU piece should be higher on the list than a TOFU one. An article in the top 11-20 may bring more traffic if we bump it to the first page than one in positions 4-10 that we bring to the top 3. Assuming people actually click on the search results for the keyword.

To address this, I’ve built a workflow that prioritizes pages to refresh based on their:

– Funnel stage (BOFU content, like listicles or X vs Y articles, gets higher scores than TOFU guides)

– Conversions (Higher conversions = higher priority)

– Keyword search volume and traffic (higher volume = higher priority)

– Organic rankings (1-3 vs 4-10 vs 11-20, etc)

– Rankings decay

– Freshness (pieces updated within the last 3 months get 0)

– Difficulty (based on competitor backlink profile strength)

The data comes from GSC, Semrush, and DataForSEO (can be Ahrefs or Moz if that’s what you prefer), and GA4 (which was a bit of a battle to set up to start with).  

Once we have the data, a sequence of prompts evaluates the article based on the set criteria and assigns a score.

The final prompt calculates the overall score and generates a report with a rationale. 

Note that to calculate the overall score, the individual criteria scores are weighted. So, the business potential affects the score more than rankings, potential traffic more than freshness, and so on. You can adjust the weighting to your priorities. 

5. How to automate content repurposing with AI workflows

Content repurposing is one of the areas that you can safely scale with AI workflows without worrying about quality. 

AI is perfectly capable of converting your YouTube into a blog post, turning it into individual social media posts or a newsletter. If the ideas in your original asset are sound, the quality won’t suffer.

Here’s an example of such a workflow I’ve built for LinkedIn. 

1. It scrapes the blog URL and identifies unique ideas.

2. I review which ideas I’d like to convert into LI posts manually
3. The prompt writes each post based on the examples I’ve saved in the brand kit. 

You can duplicate the workflow and adjust the prompts to produce social posts for other social channels, ready to plug into your scheduling tool.  

Content engineering best practices

Let’s wrap up with a few content engineering best practices that all the above workflows are based on.

1. Break the process into multiple small steps

Breaking the workflow into smaller steps almost always gives better results. 

For three reasons:

• The more complex the prompt, the higher the risk of the AI model getting confused and turning out output we’re not happy with. 
• It’s easier to find bugs in shorter prompts.
• Some LLMs, like ChatGPT, have very limited output size or context windows, so they may not be able to produce larger sections of text without the quality gradually deteriorating or the text cutting off mid-sentence. That’s why my workflows create one section at a time and only then combine them into a whole piece. 

2. Engineer human oversight into the process

Embedding human overview steps into the process generally improves the overall quality of the final workflow output.

AI is good at completing lots of content tasks at a decent level, but it’s never 100% there. So a human should step in and fix what AI can’t deliver.

For example, in my workflows, I’m always asked to choose the Reddit threads to analyze or YouTube videos to transcribe to ensure relevance. And I won’t let the workflow move directly to drafting before I review the outline.

Naturally, the more human oversight, the longer it takes to get the output. So if speed and high publishing cadence are the priority and you’re happy to ship content that may be lower quality, include fewer human revisions.

3. Embed quality checks in the workflow

LLMs still hallucinate a lot. And if you ask them to do something once, it doesn’t mean they will do it. 

That’s why all my workflows have automatic quality checks to ensure the right output quality. 

For example, my deep research workflow consists of three steps:

1. The actual deep research.
2. The source vs the report check: It goes directly to the sources cited in the first report and checks their quality, currency, and if the information was cited accurately.
3. A general accuracy check on the step 2 report.

Or when I want it to follow specific writing rules, I repeat them through the workflow. In the brand kit/system prompt, in the writing prompt, and again in the editing prompt.

Such sequences take longer to execute and are more resource-intensive, but structuring workflows like this improves accuracy dramatically. And it’s nothing compared to checking every stat or claim manually. 

4. Build context so as not to leave room for interpretation

Prompting LLMs is like talking to someone who grew up in a completely different part of the world. They may speak the same language, but you don’t share the same context, so they may interpret your words very differently. 

To avoid misunderstandings, we need to give it the 

For B2B content, here’s a good start:

– Brand information

– ICP 

– Key competitors

– Point of view

– Author persona

– Reader personas

– Tone of voice

– Style requirements

– Templates and content samples

– Reliable sources to cite

In AirOps, all this information lives in the Brand Kit, and in Claude Code, I have it all organized in neatly organized folders on my hard drive. 

5. Build reusable knowledge bases

Building knowledge bases with source information is one of the easiest ways to scale content creation. 

Let’s take creating BOFU content: listicles and X vs Y articles. 

Storing detailed information — which AI can help you gather — in a database means you don’t have to research each tool again and again whenever you or your teammate creates a content brief or wants to update the article. It’s already there, ready to access.

What else do I store my knowledge bases?

– Case studies

– Proprietary internal data

– Industry reports

– Niches newsletters

– SME interview transcripts

All in one place, ready to access. 

Wrapping up

Scaling AI content creation can mean different things. You can scale every step of the process or just individual ones, like editing. 

When choosing which phase to automate — and which to leave to experienced writers and editors – consider these factors:

• AI capabilities and limitations: Current AI models can perform all content tasks, but some of them better than others. They’re great for structured tasks that follow a checklist, like research and editing. They don’t do well on tasks that require creativity, like creating original outlines.
  
• Your prompting and content engineering skills: An experienced content engineer may easily deliver results that others find impossible to achieve.
  
• Scale: Building the workflows requires upfront investment, which only makes sense if you’re using them for big projects regularly. 

FAQs

1. What is content engineering?

Content engineering is a systematic approach to creating, managing, and optimizing content using structured processes, data analysis, and technology. It combines traditional content creation with engineering principles like scalability, consistency, and efficiency. 

Content engineers develop frameworks, templates, and workflows to ensure content quality while enabling faster production. 

Content engineering involves technical skills like working with content management systems, APIs, and automation tools, and strategic thinking about information architecture and user experience.

2. What are the best content automation tools?

The best content automation tools include AirOps and Hunch for content creation and optimization at scale, Jasper and Copy.AI for AI-powered writing, Zapier, Make, and n8n for workflow automation, and Surfer and Neuronwriter for content optimization. 

You can also create content workflows with Claude Code.

3. What does content at scale mean?

Content at scale refers to producing large volumes of high-quality content efficiently and consistently across multiple channels. 

This requires systematic processes, strategic planning, and technology assistance to maintain quality while increasing output. It involves creating repeatable frameworks and using automation tools to streamline workflows without sacrificing relevance or value.