When I first explored AI for marketing, I mostly used it for writing, brainstorming, and summarizing information. It saved some time, but I still had to do most of the work myself. I needed to research topics, check competitors, find content gaps, prepare drafts, and share everything with the team.
So, I decided to try something different. Instead of using AI only to complete individual tasks, I built a workflow where ChatGPT could handle several connected steps with less manual involvement.
The timing made sense. According to Gartner's 2025 survey of marketing leaders, 77% of organizations that had adopted generative AI were using it for creative development tasks. That shows how useful AI has become for marketers, but I wanted to explore what happens when it moves beyond generating content and starts managing a complete workflow.
Here's how I approached it, what I connected, and where I kept human review in the process.
What Is Digital Marketing Automation?
Digital marketing automation means using AI agents to handle repetitive marketing activities and save human time. This can include scheduling emails, tracking campaign performance, collecting leads, preparing reports, and managing content workflows.
A traditional automation (CRM) usually follows a fixed set of rules: when an event happens, perform a specific action. An AI agent can also interpret information, make decisions within defined limits, use connected tools, and choose the next step based on what it finds.
For example, a basic automation can notify me when a competitor publishes a blog post. An AI-powered workflow can go further by summarizing the post, identifying its main topic, checking whether my website already covers it, and preparing a content brief for review.
That was the kind of workflow I wanted to build.
Step 1: I Started with One Specific Marketing Problem
I didn't try to automate every marketing activity at once. That would have made the setup harder to manage.
I started with competitor monitoring and content research. These tasks take time because I have to visit competitor websites, review their latest posts, understand what they are discussing, and compare those topics with my own content.
I wanted one workflow that could do the initial research and prepare a useful starting point for my next article.
I called it my Competitor Spy and Content Creator workflow. Its job was simple: find a relevant new competitor post, analyze the topic, identify a possible content gap, and prepare a draft for my review.
The important part was defining what the workflow should deliver. I wasn't looking for a machine that produced more articles. I wanted it to help me find better opportunities without repeating research I had already done.
Step 2: I Connected the Workflow Using Make.com
For the automation layer, I used the visual workflow approach available through Make. n8n also supports a similar setup.
I configured the workflow to monitor competitor blog RSS feeds. RSS feeds are useful because they provide a structured way to detect newly published content without checking every website manually.
When a new post appeared, it triggered the next steps in the workflow. The automation collected the article URL and available details, then passed them to the AI model for analysis.
I kept the trigger narrow. Rather than monitoring dozens of competitors immediately, I focused on a small set of relevant websites. This made it easier to check whether the workflow was collecting useful information and avoid filling my notifications with irrelevant updates.
Step 3: I Used AI to Analyze the Competitor's Content
Once the workflow detected a new article, I passed its accessible text to OpenAI.
I asked ChatGPT to summarise the article, identify its main topic, extract relevant keywords, understand the intended audience, and list the questions the content answered.
I also asked it to identify what the article might have missed. That could mean an unexplored use case, a practical example, a comparison, or a question that the reader would probably ask next.
This step needed clear instructions. A vague prompt such as "analyse this blog" usually produces a broad summary. I got a more useful result by specifying exactly what I wanted and asking the model to separate observations from assumptions.
I also made sure the workflow could handle inaccessible pages. If an article couldn't be retrieved properly, the process needed to flag the problem instead of inventing a summary.
Step 4: I Checked the Content Gap Before Writing
A competitor publishing an article doesn't automatically mean I should write about the same topic. My website might already cover it, or the topic might have little value for my audience.
So, I added a research step before drafting.
Using Semrush, I would check the existing content, relevant keywords, search intent, and potential opportunities. Where API access was available, the workflow could retrieve supported data directly. Otherwise, I kept this check as a manual step rather than pretending the agent had access to information it couldn't retrieve.
The AI then compared the competitor's topic with the information I supplied about my existing content and suggested a possible angle.
For example, if a competitor published a broad article about AI chatbots, my next article might focus on implementation costs, integration challenges, or choosing the right chatbot for a business. The aim was to add something useful, not simply publish a response to every competitor update.
Step 5: I Generated a Draft and Sent It for Review
After the topic passed the initial checks, I gave ChatGPT my brand guidelines, target audience, preferred writing style, and the key points the article needed to cover.
I asked it to prepare a roughly 600-word first draft in Google Docs. I treated that draft as a starting point, not a finished article ready for publication.
The workflow then sent a notification to the team's Slack channel with the competitor article, the suggested content angle, and the draft link. This gave me one place to review the findings without searching through several tools.
I also kept approval as a separate step. The agent could prepare the draft and recommend the next action, but it couldn't publish the article automatically. Before anything moved toward publication, I needed to check the facts, confirm the SEO opportunity, improve the writing, and make sure the content offered something the reader couldn't get from the competitor's article alone.
Note: This article was written in collaboration with CMARIX.
Top comments (0)