How DigitalOcean updates its massive library of content 3-4x faster (without sacrificing quality)

The problem with chat-based AI

Early on, Anish’s experience with AI writing tools looked like many marketers’. Drafts could be generated fast, but they were … generic. 

“It was mostly a bunch of bullshit,” he says candidly. ​“I had to keep telling it not to sound robotic.”

Most content refresh workflows fail because AI is guessing at what you want. It doesn’t know the editorial rules, the SEO intent, your first-party research, or what has changed in reader behavior since the post was first published. Without real inputs, even the best models will poorly fill in the gaps.

So Anish made sure that the AI had all the context it needed to deliver an edit-ready first draft that:

• Loads editorial constraints, such as previous versions, a content brief, and style guidelines
• Pulls live search data via DataForSEO Model Context Protocol (MCP) servers
• Generates updates section by section, not all at once

He began using Cursor, an AI-first developer tool (also known as an integrated development environment or IDE), as the backbone of his content refresh workflow. Cursor​’s strength lies in how it handles context. Instead of pasting prompts into a chat window, Anish works directly inside a workspace containing the files that define the project: the article draft, SEO brief, editorial guidelines, checklists, and supporting documentation. 

That is a major shift from how most marketers use web-based AI tools like ChatGPT or Claude and is more in line with how developers use Claude Code or OpenAI’s Codex. In a typical chat interface, the model only sees the information pasted into the conversation. In workflows like this one, the AI can reference a broader working environment and pull from multiple connected files at once.

“I ask it to strictly follow these documents while making updates,” Anish explains. This approach mirrors how software is built. The specifications come first, execution comes second, and cleanup is minimized because the rules for output expectations are clear from the start, thanks to the provided brand standards that Cursor uses as context.

Training data goes stale. SERPs change weekly. Here’s the fix.

Even with strong constraints, something was missing. AI still relied on the static knowledge of connected files and dated training data from the AI model used in Cursor. Using stale data would produce useless results. The turning point came when Anish connected Cursor to live search data through MCP servers.

Using DataForSEO, Cursor can pull real keyword statistics, search volume, and live search engine results page (SERP) data directly into the workspace. ​“DataForSEO basically gives you all the relevant search data,” Anish says. ​“Keyword stats, search volume, SERP data, everything.”

“You can actually see the actions running in the background,” he adds. 

For example, you can verify that the MCP is pulling live search data when you see successful calls to the search volume API or organic live API, with details like location specified.

“The best part is, it’s not hallucinating,” he says. ​“Because it has the live data from the MCP server.”

Why Anish refuses to rewrite a whole tutorial at once

Getting the system running is mostly a one-time setup. Anish connected the DataForSEO MCP server inside Cursor by pasting in API tokens and account credentials, which allows the workspace to pull live data while he works.

The broad workflow is:

1. Download Cursor and connect your tools and data sources
2. Add context files to your project, such as editorial guidelines
3. Load stale content and use the chat + your folder structure to co-write with your AI that has all of your context.

To start in Cursor, navigate to Settings → Tools → MCP.

Then, add MCP servers by copying/​pasting API tokens and account details.

Anish recommends keeping it lean: mainly DigitalOcean MCP (for infrastructure tasks relating to his technical writing assignments) and DataForSEO MCP (for search data). 

“You just need to copy and paste a bunch of API token stuff… very straightforward.”

After that, most of the work happens inside a single workspace.

Before Anish begins a refresh, he opens the existing article and layers in the surrounding context for the assignment, including brand standards, updated SEO guidance, and any relevant supporting materials.

Anish then references the brief file in his natural language instructions for Cursor, where he can ​“tag” documents (such as DigitalOcean’s SEO writer checklist and AI content writing guidelines) as if he were writing a social media post:

He then prompts Cursor to pull live search data using the DataForSEO MCP server.

From there, Anish watches as Cursor runs actions to address the query in his prompt, in tandem with the instructions defined in his provided guidelines, which include:

• Keyword data + search volume calls
• Location targeting (often US, sometimes worldwide)
• SERP/​organic results calls (“organic live… API”)
• AI optimization/​keyword clustering-type calls (as supported)

Instead of refreshing an entire article at once, he goes one section at a time. Meta titles, meta descriptions, and the opening paragraph are usually the starting point. ​“Give me an updated title and description based on the SEO brief,” he’ll prompt, ​“and make sure you use actual search data.” Cursor can even apply those changes directly to the file when needed.

Here’s what this prompt looks like:

And the output:

After that, Anish works sequentially, prompting Cursor to update each one. ​“If you ask it to generate the whole tutorial, it usually generates a low-quality first draft, and most of it is AI slop,” he says. ​“I do it step by step.”

His update sequence typically looks like:

• Intro (SEO-aligned hook)
• Supporting sections (as needed based on brief + gaps)
• FAQ ideas (if relevant)
• Cleanup/​refinement editing pass

He can review and calibrate as he goes, rather than untangling a bad update of an entire article all at once. Then, with the AI updates in place, he invites subject matter experts to focus on the tutorial and technical aspects.

De-slopifying with thoughtful updates to the system 

Even with the provided context, early updates would miss the mark, and the workflow required constant calibration:

• When outputs sounded robotic, he tightened tone and style constraints. 
• When sections lacked depth or drifted from search intent, he added more structured briefing documents and live SERP data. 
• When the AI generated overly broad rewrites, he narrowed the scope to section-by-section updates instead of full-article generation.

Over time, Anish built a small set of reusable prompts that support this workflow. ​“I ended up with around nine prompts after a lot of testing,” he says. Some are used on nearly every refresh, including prompts for generating FAQ sections and identifying related topic clusters.

To surface real user questions, Anish also pairs Cursor with SEO tools like AlsoAsked and Answer the Public. The outputs become headings or FAQs that reflect how people actually search, not just how tools suggest expanding content.

And no matter how refined the system becomes, every piece of content passes through human review. 

“AI is a great tool, but it still needs an expert operator,” Anish says. ​“It needs somebody to steer the ship.” 

From days to hours: 3–4x more content updates