AI automation for blog content

Running a blog eats up hours you could spend on product work or client outreach. You stare at a blank screen, try to prompt ChatGPT, and end up rewriting the same outline three times. After reading this, you’ll have a repeatable system that turns a keyword into a drafted post, formats it, and pushes it to your CMS—plus a link to a ready‑made blueprint if you’d rather skip the build.

Why manual prompting fails at scale

When you rely on a single chat window, each new article forces you to repeat the same steps: research, outline, draft, edit. The process is tedious and error prone. You might forget to add internal links or miss a keyword you promised to target.

One‑sentence paragraph: the fatigue builds fast.

What you really need is a way to hand off each stage to a dedicated module that knows its job and passes clean data to the next step. Without that hand‑off, you waste time re‑typing prompts and fixing formatting.

How do you chain prompts without losing context?

Break the workflow into three distinct calls: research, outline, and draft. Each call receives a structured JSON blob from the previous step, so the model never has to guess what you already decided.

First, a research prompt takes a keyword and returns a list of sources with short notes. Second, an outline prompt consumes that list and produces a heading hierarchy with bullet points for each section. Third, a draft prompt receives the outline and writes the full article, obeying tone and length constraints.

Here’s a concrete named example of the research prompt we use with Claude 3 (first mention bolded):


You are a research assistant. Given the keyword "{{keyword}}", return a JSON array of up to five items. Each item must have: {"title": string, "url": string, "note": string}. Keep each note under 20 words. Do not add any extra text.


The outline prompt looks like this:


You are an editor. Given the research array, produce a JSON object with: {"title": string, "sections": [{"heading": string, "points": [string]}]}. Make the title catchy and under 60 characters. Do not add any extra text.


Finally the draft prompt:


You are a Writer. Using the outline, write a blog post of 800‑1000 words. Follow the tone: helpful, slightly informal. Include the keyword naturally in the first paragraph and twice more in the body. Do not add any extra text.


We glue these calls together in Make (first mention bolded). A scenario runs when a new row appears in a Google Sheet with a keyword. The scenario calls Claude via its API three times, passing the JSON output from one step to the next as input.

Cost note: Claude 3 API is $0.008 per 1k tokens for input and $0.024 per 1k tokens for output. A typical run uses about 6k tokens total, so the AI cost is roughly $0.14 per article. Make’s free tier allows 1 000 operations a month, which is enough for a solo blogger posting three times a week.

I think $0.14 per piece is fair for the time saved—honestly, I’d pay double if the quality stayed this high.

What most guides get wrong about AI blog automation

Many tutorials tell you to drop a single massive prompt into the model and hope it produces a ready‑to‑publish piece. That approach ignores the model’s limited context window and leads to repetitive or off‑topic sections.

They also suggest copying the output straight into WordPress without any formatting step. The result is raw markdown that looks broken when the theme expects HTML.

Another common mistake is treating the AI as a replacement for an editor. You still need a human to read the draft, fix awkward phrasing, and add internal links. The AI speeds up the first draft; it does not eliminate the need for a quick polish.

Concrete example: using Make, Claude, and WordPress API

Let’s walk through a real setup that puts the three‑step chain into production.

1. Create a Google Sheet with two columns: Keyword and Status.
2. In Make, add a “Watch Changes” module for the sheet. Set it to trigger when a new row is added.
3. Add an HTTP module that calls the Claude API with the research prompt. Map the Keyword column to the {{keyword}} placeholder.
4. Add a JSON parser to extract the research array.
5. Add a second HTTP module for the outline prompt, feeding the research array as input.
6. Add a third HTTP module for the draft prompt, feeding the outline JSON.
7. Add a final HTTP module that POSTs the draft to your WordPress site’s REST API endpoint /wp-json/wp/v2/posts. Include the title, content, and status set to draft.
8. Update the Google Sheet row to set Status to “Done”.

Each module uses the Make interface (first mention bolded) to map data fields—no custom code required.

Cost breakdown: Make’s free tier covers the scenario runs; Claude API cost stays under $0.20 per article as shown earlier. If you upgrade to Make’s Core plan at $9/mo you get more operations and error handling, which I find worth it for peace of mind.

I’ve found the WordPress API step to be the most finicky—sometimes the API returns a 401 if the JWT token expires mid‑run. That’s my concrete gripe: having to refresh the token manually every few weeks breaks the “set and forget” promise.

On the love side, the ability to see the draft appear in WordPress within seconds of the scenario finishing feels like magic. It turns a vague idea into a tangible post without opening the editor.

How to debug when this breaks

Start with the Make scenario log. If the research HTTP module shows a non‑200 status, check your Claude API key and quota. A 429 means you hit the rate limit—slow the trigger or upgrade the plan.

If the outline step returns empty JSON, the research array might be malformed. Add a temporary “Set Variable” module to output the raw research text to a log sheet so you can see what Claude actually returned.

When the WordPress POST fails, verify the authentication header. I keep a separate Make scenario that renews the JWT token every five days and stores it in a data store; the main scenario reads from that store.

Finally, open the draft in WordPress and look for stray HTML tags. If the content looks escaped, enable the “raw” option in the WordPress module or wrap the content in a CDATA block before posting.

One mild aside: if you’ve tried Zapier, you know what I mean—Make’s error handling is far clearer, which is why I switched.

Pricing and value opinion

Putting the pieces together, the ongoing cost is under $0.20 per article for AI plus $9/mo for Make if you need the extra ops. That’s roughly $2.60 per month for a three‑times‑a‑week schedule—less than a single coffee.

I think the free tier of Make is enough for solo work, but the $9/mo plan removes the anxiety of hitting limits.

For more on this exact angle, AI meeting tools coverage.

If you’d rather skip the build and deploy a working version in an afternoon, we’ve packaged this workflow as a blueprint at deepusecase.com/vault/ai-automation-blueprint.