How I Use AI to Write 6 Blog Posts a Day Across 3 Sites

Six blog posts a day. Three sites. Every weekday.

When I tell people that, the first reaction is usually: “There is no way the quality is there.” And I get it. That used to be my reaction too. But I have been running this for months now, and the quality is not the problem I thought it would be. The real challenge is something different. It is the system.

Let me walk you through exactly how I do it, what broke along the way, and what I would do differently if I were starting from scratch today.

Why Six Posts a Day

I am building three sites simultaneously. eNZeTi is a real-time intake coaching platform for law firms. Cultivate Inbox is my cold email outreach agency. And jessenavarro.com is the personal brand where I document everything I am learning.

Each site needs content to build authority in search. I am not going to rank by publishing once a week. The math does not work. If I want to outpace competitors who have been publishing for years, I need volume on top of quality. Two posts per site per day gets me there without burning out a human writer.

That is the real reason I built this. Not to cut corners. To compete.

The Infrastructure Behind It

I run a team of AI agents I call the Wolf Pack. Shakti is the content engine. Her job is to research, write, format, and publish. She does not just generate text and hand it off. She runs the full workflow from topic selection to WordPress publish.

Here is the stack:

• Topic calendar: I built a master content calendar for each site with 30+ article titles, slugs, target keywords, and internal linking destinations. Shakti reads the calendar, checks what has been published, and picks the next unpublished item. No human decision needed.
• Research: Before writing, Shakti runs 1-2 web searches on the primary keyword. She also pulls from internal research files. For eNZeTi content, she reads verbatim quotes from attorneys and intake coordinators I have collected from Reddit and industry reports. Real data. Real voices. Not made up.
• Writing: Each article is 1,000 to 1,800 words. First-person when it is my personal brand site. Third-person educational when it is a product site. The voice guidelines are written into each agent’s instructions. I have specific rules: no em dashes, no buzzwords like “game-changer” or “leverage,” no fabricated statistics.
• Publishing: Shakti formats the article in clean HTML, creates the JSON payload, and hits the WordPress REST API directly. Title, slug, category, excerpt, featured image alt text. All automated.

What the Workflow Actually Looks Like

The cron fires twice a day per site. In the morning and at noon. Each fire produces one article.

When the cron runs, here is the sequence:

1. Read the brand identity file for that site. This keeps the voice consistent even across hundreds of articles over months.
2. Pull the list of already-published slugs from the WordPress API.
3. Read the content calendar and find the next unpublished item.
4. Run the research step. Pull relevant VOC quotes if available. Run a search on the primary keyword.
5. Write the article in one pass. No drafts. No review queue.
6. Build the JSON payload and publish to WordPress.
7. Log the completed task to Mission Control so I can see what went out.

Start to finish, the whole thing runs in under three minutes.

The Part Everyone Gets Wrong

People think the writing is the hard part. It is not. The writing is actually the easiest thing to automate at this point. What takes time to build right is the context layer.

By context, I mean: does the agent actually know what you sound like? Does it know your opinions, your real examples, your verified data points? Does it know what you refuse to say?

Early in this process, I was getting articles that were technically correct but felt hollow. Accurate but not mine. The fix was not a better model. It was better input. I spent a few hours writing out my actual beliefs, my verified customer quotes, my specific brand rules. Once that was in the system, the quality jump was immediate.

If you try to run a content machine with a bare prompt, you will get generic content. The more specific context you give, the more it sounds like a real person.

The Quality Check I Actually Do

I am not going to pretend I read every article. I do not. But I do have a quality standard baked into the process:

• No fabricated data. The agent is explicitly instructed: if you do not have a real statistic from the research files or a verified source, do not include one.
• No AI tells. Em dashes, buzzwords, vague filler sentences. These are banned by instruction.
• Internal links. Every article links to enzeti.com or cultivateinbox.com at least once. This is in the instructions and I verify it in the published output.
• Consistent slug format. All slugs follow the calendar. No agent-invented slugs that diverge from the plan.

I spot-check articles weekly. When I find something off, I do not just fix the article. I update the instruction file so it does not happen again. The system gets sharper over time.

What Broke and What I Fixed

A few things went wrong early on that are worth naming.

Duplicate slugs. I had articles publishing to the same slug twice because the agent was not reliably checking the published list. Fixed by making that API call mandatory in the workflow and writing the results to a file the agent reads before picking a topic.

Wrong categories. WordPress category 1 is the default “Uncategorized.” I had a batch of articles go out in the wrong category. Now the instructions explicitly say “NEVER use category 1” and list the correct category IDs with their names.

Voice drift. After a few weeks, some articles started sounding more generic. The fix was adding a “read brand identity first” step as the mandatory first action. The agent re-reads the voice document before every single article, not just at setup.

None of these were catastrophic. They were just friction that I worked out over time.

What This Has Done for the Business

The honest answer is that it is too early to claim major SEO wins. Content takes time to index and rank. What I can say is that we have built a meaningful content library on all three sites in a fraction of the time it would take with a human writer at any reasonable cost.

More importantly, it has forced me to get clear on what I actually believe about each business. You cannot give an agent a real voice without knowing what that voice is. The process of building these content systems made me sharper about positioning, about who I am writing for, about what I refuse to say.

That clarity has already paid off in pitches and sales conversations.

Could You Do This?

Yes. But I want to be honest about what it takes.

This is not a plug-and-play tool you buy and turn on. I built this from scratch. The agents, the cron jobs, the publishing tools, the research files, the brand identity documents. It took real time to set up and it requires maintenance as things break.

If you are thinking about building something similar, start with one site and one article per day. Get the context layer right first. Make sure the agent knows your voice before you try to automate volume. Then scale once the quality is where you want it.

The goal is not to publish a lot of generic content. The goal is to publish a lot of your content. That distinction matters more than the number of posts.

The Bigger Point

I keep coming back to this: AI does not replace what makes your content valuable. It just removes the friction between your thinking and the published page.

I still have to know my audience. I still have to have opinions. I still have to collect real data and real quotes. I still have to make decisions about what matters and what does not. The machine cannot do any of that for me.

What it can do is take everything I have figured out and turn it into consistent output at a pace no human team could match at this budget.

That is the trade I made. And so far, it is working.