Writing and optimizing website content is the single most effective way to get recommended in AI search. It makes up about 70% of the work we do at TJ Digital, where we manage AI-driven content strategies for roughly 40 to 50 websites across industries like immigration law, cybersecurity, finance, and education. We’ve yet to find an industry where AI couldn’t produce content at the same quality level as a subject matter expert writing it themselves.
But it takes real effort. If you just open ChatGPT and ask it to write a blog post, the result is unlikely to rank, and doing that at scale could get your site penalized. The difference between AI content that performs and AI content that hurts you is the system behind it.
This article walks through the exact six-step process we use to go from a topic to a publish-ready article.
Why Does Most AI-Written Content Fail?
When I explain our process to prospective clients, I usually get one of two reactions. Some say AI could never write their content because their industry is too complex and requires too much expertise. Others assume we’re just having ChatGPT crank out articles, and they figure they can do that themselves.
Both reactions miss what actually matters. AI can absolutely write high-quality content for complex industries. We do it every day.
The problem is that AI doesn’t know anything about your specific business, your customers, or the way you talk about what you do. Without that context, it fills in the gaps with generic filler.
That’s why a quick ChatGPT prompt produces content that reads like it was written by someone who skimmed your website for 30 seconds. It’s technically accurate but completely forgettable. Search engines and AI models can tell the difference.
Does the Topic You Write About Actually Matter?
The topic matters more than most people realize. If you’re writing about something that AI search engines aren’t actually looking for, or if you’re targeting terms that are far too competitive for your brand, you’re wasting your time.
We use Peec.ai to see which pages AI models are already citing in their responses. From there, we look for patterns in those topics and identify which ones would be relevant for a specific client. The goal is to find topics where the client has something real to say and where AI search is actively pulling from similar content.
I have other posts that go deeper on topic selection. The short version is that picking the right topic is a prerequisite. Everything else in this process falls apart if the topic doesn’t have demand.
What Is a Brand Knowledge Base and Why Does AI Need One?
The main reason companies feel like AI can’t write their content is that AI doesn’t have the complete understanding of their business. It doesn’t have the tacit knowledge that comes from years of experience in a specific industry.
So you give it that knowledge. You build what we call a Brand Ambassador.
A Brand Ambassador is typically about a dozen markdown files, though it might also include PDFs or spreadsheets. It contains everything the AI needs to write as if it were someone on your team. That includes your company overview, each product and service in detail, your unique value proposition, your ideal customer profile, every important page on your website, your brand voice and tone, examples of content you’ve written, the canonical version of any guides or processes, and visual style guidance for AI-generated images.
Don’t try to collect all of this at once. Start with what you have, then assume you’ll be iterating on it forever.
Two years from now, this type of documentation will be a standard operating procedure. The primary audience for documentation like this is AI. When humans need information, they ask the AI.
We currently store Brand Ambassadors in Google Drive so they connect directly to Claude. Eventually, these knowledge bases will integrate directly into a client’s internal systems. If you want a more detailed walkthrough of how to build one, we wrote a full guide on AI documentation for your business.
How Does the Article Brief Step Work?
The Brand Ambassador is half the puzzle. The other half is the instructions. At our agency, those instructions take the form of Claude skills, which are structured prompts that tell Claude exactly how to approach each step.
The first skill generates an article brief. We give it the topic and any additional information we have. Sometimes the client has given us notes, a voice memo, or a video transcript on the topic.
Other times we just have the keyword. Whatever we have, we feed it all in and let Claude go.
Claude then searches the web for the target term and starts doing research. It’s trying to answer three questions:
- What’s the intent behind this search term, and what format does the searcher expect?
- What information needs to be collected to make the article thorough enough to compete?
- What specific insights from the client need to anchor the article’s point of view?
Those three answers shape everything that follows. The brief becomes the blueprint for the article, and every subsequent step references it.
What Tools Work Best for AI Content Research?
The next step is deep research, and we currently use ChatGPT’s deep research feature for this. As of right now, it’s still better than Claude’s research capabilities for this specific task.
A deep research session takes about 10 to 15 minutes and produces a comprehensive document that the first draft skill will reference. The research covers data points, statistics, competitor analysis, and broader context that the article brief identified as necessary.
| Task | Best Tool | Why |
| Deep research and source gathering | ChatGPT | Processes hundreds of sources quickly and produces comprehensive research documents |
| Article brief and intent analysis | Claude | Better at structured reasoning, following complex multi-step instructions, and working within a brand knowledge base |
| First draft writing | Claude | Produces more careful, nuanced long-form content with fewer hallucinations |
| Revision and guideline compliance | Claude | Stronger at comparing a draft against a detailed set of content rules |
| Fact-checking and link verification | Claude | More reliable at systematic claim-by-claim verification |
| Image and infographic generation | ChatGPT | Better multimodal capabilities for turning text descriptions into visual prompts |
The key insight is that no single AI tool does everything well. We use each tool for what it’s best at, then pass the output to the next step. The research document and the article brief together become the inputs for the first draft.
How Do You Write the First Draft with AI?
The first draft skill is one of our largest. It takes the article brief and the research document and produces a complete article.
This skill contains a detailed description of what success looks like for the article. It includes SEO guidelines for heading structure and keyword placement, rules for maximizing fact density (making sure every sentence carries real information), and instructions for structuring the article to maximize the chance of getting cited by AI search engines.
Throughout this entire process, Claude has access to the Brand Ambassador. That knowledge base shapes the tone of voice, determines which internal pages to link to, and provides the client-specific data and examples that make the article feel like it came from someone who actually knows the business.
The first draft is exactly that. Every article still goes through two more steps before it’s ready to publish.
How Does the Revision Process Catch Mistakes?
The first revision skill compares the draft against all content guidelines we’ve created for this brand. Those guidelines come from every piece of content we’ve previously reviewed and received feedback on.
This is one of the most important parts of the system. Anytime we or the client give feedback on an article, that feedback gets codified into a living document that the revision skill checks every new draft against. If a client once told us they don’t use a certain phrase, or that they prefer to describe their product a specific way, that correction applies to every article going forward.
The second revision skill is a dedicated fact-checker. It makes sure every factual claim in the article is backed up by a credible external source and verifies that we’re never linking to competitors. It also confirms that every link actually works and supports the claim it’s attached to.
Between these two revision steps, the article gets checked against brand accuracy, stylistic consistency, factual correctness, and link integrity. Most errors that slip through a human editing process get caught here because the AI checks every single line against every single rule, without getting fatigued or skipping anything.
How Do You Add Images and Infographics to AI Articles?
The final step is a skill that reads through the finished article and extracts descriptions for infographics. We typically add five or six per article. Each description specifies what the graphic should illustrate, what data to include, and how it should be styled.
Those descriptions then go to ChatGPT to generate the actual images. This works well because ChatGPT’s image generation capabilities handle detailed prompts effectively, especially when you provide specific brand colors, style preferences, and layout instructions.
The result is an article that’s both well-written and visually complete. The infographics break up the text, reinforce key points, and give the article a polished, professional look.
Can AI Really Write Content for Complex Industries?
Yes. We create content for immigration attorneys, cybersecurity firms, financial advisors, and education companies. The content quality depends entirely on the Brand Ambassador.
If the knowledge base is comprehensive and the client’s expertise is well documented, AI can produce articles that match what a subject matter expert would write.
The gap shows up when companies skip the knowledge base step and expect AI to fill in industry-specific context on its own. AI is very good at following instructions and synthesizing information from sources you provide, but very bad at making up expertise it doesn’t have. The fix is always better inputs.
Will AI-Written Content Get Penalized by Google?
Content gets penalized when it looks like low-effort output produced at scale. Google’s spam detection targets patterns associated with mass production, like repetitive sentence structures, thin content, and lack of original insight. If your AI content avoids those patterns, Google does not care whether a human or an AI wrote it.
Our agency manages content for roughly 40 to 50 websites, and none of our clients were hit by Google’s March 2026 spam update. Several actually gained traffic because competitors with lower-quality AI content got penalized. I wrote more about this in a post on AI for blog posts.
How Long Does This Process Take Per Article?
The full process takes between two and four hours of human time per article. The deep research step runs for about 10 to 15 minutes on its own, and the Claude skills each take a few minutes to execute. The bulk of the human time goes into reviewing and refining each step’s output before passing it to the next skill.
That’s dramatically faster than writing an article from scratch, which typically takes 8 to 15 hours for a well-researched, long-form piece. The quality is comparable because the system forces rigor at every step.
Start Building Your AI Content System
The process described in this article is the same one we run for every client at TJ Digital. Building it took thousands of hours of iteration across dozens of industries, and we’re still refining it every week.
Contact TJ Digital for a free digital marketing audit. We’ll show you exactly where your content stands and what it would take to build a system like this for your business.