My background is in XR development — 13 years building Unity applications, VR training simulators, AR experiences. Content creation was never my primary skill. But over the last couple of years I have used AI tools to build this entire website, write all the blog articles on it, produce two children's books with AI-generated illustrations, and write documentation and client-facing material for XR projects. AI has made content creation genuinely accessible to someone who builds things for a living rather than writes about them.
This is not a theoretical guide about what you could do with AI for content. This is what I actually do, the tools I actually use, and the things I have learned from building real content with these tools over an extended period.
The Most Important Thing I Learned First
AI will not replace your perspective. It will replace the blank page, the structural decisions, the time spent on first drafts, and a lot of the mechanical editing work. But the most useful thing in any piece of content is always something the AI cannot supply — your specific experience, your honest opinion, your knowledge of your audience. The articles on this site that perform best are the ones where I brought something real to the table and used AI to shape it into readable prose, not the ones where I handed a topic to AI and published what came back.
Start with that understanding and everything else falls into place.
My Actual Content Creation Workflow
Step 1 — Start with What I Actually Know
Before I open any AI tool, I think about what I actually know about the topic. For the AI tool comparison articles on this site — ChatGPT vs Claude, best AI image generators, best AI coding tools — I started by writing down what I had genuinely experienced. Which tools I had actually used. What worked. What did not. What surprised me. That raw personal knowledge is the most valuable input I give any AI tool. Without it, the output is generic. With it, the output reflects real experience.
For the children's books, I knew what kind of story I wanted — the character, the moral, the tone. I knew how I wanted the illustrations to feel from my experience with MagicLight AI images. That knowledge shaped every prompt I gave the AI.
Step 2 — Research with Gemini
For topics that require current information — recent AI tool releases, current pricing, recent developments in a field — I use Gemini. Its Google Search integration means it pulls current data rather than relying on training data that might be months old. For the IPL prediction articles I write, Gemini helps me quickly pull together team form, recent results, and current points table data before I write the analysis.
For topics where I already have the core knowledge and just need to fill gaps or check facts, I use Claude directly with web search turned on. Both work — I choose based on which tab is already open.
Step 3 — Structure and Draft with Claude
Claude is my primary writing tool for this site and for most long-form content. I give it the raw material — my personal observations, the topic, the audience, the specific angle I want to take — and it helps me shape that into a readable, well-structured piece. The key word is "helps." I do not paste a topic and publish what comes back. I give it my perspective and it helps me express it clearly.
For the blog articles, my typical process is: describe the article I want to write including specific things I personally experienced, ask Claude to draft it, then rewrite sections that do not sound like me or that miss nuance I know from experience. The final article is genuinely mine — Claude helped me produce it faster and more cleanly than I could alone, but the knowledge and perspective in it came from me.
For the children's books, Claude helped me with the story structure, the dialogue rhythm, and the pacing. The creative direction — the character, the world, the moral — came from me. Claude shaped it into something a child would actually enjoy reading.
Step 4 — Images with MagicLight AI and Leonardo AI
For illustrated content — the children's books specifically — MagicLight AI is my tool. The warm, glowing illustration style it produces is exactly what children's content needs, and the style consistency across 30+ illustrations per book is genuinely impressive. I described the character, the setting, and the scene for each page, and MagicLight delivered images that felt part of the same visual world from page one to the last page.
For concept imagery — XR project references, article header images, visual ideas to communicate to clients — I use Leonardo AI's free tier. Faster iteration than MagicLight and better suited to the kind of reference and concept work that supports development projects rather than publication-ready illustration.
For background removal and image enhancement — particularly cleaning up character illustrations from MagicLight before placing them in book layouts — Pixelcut AI handles this cleanly. The background removal quality on fur and complex edges is significantly better than manual masking.
Step 5 — Edit With Your Own Eyes
Every piece of content I publish goes through a final read where I am specifically looking for things that do not sound like me, claims I cannot personally verify, or anything that feels generic. AI writing has recognisable patterns — over-structured sections, a tendency to hedge everything, certain phrases that appear repeatedly. I remove these during editing.
The specific things I look for: sentences that start with "It is worth noting that..." or "In conclusion..." or "This is particularly true when..." — all AI filler. Paragraphs where every sentence qualifies the one before it. Lists where prose would read better. Advice that is technically correct but so general it is useless to anyone.
I also use this read to inject things AI cannot — personal examples, specific numbers from my own experience, the kind of honest "this did not work as well as I hoped" observation that makes content credible rather than promotional.
Content Types and Which AI Tool I Use
The Mistakes I Made Early On
Publishing without editing was the biggest one. The first few pieces of content I produced with AI assistance went up with minimal human review. They were coherent and well-structured — but they were also noticeably generic. The kind of content where every sentence is correct but nothing is memorable. The articles that actually get traction on this site are the ones where I brought real specificity — the Three.js project that made me switch from ChatGPT to Claude, the specific experience of generating 30+ consistent illustrations for a children's book. Generic AI writing cannot produce that.
The second mistake was expecting AI to know my audience better than I do. AI can write for a broad general audience. It cannot write for the specific developers, XR practitioners, and AI-curious professionals who actually visit this site. Only I know what that audience cares about, what their common questions are, and what level of technical detail they expect. I have to supply that — AI executes it.
What AI Cannot Do for Content Creation
AI cannot have opinions. It can present balanced perspectives, summarise debates, and explain positions — but it does not genuinely hold views. The moments in content where a real perspective comes through — "I tried Cursor for Unity and the gains were more modest than I expected" or "MagicLight AI's style consistency across a full book impressed me more than anything else I tested" — those cannot be generated. They have to be experienced and then written down. AI can help you express them clearly. It cannot supply them.
AI also cannot maintain a relationship with an audience over time. The trust that builds between a writer and readers across multiple articles, the sense that this person knows what they are talking about and will tell you the truth — that comes from consistent authentic perspective, not from consistent writing quality. Both matter, but only one can be outsourced.
Frequently Asked Questions
Google's stated position is that it rewards helpful, high-quality content regardless of how it was produced. In practice, purely AI-generated content with no human expertise or editing typically lacks the depth and specificity that ranking systems reward. The content on this site is AI-assisted — my knowledge and perspective goes in, Claude helps shape the output, I edit and add specificity. That combination performs. Generic AI content with no human layer does not.
Claude is my choice — it produces more natural, readable prose than any other tool I have tried, and it handles long-form content with better consistency. The key is giving it real material to work with. A detailed brief with your specific experiences and perspective produces something useful. A vague topic produces something generic.
Give it specific real inputs — your personal experience, specific examples, honest assessments of what worked and what did not. Then edit the output to remove AI filler phrases, add your voice, and inject the specific details only you can provide. The more specific the input, the less generic the output. And always read the final piece out loud — you will immediately hear what does not sound like a real person.
It can produce the text — story structure, dialogue, pacing — if you give it the creative direction. The character, the world, the moral lesson, the tone all need to come from you. Claude helped me with two children's books and the text is good. The illustrations came from MagicLight AI based on scene descriptions I wrote for each page. The overall creative vision — what the books are about and how they should feel — came from me. That is the right division of labour.