AI blog writing got much better in less than one year. That is my experience, not a claim that every AI draft is now good. Less than a year ago, I could ask AI to write an article quickly, but the result was generic, poorly structured, and disconnected from the company. Today, I can use Codex, local project context, reusable Skills, online research, and an interview process to create something much more tailored.
The most important change is not that AI can produce more words. It is that I can now give the AI agent enough context to help me turn my own experience into an article that sounds like me.
I still review every draft. I correct anything inaccurate, unclear, or too generic. I provide feedback until I am happy with the article, and I decide when it is ready to move forward.
AI has made the process much faster for me. It has not removed my responsibility for the final result.
A year ago, speed was not enough
Around a year ago, I was working for a company and wanted to test whether AI could help us create blog articles faster.
The opportunity was obvious. Instead of starting with a blank page and spending hours building an article, we could ask AI for a draft and accelerate the process.
But the quality was not good enough. The drafts were too generic. The structure was often weak, important information was missing, and the writing was not tailored to the company.
I do not mean that the drafts only needed a final polish. The deeper problem was that they repeated information already available online. They did not contain enough company-specific knowledge, experience, or point of view to justify publishing another article.
I kept experimenting because I expected the technology to improve. I wanted to understand its limitations early, learn what it could already do, and be ready when the workflow became more useful.
I had wanted to publish my experience for years
My interest in AI writing was also personal.
For a long time, I wanted to document my experience and share what I had learned. Writing gives me a public record of my knowledge. It also lets other people see how I think and what I have worked on.
The problem was time. A complete article can easily reach around 1,500 words. English is not my first language. I can write in English, but turning an idea into a clear, well-structured article could take me too long.
Many ideas stayed in my head because the distance between having the experience and publishing the article was too large.
AI looked like a way to reduce that distance. The early tools made drafting faster, but the result still did not feel personal enough. Today, the process feels very different.
What changed was more than the model
The models are better, but I do not think the model alone explains the difference.
I now use Codex inside the ChatGPT desktop app. OpenAI’s documentation explains that when I open a folder, ChatGPT can use the files and context in that location. That matters because the agent does not have to work from one isolated prompt.
My local project can contain the content strategy, topic bank, existing articles, voice rules, SEO decisions, company information, research notes, and publishing requirements. The agent can read the relevant material before it starts writing.
I can also use Skills. OpenAI describes Skills as reusable workflows that package instructions, references, and optional scripts. For me, a Skill is the process I want the AI agent to follow each time I create an article.
The combination is what changed my experience. The agent has a better model, a better place to work, more relevant local context, and a repeatable process designed around how I want to create content.
AI did not replace my experience. It gave me a much better way to turn that experience into an article.
My current workflow starts before the draft
I do not begin by asking the agent to write 1,500 words about a broad topic.
My current process starts much earlier:
- Check the content strategy and topic map. The article should have a reason to exist and should not compete with something I already planned.
- Research what people are asking. I want to know whether the question appears in search, forums, professional discussions, or buying conversations.
- Validate the search intent and keywords. Keyword research helps me understand the language and opportunity, but it should not invent a generic angle for me.
- Interview me one question at a time. My answers give the agent the experience, opinion, examples, and judgment that it cannot find in a general search.
- Create and review the draft. I read the article, give feedback, and keep revising until it reflects what I want to say.
- Move to visuals and WordPress only after approval. The article remains a draft until I approve the writing and the rest of the package is complete.
This article followed that process. It started as a rough idea about how different AI blog writing feels compared with a year ago. The agent checked my topic bank and content map, researched the questions people ask online, proposed three directions, and then interviewed me.
I could answer those interview questions between other tasks. I did not need to find a free afternoon to write the entire article from the beginning.
The Skill keeps the process in one place so I do not have to explain it every time. I have written separately about why Codex Skills save me repetition but not review. The important point here is that the Skill gives the agent a process, while my answers give it the substance.
SEO also changed what is worth writing
The writing workflow is not the only thing that changed. Search changed too.
Many SEO strategies put a lot of weight on broad top-of-funnel keywords because those articles could attract large amounts of informational traffic. That approach is weaker now. AI tools and search summaries can answer many common questions without requiring someone to visit another general article.
Google’s current guidance does not say AI-assisted content is automatically bad. It says that generating many pages without adding value can violate its scaled-content policy. Its newer AI-search guidance goes further and recommends first-hand, non-commodity content that provides more than common knowledge.
Research points in the same direction. Ahrefs found that AI Overviews appear especially often for question-based and informational searches. That creates more competition for the click, although useful informational content can still earn citations and support a topic cluster.
I would not say top-of-funnel SEO is impossible. I would say generic top-of-funnel content should no longer be the default investment for a small business.
I would put more attention on specific questions connected to a real decision. That can include comparisons, product fit, service limitations, implementation questions, customer objections, local considerations, and the tradeoffs a buyer wants to understand before taking the next step.
One B2B SEO specialist writing for Search Engine Land describes a similar shift. She says she now puts most of her effort into mid- and bottom-of-funnel content, while keeping differentiated top-of-funnel articles as supporting structure.
Bottom-of-funnel content is not automatically protected from AI summaries either. Semrush found that AI Overviews on commercial-intent searches grew during its recent study. Choosing a lower-funnel keyword is not enough. The article still needs information that comes from the company and is genuinely useful to the reader.
That is where local context becomes valuable. Codex can help turn the company’s knowledge into useful content, but the company still has to provide that knowledge.
I still approve every article
A better workflow does not mean I publish whatever the agent writes.
I review the facts, sources, wording, examples, structure, and tone. If something does not sound like me, I say so. If a claim is too broad, I narrow it. If a paragraph is confusing, I ask for a simpler version.
The AI agent can suggest corrections, but I decide whether those corrections improve the article. The same applies to the final title, SEO information, visuals, and WordPress draft.
The review also improves future work. When I notice a problem that could happen again, I do more than correct the current article. I update the Skill that owns that part of the process.
That happened while planning this article. We agreed that article titles should normally aim for 45 to 60 characters when the wording remains natural. We also added a visual rule so the featured-image title should work in two or three readable lines. The title of this article is 53 characters.
That small decision will now help future articles too. That improvement now applies to future articles as well.
I think this is the right time to start
I expect AI writing tools to keep improving. I have already seen a dramatic change in less than a year.
That does not mean someone should wait until the process becomes perfect. The current tools are useful enough to begin learning how to work with them now.
My preferred starting point is a simple Skill that reflects the way I want to create an article. It can include question research, keyword validation, the interview, the writing style, the review steps, and the approval steps. It does not need to be perfect on the first attempt. Real use will expose what needs to improve.
This can help a founder, marketer, creator, small-business owner, or non-native English speaker who has useful experience but has struggled to publish it consistently. The AI can help with the process and the language. The experience and judgment still have to come from the person or company.
After years of wanting to document what I know, I can now develop articles through focused conversations between my other tasks. That is the change I care about. It is not more content for the sake of producing content. It is a practical way to share my ideas, keep my personal site active, support search, and build awareness around the work I actually do.
Less than a year ago, the AI blog-writing process was not good enough for me. Today, it is.