The frequent perception is that weblog posts generated by AI are inherently decrease high quality and inferior to their human-authored counterparts.
We imagine that corporations that scale AI-generated content material are doing so knowingly, selecting pace and scalability on the expense of high quality. Whereas we agree that AI is quicker than any human and can go the primary draft, we additionally know that through the use of AI we’re nonetheless sacrificing one thing necessary.
I feel this perception is now outdated. I feel now we have reached a degree the place generative AI can create content material that’s indistinguishable from the huge corpus of human-written content material beforehand created by content material entrepreneurs like me.
AI is extra totally researched, adheres to model and opinion pointers, and is extra versatile, quicker, and extra environment friendly in responding to suggestions. There are now not trade-offs inherent in utilizing AI for content material creation.
Not all AI content material is sweet by default. The obstacles that had been stopping AI content material technology are merely gone. Entry to world-class writing by means of LLMs remains to be uneven, however this would possibly not stay the case for lengthy. Functionally “good” AI content material is simply across the nook for all of us, and it’s in our curiosity to acknowledge that.
Here is why:
Many imagine that human writing has an inherent high quality that AI can by no means attain, and that it has a artistic spark that silicon variations won’t ever attain.
I will not declare that AI will strategy the depths of Shakespeare, however I’ll argue that “good writing” is less complicated and extra mechanical than most individuals assume. Many of the elements of ‘good writing’ are issues that LLMs can do very effectively.
I’ve spent my whole profession making an attempt to develop into a greater author by introspecting my writing course of and asking why some issues work and others do not. I am no professional on this discipline, however I’ve developed an efficient worldview of how writing works and writing ideas I observe again and again.
A small snapshot of an modifying guidelines for coaching new writers.
For instance, under is a random excerpt from my modifying guidelines.
- Have we addressed the obvious objections to this concept?
- Did you utilize the densest language potential (e.g., “new” as an alternative of “one thing new”, “worldwide” as an alternative of “globally”)?
- Did I exchange Itachi’s phrases with a concrete instance? (e.g., “enterprise outcomes,” “specialists imagine,” “analyzing information,” “making selections,” and so forth.)
- Can we keep away from making tough issues sound straightforward?
- Do you begin with a very powerful info? (Introduction, starting of paragraph)
- and so forth.
These ideas are how I write, how I edit my writing, and the way I educate writing. They’re quite simple, however finished in unison, they find yourself being nice on a reasonably common foundation. great,I’m writing.
In reality, these ideas are so easy that LLMs can execute them completely and normally higher than I can. I typically apply these ideas erratically as a consequence of fatigue, boredom, laziness, and so forth. Nevertheless, for an LLM, these ideas could be set as soon as and adopted indefinitely. Scales evenly to lots of or 1000’s. tens of millions of The content material of the output codified in system prompts and SKILL recordsdata (these are mentioned in additional element within the subsequent part).
If we settle for {that a} primary recipe for good writing exists (and I imagine it does), then an LLM can efficiently observe it. Many of those heuristics could be chained collectively in a dependable approach to create AI creation course of.
And eventually, now we have the know-how to make it potential.
For a lot of, the worldview of AI remains to be rooted within the chat expertise. Nevertheless, LLMs, and extra importantly the infrastructure surrounding them, have come a great distance in current months.
Even of their early days, large-scale language fashions confirmed brilliance of superhuman genius in small areas. However like a precocious youngster who imitates his mother and father with none understanding of his personal actions, it was tough to think about that these sparks would develop right into a flame of real writing means.
Reliably producing 1000’s of phrases of correct, helpful, concise, on-brand writing with only a few constant sentences appears a great distance off. Establish and fill subject gaps, perceive key search intent, differentiate your articles out of your rivals, and extra.
Once I wrote about it final time, While the AI writing process (using a custom GPT based on my editing principles) showed a lot of brilliance in the output, human intervention was still required to create the final product.
An early version of the AI content pipeline built using the ChatGPT project and custom GPT.
But that’s no longer the case. After just seven months, that process no longer has any limitations. Now, my $20/month Claude subscription gives me access to almost sci-fi-like capabilities. I can:
- Chain multiple LLM processes into one continuous workflow (Claude Code, OpenAI Codex, and other agent models).
- It provides guardrails to avoid much of the stochastic “wobble” that LLMs typically see when trying to follow processes (skills), and encourages recursive performance benchmarking and self-improvement.
- Integrate AI into existing workflows across other tools (MCP).
- Create content based on your research, existing writing samples, tone, and brand guidelines (RAG, memory, context).
(And this ignores the significant improvements that the flagship models themselves have shown in recent years.)
Part of a custom SKILL file that I built for the Claude code.
All the vibecoding infrastructure developed over the past year has had a transformative impact on the overall utility of LLM. While LLM is still “just” a sophisticated autocomplete and certainly hasn’t achieved AGI, companies like Anthropic and OpenAI have managed to leverage its behavior in ways that seem far more useful than the sum of its parts.
And importantly, the task set before them, content marketing, is not particularly complex.
Most content material entrepreneurs spend most of their time creating informational content material that targets key phrases, i.e. useful “how-to” articles and content material. Comparison list. These are proven content marketing archetypes and are generally very easy to create.
As always, we believe there is a basic recipe for effective search content. Here are some of the core principles we try to follow in our search content:
- Address key search intent
- Build on consensus of existing search results
- Fill in the obvious topic gaps between your articles and your competitors’ articles.
- Go beyond existing results and add new and novel information.
- Reference existing, related content you’ve already created around the topic.
- Reference relevant external content that may help readers continue their exploration.
- Prioritize topics that naturally reference your product
- Make sure your article has the following structure: mutually exclusive, collectively exhaustive
- Ensure that the construction of your article really delivers on the promise of the title
- Seize the reader’s curiosity with the title and introduction
- Embody key phrases and key phrase variations naturally in necessary components of your articles
- and so forth.
These are equally easy ideas that may additionally provide help to arrive at efficient search content material. When you can observe these processes, your search content material will normally work effectively. The identical goes for LLMs. If Opus 4.6 or GPT 5.4 can observe these processes, their output can even work effectively.
Even essentially the most opaque of those processes is pretty straightforward for LLMs to observe by offering express steps to observe (“I exploit WebFetch to run my website. Search ahrefs.com/weblog and return the primary 3 articles…”), examples of the specified output (resembling a reference file on your favourite article introduction), or entry to an authoritative information supply (resembling Ahrefs MCP).
Preview of SKILL recordsdata to retrieve current Ahrefs content material for specified key phrases.
Though we would want in any other case, efficient search content material could be very formulaic (therefore the success of the skyscraper methodology). It does not require nice complexity or novelty. No have to disagree with poetry or SERPs both.
There’s some room for innovation and experimentation, however lower than you may think. Deviating too far exterior the Overton window normally hurts efficiency as an alternative of enhancing it (I say this after many failed makes an attempt to create “good” search content material).
If Claude can refactor a 100,000-line codebase, it appears conceited to imagine that a big language mannequin cannot create nice, search-optimized content material. AI can’t write Shakespeare, nevertheless it doesn’t have to.
remaining ideas
Whether or not I handle to persuade you or not, as of this writing, a good portion of my position has already been delegated to generative AI. I am utilizing Claude code. Ahrefs MCP and a set of about 15 customized SKILLs chain collectively in sequence to replace previous articles and create helpful, high-quality content material.
Claude begins the content material replace course of.
These articles sound the identical. they do the identical factor. They embody my experiences and views. They’re simply nearly as good as something I’ve written. In any other case I would not have had time to create it. There aren’t any trade-offs.
There’s nonetheless an enormous hole within the high quality of what is potential between a talented author who takes full benefit of generative AI and the typical beginner who prompts ChatGPT to “write a weblog put up.”
Nevertheless, this hole is now a lot smaller than earlier than. It is going to be shut down in the long term as AI platforms proceed to democratize entry to all of this nice performance. The “Content material Engineer” talent will probably be one of many workflows in all main LLM platforms. Functionally “good” AI content material is simply across the nook for all of us.
There are nonetheless many components of my job that may’t be outsourced to AI, so I am comfy having this dialogue. Even when they might, some individuals do not need to do it (like this text).
Provided that we expect truthfully about the place AI can and must be used will we discover a path ahead. Till lately, there wasn’t sufficient AI content material. Nicely, that is proper. The earlier you may acknowledge that, the extra time you need to deal with the components of promoting that may maintain people pleased for longer.
(I additionally do not miss writing content material about skyscrapers.)
