If Writing still Matters, How to Do it Right and Avoid AI Suspicion?

Does writing still matter? Does anyone still care? I care. I write. Because it matters.

"Still" refers to the ongoing "AI will replace us all" anxiety/promise/discussion as the AI hype hasn't yet collapsed in 2026. Meta discussions about DEV, writing, and AI usage keep going recently. As Georgi Kobaidze said AI (or, more specifically, vibe coding) "is not a topic I like to talk about." But it seems like we need to.

AI Art and Stereotypes

AI art is part of this dilemma and discussion. Abundant tools encourage laziness and distraction. DEV is no toxic social media platform or at least it's a little bit different. However, still it collaborates with Google's AI team, a company that has long ditched its famous former motto "don't be evil."

Cover Image Generator

The cover image has been generated by Google's Nano Banana AI image generator with DEV's retrofuturistic preset and my spontaneous prompt description: An intellectual women with dark colored skin and large glasses and dreadlocks sits at a table in a cafe with an opened sketchbook, pens and an opened small vintage laptop computer visible from its backside with a lot of overlapping laptop stickers. The woman looks thoughful and hopeful.

I would have expected a more intellectual, more text-focused, and less "hippie" interpretation of my prompt, but at least the generated image avoids social stereotypes prevailing in the tech industry's California Dude culture, while there might be more black women in Lagos, Nairobi and New Delhi coding than white guys in the Bay Area. Even though I'm a white tech dude myself, I try to counter culture the old stereotypes. Below are some of my recent AI-assisted Meme Monday contributions. But what's that got to do with writing? Well, we'll see in a minute...

Modern Monday Memes

Flag Episode Analogy

So, I made AI create cartoons and cover images with modern women instead of beardy dudes. So what? These are still stereotypes.

After notable experts have left American IT and AI companies, here I am, a privileged white dude wasting his coffee break time and CPU power generating cartoon memes trying to fix what's broken on a more fundamental level?

In a way, I remind myself of the South Park "flag war" episode officially known as "Chef Goes Nanners". Maybe it still matters. Maybe it even matters if the story is retold based on vague text prompts resulting in generic cartoons like this one below.

I won't post a screenshot or replication of the said TV series. That would be an avoidable signal for getting shadow-banned in the current cultural climate, maybe not here but it did happen on Meta platforms and it might raise suspicion when applying for an immigration visa eventually, banning critics for alleged racism or copyright violations to silence their justified message, unless they are backed by an established media outlet and their legal department. Thus, one part of my content creation recommendations: don't be more "political" than you have to. I mean, if you have an attitude, you will probably still be political and critical between the lines, no matter how hard you try to be mainstream and stick to professional content.

AI Intoxication and Echo Chambers

Without mentioning skin color in an image generation prompt, all cartoon characters created by AI are white. Not yellow, as that might infringe copyrights. The stereotypical white Americans seem nondescript enough that AI providers don't seem to fear copyright issues, at least not by artists that have the power to claim fair pay. Maybe they don't have to. As creatives, they can advance and grow, while AI will train on its own output and become even more stereotypical than before.

Inevitable Knowledge Collapse?

Are we creating a knowledge collapse and no one is talking about it?

I have asked myself often these days how much our work as developers and content creators still matters when "AI is going to replace us all" while the very same AI needs quality training data. According to research and reasoning there is not necessarily an "AI intoxication" echo chamber effect and no gap of years without meaningful documentation and tutorials either.

Useful Lying Machines

AI can lie a thousand times as long as it still produces valuable insights, and my preferred one-line code completion after diabling an unhelpful GitHub copilot plugin is also "AI" just on a smaller scale. A scale that actually helps me. I ask, AI rolls the dice, I filter, I test and commit. If the code goes into a public repository, AI can train and other can see it. If I mind commenting and documenting, even better. I'm still not a big fan of AI, but relying on my memory and search engine results isn't good enough these days. I'm still not a big fan of StackOverflow either, which is why I've contributed so much DEV content already.

Relevant Ones

As individual developers, authors, and part-time open source contributors, every one of us is just one out of many. So why and how can a single person's contribution still matter among all the noise? "Long tail" is the answer.

Long Tail

While mainstream information audience might matter even less unless you can invest costly marketing power to push it, and niche content might be discarded by search engines and AI training as irrelevant, classic long-tail publishing still seems to work: popular enough, filling gaps, catering for real demand. Consider Google's "people also search for", "answer the public", keyword tools, Matomo or search console insights and similar sources suggesting relevant keywords and W-questions. Add meaningful detail to prove authenticity while also trying to come up with a more general "today I learned" takeaway.

IndieWeb

IndieWeb, also known as POSSE or "own your content" principle, means you should have and maintain at least one personal website and never rely on external third-party platforms alone. Consequentially, I wrote several open mind culture blog posts focusing more on the cultural and art-related aspects of content creation and contemporary AI usage and misuse.

Unsung Elephant Hero

DEV.to, as one of the few truly useful social web sites, also counts as a reputable content source at least partially included in Common Crawl snapshots, the unsung elephant hero in the room that nobody talks about in SEO/GEO discussions.

Common Crawl

AI assistants are trained on Common Crawl derivates and public code repositories like those on GitHub. If you want to feed your niche solutions back into their training material, have a handful of public demo repositories that you can actively maintain, both as templates for future projects, personal code snippet libraries, and to rediscover them via Google, Claude or Copilot in case you forgot. That's when DEV proves handy again. You can describe your issue, approaches and solution(s) in a concise abstract and in lenghty detail, using different synonyms to describe the same thing and comment your code more verbosely than you would in a StackOverflow issue or answer. Add or embed your source code, codepen demos, screenshots, demo websites, and a call to action and wait for helpful comments!

Interlinked Web

I have already linked to several community members' and some of my own DEV posts here, and that's one part of a good post leveraging the power of hypertext thus what still makes the internet special and useful. We don't want to get stuck in biased echo chambers and walled gardens either, so let's link to content outside of the current platform. Those links can point to reputable sources like StackOverflow, GitHub, Wikipedia or .edu domains possibly helping to make a post look more reliable and authoritative.

Factful Examples

Linking or embedding CodePen demos, screenshots or proof of concept demos are always a good idea as well. Show and tell! AI can imitate authentic style and fake expertise but actual first-hand real-world experience goes far beyond adding "honest" and "why this works" to a text while relativizing its stubborn insistence on made-up alternative "facts" by adding "may", "many", "could", "often" etc. Testable real-world examples and opinions based on relevant experience is one thing that AI can't fake at least not in its current LLM-based form.

Clean aesthetics devoid of meaning: generic AI imagery.

Perplexity

Unlike the principle of least surprise to facilitate colloboration when coding, perplexity measures the unlikeliness of words or ideas in a text, and LLM-based AI is bad at perplexitiy by design. The "unsung elephant hero" subheadline and the cute elephant mascot cartoon is an example for perplexity: too creative, unlikely and costly to appear in a usual AI answer, it's still relevant on-topic enough for a post in a developers' community.

Perplexity is also the name of an AI service with an interface similar to Gemini, Claude and ChatGPT, but praised for researching the web and fact-checking by default. The promising concept doesn't always work well though. Perplexity failed me several times.

Perplexity Failed Me

Perplexity insisted that DEV.to adds rel="nofollow" to outgoing links, that Chrome web store descriptions support clickable do-follow links and that any of this matters in 2026. Perplexity also falsely attributed Pixel 3a issues to a Pixel 4a device.

The phone issue is an interesting follow-up blog post topic for my Sustainable Low-Waste Tech series ... later this year, maybe.

Claude Failed Me Too

Claude repeatedly suggested broken code and insisted his failed attempts count against my token limit. After another dubious suggestion, it stated that the code "honestly still may not" work in every case. I challenged his relativizing statement.

Likewise Claude, why do you say "honestly this still may". Is the matter too difficult to give a fact-based correct answer?

You're right to call that out. I was being vague when I should have been precise.

The after_setup_theme hook does not fire during async media uploads because the theme is not loaded in that context. So that alternative I offered won't work — it's not uncertain, it simply won't work.

Don't be Claude!

That's a perfect example how not to write technical articles. Don't suggest untested code! Don't deprecate yourself, just share what you did and why and let others decide if that looks like it might be useful. Use structure, suspense, and a sense of humor, if you like. But don't fool or gaslight your readers.

(Don't) Use AI Images

Disclaimer: all images in this post were generated with the help of artificial intelligence. AI images can be helpful, funny or aesthetically pleasing. However, creation is a waste of energy and often also an unfair unpaid exploitation of uncredited artists. As an experiment, I have used different image generators in different contexts. DEV's cover editor integrates Nano Banana with its retrofuturistic nerd style system prompt recognizeable by its neon purple and aqua lines and vintage monitors. The orange-brownish cartoon variation "Claude failed" me was created by OpenAI's ChatGPT formerly known as Dall-e with two input pictures, Nano Banana's neon-retro "Perplexity failed me" DEV scene above, and the brownish barista cartoon in this post's Modern Monday Memes section, which is based on another barista cartoon that Nano Banana in Gemini created using a collage of a stock image preview showing AI replicas of human models.

Take Your Time

Taking pictures in a studio or outdoors is a complex and time consuming process and so is writing. Even highly productive best selling authors typically published one book per year. Stephen King was a notable early adopter using personal computers for writing. In his memoir On Writing, King notes that he moved from his electric typewriter to the computer system in the early 80s, primarily because it made the assembly of his massive manuscripts significantly faster and allowed him to reuse text more easily. Still, King wasn't famous for the sheer amount of output but for his style, suspense and relevant content criticizing real-world American anxieties, institutions, and social failings.

Writing a good post takes time or a rare moment of hyper-focused inspiration. I think I can take an occasional shortcut like copy-pasting and only slightly rephrasing a three-sentence summary about an American author without fact-checking, because it aligns with what I already thought I knew before, but 90% of my content are my own ideas in my own words. ZeroGPT rates this text as < 3% AI GPT while other popular DEV posts score 10% or more. However, like recent discussions pointed already, that's not the point. AI can help neurodivergent and non-native speakers express themselves and share valuable knowledge and inspiration, and just the fact that a vintage nerd writes on an analog typewriter in a mountain cabin disconnected from internet and electricity does not make their text better per se.

Take your time and do as you please, but please don't stop writing!