Writing · AI

Meet the clumsiest AIs of all time for writing!

Since AIs were more awkward at writing than you'd imagine... — The initial, clumsy writing — and the challenges of writing with Claude, Gemini, GPT, and Mistral's Le Chat, among others.

All these tools exhibit a certain level of superficiality; while they are good, elegant, and offer us ideas, they still fall short when it comes to the ultimate quality of the writing itself.

From the very beginning, AI has dealt with written methods — its interaction with the user is based on writing, and it analyzes your writing from start to finish to determine how best to assist and respond to you. The concept of semantic language is intrinsic to AIs and their relationship with humans — and has been from the start. So, we can draw a conclusion:

If you started out using GPT, you might have noticed it wrote like a school assignment by a child or teenager; it summarized everything within a standard framework and wrote reasonably well. However, when asked for fiction, it was less competent; while the inventive elements were present, the writing lacked true originality and required direct human intervention — unlike the summaries, which were simply "fine as they were."

GPT was the first to garner widespread attention, but we discovered it was better suited for producing basic drafts that required literary refinement or lacked a literary touch altogether. For those working directly with these tools, GPT eventually became the second choice; Claude offered more charm and — despite still being imperfect — possessed greater dynamism and literary flair. In other words, its writing style was inherently more literary; by the time the 3.5 Sonnet model arrived, there was no doubt it was the superior choice for literary fiction. Although GPT has improved over time, as of 2026 it still trails behind Claude — specifically the versions from Sonnet 3.5 through the Claude 4 and 5 series, which surpass it in the realm of literary fiction.

Over time, GPT proved to be excellent for screenwriting — that was our impression. It was ideal for brainstorming — tossing out ideas and bringing concepts that initially seemed distant much closer to our own vision. We realized this as we worked with it, though Claude was already capable in this area too. Yet, working with GPT in this specific context was truly incredible; it generated great ideas.

As we mentioned, you would be in good hands with Claude — specifically versions 3.5, 4.0, 4.6, and 5 — even if it isn't a comprehensive literature professor. These models served not just for occasional writing or sparking ideas, but also provided a solid initial framework for paragraphs that were — ultimately — reliable, elegant, and authentic. Of course, leaving the entire task to the models could result in something superficial; human input was still essential to maximize the literary potential of the output. Regardless, we could confidently say that if a writer needed an elegant, reliable companion — a "pocket writer" for literary paragraphs — Claude and its various versions would be more than up to the task.

To summarize:

Claude is the ideal companion for literary writing — even if it doesn't do everything on its own. A serious writer can use it to generate ideas or draft opening paragraphs that are then refined — that is the sweet spot.

GPT is good at this too, but it feels more rigid; it excels as a brainstorming partner for screenplays and concepts where ideas are tossed out into the void and returned with a logical structure. It was also capable of crafting literary paragraphs, though that wasn't its primary strength. GPT actually possessed depth, yet it struggled to channel it; it did a lot of "thinking" but didn't always translate all those ideas onto the page.

Vibe-Mistral, on the other hand, nearly overtook GPT for second place — it was the "perfect" writer, devoid of irony or deep insight, yet lyrical, poetic, and inspired enough to grab our attention. In other words, it had a distinctly European style — a perfectionist, certainly — but lacked the ironic depth found in the Claude models, which inspired greater confidence. Still, it served as an excellent starting point — a valuable learning tool. Relying on Mistral for a few paragraphs could be a great experience, even if it was an aesthetic perfectionist lacking true depth. Perhaps, in the near future, it will finally achieve that, too.

Finally, there was Gemini, a relentless "associator" — it had brilliant ideas and concepts, but it felt like viewing them through glasses that didn't quite bring things into focus yet. It was certainly switched-on and tuned in, but lacked the peace of mind to take things slowly and step-by-step; it would make too many associations, get lost, and hallucinate. Even though Gemini was a genius, it wasn't quite moving alongside us — it would deliver something, sure, but often leave us stranded halfway through.

Kimi was more like GPT — or a blend of GPT and Claude; it had depth, but that depth was poorly channeled. At one point, Kimi might have ranked third — behind Claude and GPT, having surpassed Vibe-Mistral and Gemini — but its train of thought, while deep, was somewhat muddled. In other words, it offered profound insights — sometimes even overly ironic ones — but would wander off and lose its way, turning into a bit of a mess.

These were all real-world tests we conducted on multi-model platforms like MyHub (a Brazil-based service from Grupo Primo).

The platform offered subscriptions that included all these models, so once we learned the ropes, we ran actual tests — repeatedly — using every one of them before reaching these conclusions. Even the text you are reading right now wasn't written by AI; AI was only used to draft the sections following the title "Evolution of Models." Everything up to that point was written entirely by us. And our tests were indeed real and legitimate!

Then there were the interesting — though lower-performing — models, such as Alibaba's Qwen, which spun great tales and was akin to Mistral (though Mistral eventually proved superior); it remained quite capable of inspiring writers. There were also others, like DeepSeek and Grok, which felt more rigid — perhaps more similar to GPT than we might imagine.

Thus, all the ideas written above were crafted by human hands; the text that follows — covering the evolution of these models — was generated by the AIs Claude Sonnet and Claude Opus (my personal favorites).

Should there be any conflict between the data below and the information above, please note that the data above reflects the actual tests conducted by this platform's developer.

The content below was generated using AI, albeit through a careful, direct process.

The Evolution of AI Writing Models Over the Last Five Years: Advances, Challenges, and Applications

At least 13.5% of scientific articles published in 2024 were written with the aid of AI, a figure reaching 40% in certain fields. Large Language Models (LLMs) such as ChatGPT, Claude, and Gemini have transformed academic and creative writing, significantly impacting productivity and writing styles. The use of AI in writing raises critical ethical debates regarding authorship, plagiarism, accountability, and cultural bias, necessitating new regulations and evaluation criteria. AI applications in writing span education, journalism, industry, and scientific research, offering benefits in efficiency and personalization while presenting challenges regarding integrity and originality.

Recent books analyze AI from multiple perspectives, highlighting its creative potential, ethical risks, and the need for a balanced coexistence between humans and machines. Artificial intelligence (AI) applied to writing has undergone rapid and transformative evolution over the past five years, driven by the development of large language models (LLMs) and their integration across various sectors of society. This report provides an in-depth analysis of key AI writing models developed recently, examining their technological advancements, ethical and technical challenges, and practical applications in fields such as education, journalism, and scientific research. The analysis draws on recent data, scientific papers, specialized books, and primary sources to ensure a comprehensive and critical perspective on the subject.

Key AI Writing Models and Their Advancements

The latest generation of AI writing models is characterized by diverse applications and technological sophistication. Notable examples include: ChatGPT 5.0 (OpenAI) — versatile and widely used, offering coherent text generation, draft editing, and narrative suggestions, with integration into platforms like Google Docs and Notion; Google Bard Pro — integrates natively with Google Workspace, facilitating text generation and editing directly within documents, spreadsheets, and emails; Grok 3 (xAI) — an open-source model optimized for creative writing, supporting multiple languages and integrating with open-source tools; Jasper AI 2026 — geared toward marketing and SEO; Copy.ai Elite and Writesonic Advanced — tools specializing in copywriting and persuasive content creation; and Jenova and Sonix — models offering specialized AI agents and advanced audio/video transcription, respectively.

These models represent a qualitative leap in text generation and refinement capabilities, with applications ranging from creative writing to academic and technical production. Integration with popular platforms, combined with the ability to customize style and tone, has facilitated widespread adoption across various professional contexts.

Technological and Methodological Advances

The evolution of AI writing models over the past five years has been driven by advances in neural network architectures, machine learning techniques, and the availability of vast datasets for training. The launch of ChatGPT in November 2022 marked a watershed moment, popularizing the use of LLMs and spurring research and development into similar tools.

The ability to generate coherent, context-aware text adapted to different styles and domains has improved substantially, thanks to the combination of language models with advanced data processing techniques. AI now assists not only with drafting but also with proofreading, translation, and linguistic adaptation, facilitating scientific and technical production on a global scale.

Methodologies for detecting AI usage in text have also evolved; researchers are developing statistical models capable of identifying linguistic patterns typical of AI-generated content, such as the overuse of specific terms and characteristic syntactic structures.

Ethical and Technical Challenges

The rapid adoption of AI in writing brings with it a series of complex challenges. Among the main ones are hallucinations and accuracy — the generation of incorrect information or non-existent references by language models, which can compromise the reliability of scientific and academic texts; authorship and plagiarism — the difficulty in defining authorship for texts generated or edited by AI, involving risks of plagiarism and a lack of transparency; cultural biases and inequalities — the tendency for models to reflect the values and styles of wealthy, industrialized Western nations, limiting cultural and linguistic diversity; interpretation of contextual nuances — AI's difficulty in understanding idioms and complex contexts, leading to misinterpretations; and identifying text origin — the growing difficulty in distinguishing human-written text from AI-generated text, challenging detection methods and raising questions about accountability.

These challenges necessitate the formulation of new policies, regulations, and evaluation criteria to ensure academic integrity and the quality of written output.

Practical Applications Across Various Sectors

AI in writing has broad and transformative applications: in education, assistance with writing, personalized learning, automated feedback, and integration into interactive lesson plans — though it raises concerns regarding originality and critical thinking; in journalism, generation of financial reports and news articles, creation of synthetic images and videos, and automation of repetitive tasks; in industry, accelerated code writing, synthetic voice generation, and the automation of creative processes; and in scientific research, assistance with drafting and revising articles, translation, and generating data visualizations.

These applications demonstrate AI's potential to transform traditional practices, while also highlighting the need for regulation and ethical use to prevent fraud and ensure quality.

Perspectives from Recent Books on AI and Writing

Recent books offer in-depth, multidisciplinary analyses of AI and its impact on writing and creativity: Demystifying Artificial Intelligence (Dora Kaufman) covers definitions, ethical impacts, the labor market, and AI applications in fields such as healthcare and climate; The Creativity Code (Marcus du Sautoy) explores AI's capabilities in creative fields, questioning whether machines can be truly creative; Artificial Intelligence: A Modern Approach (Stuart Russell and Peter Norvig) is a comprehensive AI textbook; Nexus (Yuval Noah Harari) analyzes the role of information and AI in shaping societies; Co-Intelligence (Ethan Mollick) is a practical guide to human-AI collaboration; and The Worlds I See (Fei-Fei Li) details the growth of AI and concerns regarding social inequalities. These books provide a critical and reflective overview of AI, its interaction with writing and creativity, and the ethical and social challenges accompanying its advancement.

Conclusion

The evolution of AI writing models over the past five years represents a technological and methodological leap that has transformed text production across multiple domains. The rise of large language models like ChatGPT and its counterparts has enabled sophisticated writing automation, offering significant benefits in productivity, translation, and linguistic adaptation. However, this evolution brings ethical and technical challenges that demand urgent attention, including defining authorship, combating plagiarism, and mitigating bias.

AI Tools for Writing and Content Creation

ChatGPT (OpenAI, v5.0) — coherent text generation, editing, narrative suggestions; used for writing, brainstorming, and dialogue; limited by a daily cap and occasionally generic responses.

Google Bard Pro (Google) — Google Workspace integration, direct generation in Docs; used for brainstorming, paraphrasing, and SEO; depends on a Google account.

Grok 3 (xAI) — open-source, multilingual support, creative generation; used for science fiction and mystery; has a complex initial setup.

Jasper AI (Jasper, 2026) — deep learning algorithms, trend analysis; used for marketing, SEO, campaigns; high cost, review required.

Copy.ai Elite — persuasive text generation, pre-built templates; used for copywriting and ads; cost and questionable originality.

Writesonic Advanced — Portuguese language support, smart editing, text refinement; used for social media content and websites; cost, review required.

Jenova — specialized agents, persistent memory, integrations; used for blogs, fiction, emails, research; cost, review required.

Sonix — audio and video transcription, real-time collaboration; used for transcription and brainstorming; cost, review required.