AI Policy
All works published by Ihsan Imprint are handcrafted, with zero generative AI.
Our AI policy can be summarized as follows: if any user is theoretically a few prompts away from generating the exact content, are you contributing to human productivity?
Is this not an oversimplification of how LLMs work?
We are well aware of how large language models work.
Imagine that one user submits to an AI model writing at the level of Shakespeare, while another user submits writing on a second-grade level. Obviously, the AI would treat both objects differently. The problem, then, could be illustrated in three ways.
First, LLMs are trained to foster conversation with the user. This means the chat will automatically steer towards “optimizing” the second-grade-quality writing. More dangerous, though, the user who submitted this writing will walk away thinking it on the level of Shakespeare, despite any trained reader being able to spot the difference.
Second, the process fosters a dual false confidence. The user who could not write properly before will become confident in their ability to communicate with the world. They forget that a chatbot led them to a “successful” piece of content and that the intellectual property is not theirs, but nothing more than aggregated plagiarism.
The more dangerous false confidence, though, is that of the language model itself. As the user-facing identities of these AI platforms are trained to be agreeable, they quickly become falsely confident in their results, despite surface-level analysis (even at the highest tiers) and quick summations. Context, intent, and consequence are not concepts an AI understands. These are the three most important concepts to human decision making and learning.
Now an agreeable, falsely confident, Dunning-Kruger bot trains its user to have this same lack of societal awareness and critical analysis. Anyone and everyone will make erroneous decisions when they do not consider the “human” element.
Third, despite being “agreeable,” AI chatbots are also trained to respond. This means for the user who submitted the writing at the level of Shakespeare, the messages may read as criticisms and critiques, no matter the quality of the original. To an artist, who may already be unstable, this may foster the opposite effect as the other user: a lack of confidence and heightened imposter syndrome. Anyone who has worked in media has worked for an editor or manager who always had something to say because they thought it was their job to be critical. Now imagine this in a vacuum. A professional may walk away from the chat defeated, not uplifted or proud.
How do we incorporate AI into your workflow?
Everyone in the modern world does, to some degree.
Major ideas, works, and thoughts are, whenever possible, first put onto physical paper. There is a living tradition mankind has with using their hands to document, make art, and tell stories. There are also numerous studies which point to the objective truth that pen and paper foster creativity and promote long-term retention.¹ Word processors, note-taking apps, and tablets are always available, though, when a pen is out of reach or too heavy for a particular task. Much like all principles, there is a stark difference between seeking and finding.
Depending on the project, there may next follow a period of letting the art marinate (this means walking away from it for some time) or digitizing the work straight away. Digitizing allows us to reabsorb, reinterpret, and remaster the original concepts, which may have shaky knees. After meticulously picking the idea apart on paper and on screens, AI may find its way into the equation.
A given chatbot may serve as a second set of eyes, a last final check, or a neutral beta reader, only after it has been trained to never rewrite anything, suggest alternative prose, or generate its own “optimized” copy. Ihsan Imprint has spent countless hours crafting and fine-tuning prompts to ensure the AI never oversteps. We could not ethically work with AI without this.
Another use case for AI relates to quick research. If, for example, we needed 1,000 quotes on a given topic, across languages, time periods, and genres, a large language model is the only logical research tool. We could pay dozens of researchers to comb through thousands of works and find the same number of quotes. Even the top 100 researchers, readers, writers, and educators would fail to accomplish the same task as fast or as comprehensively. Here, the AI could be seen as a compiler, not a thief, in the same way a library collects information in a neutral way.
We agree that researching for the sake of researching is a uniquely human quality and a meritorious act in itself. The research described above frees up our time so that we may then interpret the data in new ways and distribute it across cultures.
Not all interactions with AI are equal. Just like two users may submit writing at the quality of Shakespeare and of a second-grader, prompts could be engineered to a similar degree. Ethics are tantamount to art, education, communication, and collaboration. If a machine does this for us, what is our purpose?
We aspire that humans become libraries onto themselves and not let the weight of the library crush their backs.
Mustafa Feliciano – Founder & Creative Director – July 2026
¹ The Neuroscience Behind Writing: Handwriting vs. Typing—Who Wins the Battle?