Many people believe the quality of AI-generated content hinges primarily on prompt engineering. Having used AI-assisted writing for some time, my experience shows that prompts have surprisingly limited influence—they only control surface-level aspects like format and tone.

What truly determines content quality are three other factors.

If we liken AI writing to cooking a dish, these three elements are: ingredients, culinary skill, and flavor profile.

Ingredients Are the Raw Materials: You Provide the Ingredients, AI Can Only Cook What You Give

Many users approach AI writing by simply throwing a topic at it and letting it generate freely. The result is often an article that’s technically correct but empty—filled with generic platitudes devoid of substance.

If you give nothing, AI cannot create anything.

For example, if you ask your assistant to write a speech but only say “write something about our company’s strategic direction” without any further context, the output will likely be a string of polished but meaningless statements.

But if you clarify: “Our core message this time is shifting focus entirely to domestic markets, because overseas customer acquisition costs have tripled; I’d like to open with a failed case study from last quarter in Southeast Asia,” the resulting content will be dramatically different in quality.

AI writing works the same way. You don’t need to provide a full draft, but you must at least supply the main thread and key points. What argument are you trying to make? Which examples will support it? What should readers remember after reading? Once you’ve clarified these, even just three or five sentences fed into AI will produce far better results than vague background information.

The value of input isn’t in volume—it lies in curation and judgment. Your decisions on what to include, what to omit, which example to use, and which to discard—these choices form the foundation of content quality. AI excels at filling in details, structuring arguments, and polishing language *after* you’ve defined the direction. But the direction must come from you.

For me, the most effortless tasks are translation and transcribing video interviews—because both the input materials and formats are fixed, so AI reliably produces solid output, as long as the source material is strong.

But for pieces requiring original viewpoints—like this one—I can’t just say “write an article on improving AI writing quality.” I must break down each critical point clearly, explicitly stating that the entire piece should weave through the themes of ingredients, culinary skill, and flavor. Only then does AI know how to proceed.

With ingredients ready, the next step is who’s doing the cooking.

Model Is Culinary Skill: Same Ingredients, Vastly Different Results

The differences between models in writing performance are far greater than most people realize.

Same ingredients, but a Michelin-starred chef produces something fundamentally different than a street vendor. You can’t take premium ingredients and hand them to an average cook expecting a masterpiece.

Currently, I believe Claude Opus 4.6 delivers the highest-quality writing. Its grasp of language rhythm, contextual understanding, and nuanced tone control are clearly superior. That said, models evolve rapidly—the rankings shift constantly. I once favored GPT-4.5, but GPT-5 series underperformed; Gemini 3.0 Pro’s style became too monotonous—reading multiple outputs felt like eating the same dish over and over.

Smaller models generally perform poorly. Larger parameter models tend to be better. For instance, while many praise Sonnet, it still falls short compared to Opus.

Naturally, these are subjective impressions.

Editing Is Tasting: If You Can’t Taste the Difference, You Can’t Serve Good Food

Initially, I considered writing “author’s skill” as the key factor. But increasingly, authors themselves are AI Agents. Whether reviewing is done by humans or agents, the principle remains: you must be able to discern what works and what doesn’t in a draft.

This is a hidden threshold often overlooked—and also the current bottleneck limiting my own writing quality.

After AI generates an article, you think “it’s okay” and publish it directly. But **“okay” is a dangerous judgment**. You might miss telltale signs: a typical AI hook at the beginning, a logical gap in the middle, or a forced rhetorical climax at the end. If you can’t spot these flaws, you can’t provide meaningful feedback for revision—so the final quality becomes a matter of luck.

When lucky, AI produces something decent on the first try. When unlucky, the published content is bland or even contains critical errors.

Conversely, if you possess strong editorial skills, you can instantly identify filler that needs cutting, mismatched metaphors needing replacement, or abrupt transitions requiring smoothing. Feed those insights back into AI, request a rewrite of the problematic section, and after several iterations, the draft can rise from a mediocre 60 to a solid 80.

This is like a skilled chef tasting each bite, detecting where it’s too salty or missing a key spice, then making precise adjustments. A chef without taste may take one bite, think “it’s fine,” and serve it blindly—relying entirely on luck.

This issue is even more pronounced when using AI Agents for writing. An Agent acts like a chef without taste: cooks a dish, declares it finished, and never questions whether the flavor is right. Even if instructed to self-review, its critique stays superficial—something like “this dish has great color, aroma, and taste”—empty praise lacking real insight. It lacks the internal sense to judge whether a sentence flows naturally or whether an example is persuasive.

Editorial ability stems from extensive reading and writing experience, and from an intuitive sense of what constitutes high-quality content. This capability cannot be replicated quickly by AI—precisely why I never automate content generation. I’d rather publish less, but invest time carefully reviewing and refining each piece.

Ingredients, Culinary Skill, Flavor—None Can Be Missing

Good ingredients but poor cooking ruin the potential. Excellent cooking but poor ingredients leave no room for mastery. Even with both ingredients and skill, flawed flavor leads to inconsistent results. One weak link spoils the outcome.

As for prompts? Prompts are seasonings—salt, pepper, spices. They enhance a dish when everything else is already in place. But you can’t rely solely on seasoning to transform a meal lacking ingredients, cooking skill, and proper tasting.

Author: Baoyu