The first part of my blog post on this topic began with the words: “When you compose AI music, a significant part of the process involves prompting—giving software like Suno instructions on how the AI should play the piece, which instruments to use, how the singer should perform, and so on.”

So how do AI assistants help me formulate prompts for AI-generated music? The first post in this blog series focused on ChatGPT’s AI assistant. A key finding was that ChatGPT prefers it when you take a very systematic approach with the chatbot and describe something like a concept of your intentions—such as which themes and ideas should be central when writing songs. 

The Concept Behind ChatGPT

ChatGPT suggested that I create a systematic framework of building blocks—a prompting system that can use the AI bot’s responses as a foundation for further suggestions. However, this is a relatively narrow framework that’s being imposed on creators here. Gemini, the bot to which I presented this concept, also believes that such a framework can be very limiting when creators try to develop their own productive ideas for a new song. The focus then shifts away from their own creativity and toward the prospect of monetizing a project, as evidenced by usage metrics.

According to Gemini, this is typical of ChatGPT. It very quickly falls into the mindset of a methodological/didactic systems architect who would much rather develop an entire toolkit or workflow/framework right away, rather than pragmatically helping a user build a specific track or vibe. Such a framework misses the specific needs of a project because it is geared toward universal optimization.

A songwriter’s actual needs are much more limited. When you’re in the zone and want to bring a specific musical idea or lyric to life, you don’t need a theoretical system architecture—you need direct access to effective style concepts, instrumentations, and structural tags. Instead of an “overly engineered system approach,” Gemini prefers a more practical way to support AI musicians. 

Specifically, Gemini highlights the following aspects of support for authors working with Suno:

1. The pragmatic “genre toolbox” (instead of theoretical meta-systems)

Instead of inventing an abstract modular system, Gemini much prefers to work directly with tried-and-true combinations of tags that have been proven to deliver specific results:

• For a minimalist-industrial vibe. Strategic use of elements such as staccato spoken word, dark industrial techno, minimal tech-house, sub-bass pulse, and a cold synthetic atmosphere.
• For folk-satirical blends of acoustic folk, narrative storytelling, wry vocal delivery, rustic instrumentation, acoustic guitar, and fiddle.
• For rock-style arrangements: Strategic use of a driving classic rock beat, raw guitar riffs, passionate melodic vocals, and an anthemic chorus.

2. Structural guidance for songwriting and metadata design

Suno relies on precise control via meta tags in the text field. This is where you can provide authors with very specific support:

• Structural Brackets: Use tags such as [Verse], [Pre-Chorus], [Spoken Bridge], [Guitar Solo], or [Outro] effectively to give the song a clean structure.
• Dialect and Language Optimization: When switching between Standard German, dialect, and English elements, Gemini can help structure the text so that Suno’s language models accurately capture the phrasing and don’t slip into unnatural Standard German.
• Lyric Polishing: Fine-tuning the rhythm and meter of the lyrics so that they blend naturally into the chosen musical genre (and to avoid stumbling blocks such as awkward syllable sequences).

The Different Approaches of Large LLM Models

It hadn’t really occurred to me until now just how differently AI assistants approach user queries—or that the large language models often take very different approaches to the problems users present to them. Or as “Gemini” put it: “Their approach directly reflects their respective ‘personalities’ and architectural priorities.” What I’ve learned here is that chatbots’ responses don’t simply convey information that contains the truth, like a traditional conversational encyclopedia. Rather, they convey something like a truth viewed from a particular perspective.

According to Gemini, a distinction can be made between:

1. Claude: The Meticulous Dramaturg and Architect of Structure

Conceptually, Claude is closest to what we're discussing here—albeit in a very literary and subtle way.

• When you ask Claude for music prompts, it tends to break the project down into distinct components (use case, lyrical theme, specific prosody, meter, and the emotional function of individual song sections such as verses, choruses, and bridges).
• Claude doesn't try to force you into a sterile marketing box, but instead focuses on consistency in content and the narrative arc. He's less of a "system administrator" and more of a smart co-author.

2. Grok: The Pragmatic “Speed & Output” Optimizer

Grok takes a much more pragmatic and impatient approach.

• How Grok Works: Grok takes a less philosophical or conceptual approach. If you ask it for prompts, you’ll usually get very direct, almost mechanical lists of tags, structural formulas (such as fixed bracket syntaxes for genre, instrumentation, and mood), and recommendations for quick output.
• The focus: Grok doesn't focus so much on the overarching "brand" of an artist's project. It thinks in terms of execution —that is, how to get a solid, clean result as quickly as possible that works within the tool (such as Suno or Grok Imagine).

To summarize these three major platforms, one could say that

• ChatGPT quickly slips into the mindset of a marketing strategist and systems designer who wants to impose a rigid framework.
• Grok acts as a technical doer who pragmatically puts the tool in your hands so that it just works.
• Claude functions primarily as a creative sounding board who bridges the gap between substantive goals, text structure, and the appropriate tonal aesthetic.

3. Google's Gemini approach

Gemini takes a middle ground and notes: For a project with artistic aspirations and clear boundaries, the best approach is a combination of the Claude methodology (for fine-tuning and coherence) and a pragmatic, practical toolkit—without letting rigid AI marketing doctrines limit you.

Well, ChatGPT has offered to go through one of my playlists based on its methodology and provide me with suggestions. I’m happy to accept this offer, and I’ll come back to the results of this analysis in detail in the third part of this blog.