Talking to no-code AI tools: my 4-step cheat sheet
Before diving into a no-code AI tool, consult its “cousins” — the LLMs that know exactly how it likes to be spoken to.
6 min read

Just like in any new relationship, before committing to someone we want to be with, we typically consult friends who know them well. The same approach applies to no-code AI tools. Before diving in, we can consult their “cousins” — LLM providers like ChatGPT or Claude — who can tell us how these tools prefer to consume content, which formats they respond to best, and where to place emphasis for optimal results.
Step 1: Spill. Everything.
Begin by dropping all your thoughts, problem statements, requirements, and gaps into the conversation. Provide as much context as possible about your project — imagine you're starting a new meeting with teammates who need a thorough briefing on the project mission.
In most cases, this information already exists in files, documents, or Slack channels. The key challenge isn't writing new information but rather organizing and articulating what you already know.
- Don't hesitate to raise open questions or unresolved issues during this phase — it is the perfect opportunity to engage the AI in deeper thinking as a complementary task.
- Adding images and visual context at this stage is highly beneficial for the quality of the output.
Step 2: Actually read the response
Take time to thoroughly read the AI's response — don't just scan it. While the output may appear informative and well-structured, subtle nuances can sometimes be missed on a quick review. LLMs are powerful but occasionally take ownership in directions that slightly alter what you originally envisioned. After reading, adapt, modify, and refine as needed. Leverage canvas capabilities for more efficient editing without resending your entire query.
Step 3: Copy-paste magic
Once you've had sufficient back-and-forth with the AI chat and finalized a prompt that effectively demonstrates your challenge and covers all relevant cases, copy and paste it into your no-code AI tool.
This is where the magic happens: instead of multiple iterations using simple instructions that might eventually work, you now have a single, specific prompt that functions according to your vision. It's like a game where you're the visitor learning the rules to achieve better, more satisfying results.
Step 4: Tweak, tweak, repeat
Continue modifying your approach until you achieve increasingly better results. While you may get excellent results on your first attempt, nothing is perfect. Using tools in real-world scenarios often sparks new ideas and approaches — concepts that might need slight redesign to better align with your goals.
Real talk: how this actually works for me
Just recently, I used this exact approach and got my four design alternatives as planned. But during testing, we hit a snag: the button in design #2 was being stubborn.
Instead of banging my head against the no-code tool, I went back to Claude with a simple: “The button in design alternative 2 doesn't trigger the expected action when clicked. The animation plays but the form doesn't submit.”
Please modify design alternative 2 to ensure the submit button properly triggers the form submission event after the animation completes. Add a 200ms delay between animation end and form submission to prevent timing conflicts.
I copied this into my tool and it was fixed on the first try. Later, while playing with alternative #4, I noticed some UX wonkiness — inconsistent hover states and a navigation path that wasn't intuitive for first-time users. Claude came through with a precise instruction to implement consistent hover effects using the primary brand color at 20% opacity and restructure the navigation into a linear progression with clear visual indicators.
This constant ping-pong between LLM consulting and no-code implementation helped me refine all four alternatives without wanting to throw my laptop out the window.
Bonus: reuse isn’t a bad word
When you find yourself using LLMs for repetitive tasks as a vibe coder, here's a pro tip: reuse your initial prompt across tasks rather than repeatedly re-explaining the same request in new chats.
For example, in my day-to-day work I contribute to a feature that enables users to generate React components using AI. During regular testing and QA, I needed to stretch the product's technical limits. To do this efficiently:
- Gather inspiration: explore the web for cool, interactive components built by others, capturing screenshots and video recordings of the complete experience.
- Leverage media capabilities: upload these visuals to a model that excels at handling media files.
- Systematize the prompt: instead of repetitive back-and-forth, create one reusable custom assistant with a dedicated system instruction.
Analyze the provided video or image inspiration. Identify structure, layout, components, spacing, typography, colors, motion, and interactions. Infer missing details logically and keep assumptions minimal. Translate insights into a single, smooth, unambiguous prompt that describes how the element looks and behaves. Avoid technical jargon and frameworks. Be concise, precise, and copy-paste ready.
If I had the time and bandwidth, I could write prompts like Claude does. It isn't difficult to notice how LLMs prefer to receive instructions and adapt to that phrasing. However, as a non-native speaker focused on using AI to work faster, I prefer straightforward communication over fine-tuning every single word. That extra effort is better directed toward solving the actual problem at hand. I'm a designer and problem-solver, not a professional prompt engineer.
That's the sweet spot where AI seamlessly integrates into our workflow and helps us become 10X more effective.