“We need a system,” she told her team in March 2024 at their San Francisco office. Within two weeks, they had one, and their success rate jumped from 23% to 81%. Sarah Chen stared at ChatGPT’s fourth attempt to write product copy for Figma’s new collaboration features. Each version sounded like it came from a corporate buzzword generator.
\n\n
The problem is not the AI, but how we talk to it.
\n\n
The Three-Layer Prompt Structure That Changed Everything
\n\n
Most people jump straight to asking for what they want. It’s like hiring a designer and saying, “Make it good.” Figma’s marketing team discovered that effective prompts need three distinct layers: context, constraints, and criteria.
\n\n
These numbers aren’t made up—they’re taken straight from Datadog’s analytics. For a product launch email, they’ll say, “This goes to 50,000 developers who already use Figma. They receive our emails weekly. 68% open on mobile.” Context is everything. The team now starts every prompt with two or three sentences about who will read the content, where it will appear, and what action they want readers to take.
\n\n
The Figma team maintains a shared Notion document with a list of 47 constraints, including: “Never use ‘seamless’ or ‘game-changing’. Always write ‘plug-in’, not ‘plugin.’ Keep sentences under 25 words.” sentence The constraints define the boundaries. Word count matters, but so do brand voice rules, required terminology, and forbidden phrases.
\n\n
For blog posts, they require a Flesch-Kincaid reading score between 8 and 10. For social media copy, they require at least one concrete number or statistic. For product descriptions, they require at least one customer pain point and one specific feature that solves it. The team asks: “What makes this output actually good?”
\n\n
Why The First Draft Is Always Wrong (And What To Do About It)
\n\n
The Figma team learned that iterative refinement of AI outputs improves the quality of the results. Their process now includes mandatory revision rounds built into the project timeline.
\n\n
In April 2024, when launching FigJam’s AI features, their first prompt produced technically accurate copy that completely missed the emotional hook. The second attempt, with added context about user frustration with traditional whiteboard tools, nailed it. The first output establishes the direction. They never use it as-is. They treat it like a rough sketch, useful for identifying what’s missing.
\n\n
The old way, starting from a blank page, took at least 90 minutes. Now we budget forty minutes per piece: ten minutes for the initial prompt, fifteen for the AI generation and first review, and fifteen for the final prompts and final revisions.
\n\n
The revision strategy follows a pattern: first round: structural feedback (“Add more specific examples in paragraph 2”), second round: voice and tone adjustments (“Make it sound less corporate, more conversational”), third round: polish (“Vary the sentence length, remove any remaining jargon”), fourth round: rarely.
\n\n
The parallel is not accidental. Good writing and good code both improve through systematic iteration. This is how GitHub’s 100 million developers approach code review: multiple passes, each with a different focus.
\n\n
The Role Library That Saves 6 Hours Per Week
\n\n
They built a role library, a collection of pre-written prompts that they could mix and match like Lego blocks. The breakthrough came when Chen’s team realized that they were retyping the same blocks of text dozens of times.
\n\n
Each role defines a specific writing persona with detailed characteristics:\n
- \n
- Tech Journalist Role: “You write for The Verge. Your articles connect individual product launches to broader industry trends. You cite specific studies and data points. You interview real people and open articles with their stories.”
- Product Marketer Role: “You write product copy for B2B SaaS tools. Your audience includes engineering managers and startup founders. You lead with customer pain points, not features. Every claim includes proof – customer quotes, usage statistics, or third-party benchmarks.”
- Technical Educator Role: “You teach developers how to use new tools. You assume intermediate knowledge – no hand-holding, but no unexplained jargon. Every tutorial includes code samples, expected outputs, and common mistakes to avoid.”
\n
\n
\n
\n\n
The consistency of the content has improved significantly, so that different team members now produce content that sounds like it was written by the same writer. The library now contains 23 roles. Team members copy the relevant role block, add project-specific context, and paste it into their AI tool.
\n\n
Treating prompts as infrastructure, not ad-hoc requests, changed the way the team thought about AI tools. Prompts became versioned, reviewed, and improved systematically. They store these roles in HashiCorp Vault alongside API keys and other sensitive configuration data.
\n\n
The quality of the content has also improved, and it now consistently hits the target reading level and includes the specific, data-driven details that their technical audience demands. The time savings compound. Chen estimates that her seven-person team saves 42 hours a month, which they now use for strategy and campaign planning.
\n\n
What Actually Works Right Now
\n\n
The three lessons from Figma’s experience apply to any team writing AI prompts today.
\n\n
The team now includes brand names (“Mention Supabase as an example of backend-as-a-service”), specific numbers (“Include the statistic about Nvidia’s $3 trillion market cap in June 2024”), and exact formatting requirements (“Use H2 headings, not H3”). The more precise the input, the less editing is required afterward. First: specificity wins. Vague prompts produce vague outputs.
\n\n
/ Second: constraints liberate. Counter-intuitive, but true. When the team added their list of 47 constraints to every prompt, the outputs became more creative, not less.
\n\n
The difference is that AI iterates in seconds instead of hours. Third, treat AI like a junior colleague. You wouldn’t give a junior writer a vague assignment and expect perfection. You would give examples, explain the audience, define success criteria and review drafts.
\n\n
Clear, precise writing is more important than ever in a world running on increasingly complex systems. The CrowdStrike outage in July 2024 cost Fortune 500 companies $5.4 billion, according to Parametrix Insurance, a reminder that when technical communication fails, the consequences multiply quickly. The stakes keep rising.
\n\n
The role library grows. The constraint list evolves. But the core principle remains the same: AI is a tool, not a replacement. The human defines the target, and the AI helps to hit it faster. Chen’s team continues to refine their approach. They test new prompt patterns every week, document what works, and share the results across Figma.
\n\n
Sources and References
\n
n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n n
