Short answer: open aistudio.google.com, click into the system instruction or prompt field, press ⌃+⌥+R (Mac) or Ctrl+Alt+R (Windows/Linux), speak as long as you need, press again. AICHE inserts cleaned-up text in 2-3 seconds. Run the prompt.
Useful Gemini configurations are long. A real review agent needs role, expertise areas, evaluation criteria, response format, severity rubric, tone guidance, and "what never to approve" constraints. A documentation generator needs input types, output structure, style rules, and edge-case handling. Typing it all out is a 15-20 minute job, which is why most prompts end up at three sentences and underperform.
Voice closes the gap. The same comprehensive instruction takes 3-5 minutes to speak.
How It Works
- Open Google AI Studio.
- Create a new prompt or open an existing one.
- Click into the system instruction field, the prompt body, or a few-shot example.
- Press ⌃+⌥+R (Mac) or Ctrl+Alt+R (Windows/Linux).
- Speak. No length cap.
- Press the hotkey again. AICHE transcribes, applies AI cleanup, inserts.
- Run the prompt against test inputs, refine.
Where Voice Pays Off in AI Studio
System Instructions That Cover Real Cases
A code-review agent that actually works specifies what to look for, how to score severity, what tone to use, and what never to approve. Speaking it through, you cover each section: "You're an expert Python/Django reviewer. Focus on SQL injection, XSS, CSRF, N+1 queries, missing tests on business-critical paths. Severity scale: critical / high / medium / low with justification. Cite line numbers. Provide a fixed-code example, not just a description. Constructive tone, acknowledge good practices. Never approve code with SQL injection or missing auth on API endpoints."
Five short sentences spoken, comprehensive instruction in the field.
Few-Shot Examples Without the Typing Tax
Quality few-shot examples are what makes Gemini consistent. They're also the part most people skimp on because they take forever to type. Speaking them is fast: dictate the input, dictate the ideal output, repeat for two or three examples. Voice + AI cleanup handles formatting; you focus on the example content.
Documentation-Generator Prompts
Specs for a doc generator need style guide, section structure, tone, and "verify by tracing" instructions. Dictate: "Generate Markdown docs in Google's developer-docs style. Sections: overview, parameters table with type and required/optional, returns, raises, two or three usage examples, implementation notes with time complexity. Active voice, present tense. Concrete examples not foo/bar. Trace through the code to verify accuracy."
Data-Analysis Prompts
Speak the analysis spec the way you'd brief an analyst: data structure, what to compute, evaluation metrics, output format, visualization suggestions, what to flag as anomaly. The result reads like a real analyst brief, and Gemini follows it more consistently.
Code-Generation Prompts
Specifying the dependencies, error responses, security requirements, and testing surface up front means Gemini doesn't have to guess. Voice makes the up-front spec affordable instead of being the thing you skip to "save time".
What You Get
- Unlimited voice notes with AI cleanup - filler words removed, punctuation and paragraph breaks added.
- Software Development profile (Pro) - recognition tuned for code, APIs, library names, framework jargon.
- Custom vocabulary - drop in product names, internal libs, model names. Spelled correctly.
- System-wide dictation - same hotkey works in AI Studio, ChatGPT, Claude, your IDE, anywhere.
- Multilingual voice input - speak in any supported language; auto-translate to English if your prompts are in English.
- Zero-retention audio - audio purged immediately after processing, within 1 second.
Plans start at $3.99/mo (annual) with a 7-day free trial, no credit card. See pricing.
Common Questions
Q: Does this work in Vertex AI Studio (the GCP-side tool) too?
A: Yes. The hotkey works in any text field on Vertex AI Studio's UI as well.
Q: Can I dictate JSON schemas or function-calling specs directly?
A: Mixed approach: dictate the natural-language description of each tool/parameter, then add JSON syntax. Voice for prose, keystrokes for punctuation-heavy structure.
Q: My prompts use Markdown formatting. Will AICHE keep it?
A: Plain Markdown (headers, lists, code fences) survives. Speak "hash hash" or "bullet" if you want literal Markdown markers in the output.
Q: How do I dictate code blocks inside a prompt?
A: Speak "code block" and AICHE inserts triple-backticks, or paste the code afterward. Voice is best for the prose around the code, not the code itself.
Q: Will the audio be sent to Google?
A: No. AICHE handles transcription on its own infrastructure (audio purged immediately after processing, within 1 second). Only the text prompt is submitted to Gemini. AICHE never sends audio to Google.
Result: detailed Gemini configurations in 4-5 minutes instead of 18-20. Few-shot examples actually included instead of skipped. Prompt quality goes up because the spec is finally thorough.
Try it now: open Google AI Studio, create a new prompt, click into the system instruction, press your hotkey, and dictate the agent's role, expertise, response format, tone, and constraints in one pass. Run it against a test input. Compare to your usual three-sentence version.