A Gemini prompt generator that follows Google’s prompting guides
This Gemini prompt generator builds prompts the way Google’s own guides describe: a persona, a task, the context and the format you want, plus few-shot examples — which Google’s Gemini API guide recommends in every prompt. Start from a template or a blank form and the prompt updates as you type.
Four views cover the places you use Gemini. Plain text is a natural-language prompt for the Gemini app or Gemini in Google Workspace. Markdown and XML follow the two structured layouts in Google’s Gemini 3 guidance, rules first and task last. JSON is a request body for the Gemini API’s Interactions endpoint, with a placeholder part for each image, PDF, audio or video file you plan to attach.
How to build a Gemini prompt
- Pick a template or start blank, then fill in the variables.
- Write the Persona, Task, Context and audience, and list any Files you will attach.
- Set format, tone, length and language; tick Tell Gemini today’s date or Answer only from what I provide if useful.
- Add a few examples and read the checklist.
- Choose Plain text, Markdown, JSON or XML tags, then copy, download, share or save.
Gemini Prompt Generator features
Persona, task, context, format
The plain-text view follows the four-part pattern of Google’s Workspace prompting guide, in full sentences.
Few-shot first
No examples is a warning here, not just a tip.
Context before the request
Beyond 1,500 characters of context, the request moves to the end.
Markdown or XML structure
Google’s two layouts: # Identity / # Constraints / # Output format, or <role> … <task> tags.
Files for multimodal prompts
Listed images, PDFs, audio and video are named in the prompt and become typed parts in the JSON.
Date and grounding options
Two one-click lines that Google’s system-instruction examples suggest.
What Google’s guides recommend
- Clear, specific instructions, with constraints and the response format spelled out.
- Always include few-shot examples. Google writes that prompts without them “are likely to be less effective”. Show a pattern to follow, format every example the same way and experiment with how many to include.
- Consistent delimiters. The Gemini 3 section shows XML-style tags and Markdown headings; pick one.
- Role, rules and format first, in the system instruction or at the very beginning.
- Large context first, question last. Supply documents or code first and put the specific request at the very end.
- Tell it the date when time matters; Google’s example system instructions state the current year.
- Natural sentences in the app. The Workspace guide frames prompts as persona, task, context and format.
Sources, checked 14 October 2026: Google AI for Developers, “Prompt design strategies” (updated 13 October 2026), “Files” and “Image understanding” (updated 23 September 2026); Google Workspace, “Writing effective prompts”. Google also notes that its newest models ignore custom temperature, top-p and top-k values, so improve the prompt rather than the sampling settings.
Prompting with images, PDFs, audio and video
Gemini treats text, images, audio and video as equal inputs. Google’s file tips: be specific, add a few examples, break complex tasks into steps and name the output format. If answers stay generic, ask Gemini to describe the image first or point it at the part that matters; for invented detail, ask for shorter descriptions.
Google’s pages currently disagree on order: the Files guide suggests a single image before the text, while the newer image-understanding page puts the text first in the API’s input list. The JSON view follows the newer page — if results are weak, try the other order. To pull text out of a picture first, use image to text.
The four views
| View | Use it for | Layout |
|---|---|---|
| Plain text | Gemini app, Gmail, Docs | Persona and task in sentences, then context, steps, rules, examples, format |
| Markdown | AI Studio, system instructions | # Identity, # Constraints, # Output format, # Context, # Examples, # Task |
| XML tags | The same, with tags | <role>, <constraints>, <output_format>, <context>, <task> |
| JSON | Gemini API (Interactions) | system_instruction and input — add your model id |
Worked example
The “Product description” template with a photo, one example and the variables filled gives this plain-text prompt:
You are an e-commerce copywriter. Write a product description for the steel water bottle in the attached photo. The audience is commuters. I've attached bottle-photo.jpg (image). Refer to what is in it when you answer.
Context:
Key features: keeps drinks cold for 24 hours, leak-proof lid
Price: $29
Rules:
- Lead with the main benefit, not the feature list
- Use only the features given; do not invent specifications
- End with one sentence that invites the reader to buy
Examples:
Input: Bamboo lunch box, 1 litre, dishwasher-safe, $18
Output: Pack a proper lunch without the plastic. This 1-litre bamboo box goes straight into the dishwasher, so it is as easy to clean as it is to carry. Order yours for $18.
Write in flowing prose paragraphs. Use a persuasive tone. Length: about 120 words.Gemini Prompt Generator FAQ
What is the persona, task, context, format pattern?
It is how Google’s Workspace guide breaks down a prompt: who Gemini should be, what it should do, the background it needs and how the answer should look. This Gemini prompt generator writes those four parts as natural sentences.
Should I use XML tags or Markdown with Gemini?
Either. Google’s guidance shows both and asks only that you stay consistent. Copy the Markdown or XML tags view, whichever suits your code or notes.
Why did my context move before the request?
Google advises supplying large context first and the instructions at the very end. Past 1,500 characters, the plain-text view does that; structured views always put the task last.
Can I attach images or PDFs?
Yes — attach them in Gemini as usual and list their names here, so the prompt can refer to them. The page never reads or uploads your files.
Does it work for prompts written for Bard?
Yes. Google renamed Bard to Gemini in February 2024, so the same advice applies.
Is this tool made by Google?
No. It is an independent page that follows Google’s published guidance and does not connect to Gemini.