Google Gemini prompting techniques explained (with examples)

Knowing the names of the prompting techniques is what separates a quick, reliable answer from a frustrating round of back-and-forth. Each technique solves a different problem: some give the model a role, some show it a pattern, some make it reason out loud, and some break a big job into stages. This guide walks through the main Google Gemini prompting techniques one at a time, with a short definition and a worked example for each, then shows how they fit the PTCF framework we cover in our guide to writing Gemini prompts. If you are studying for a Gemini or AI course, the same techniques and the scenarios below are exactly what most quiz questions test.

Zero-shot and few-shot prompting

Zero-shot prompting means giving Gemini an instruction with no worked examples and asking it to respond directly. It is the right choice for clear, common tasks the model already understands, such as summarising a paragraph or translating a sentence. A zero-shot prompt might read: classify this customer message as a complaint, a question or praise, then reply with one word.

Few-shot prompting adds a small number of example input and output pairs inside the prompt so the model copies the pattern instead of inferring it. This is the technique to reach for when the format, tone or style has to be consistent. Google's developer guidance recommends including a few examples to steer the model, and notes that the examples teach structure more reliably than instructions alone (Google AI for Developers). A worked few-shot example:

Classify each review as Positive, Neutral or Negative.

Review: "Coffee arrived fast and tasted great." -> Positive
Review: "It was fine, nothing special." -> Neutral
Review: "Cold, late, and the wrong order." -> Negative

Review: "Lovely beans but the bag was torn." ->

The three examples lock the labels and the output shape, so Gemini returns a clean single word for the fourth review rather than a paragraph of its own choosing. As a rule, start zero-shot, and switch to few-shot the moment the output drifts from the format you need.

Persona and role prompting

Persona prompting, also called role prompting, tells Gemini who to be before it answers. Setting a role shifts the model's vocabulary, assumptions and depth, because an experienced tax accountant and a primary school teacher explain the same idea very differently. This is the P in PTCF, and a good persona names a field and a seniority level rather than a bare job title.

A worked example: you are a senior conversion copywriter who writes for Australian small business owners. Explain in plain language why a slow website loses sales, in three short paragraphs. The role steers tone and reading level in one line, so you do not have to spell out every stylistic choice. Google's Workspace prompting guidance highlights persona as one of the most effective single additions you can make to a prompt (Google Workspace).

Providing context and constraints

Context is the background that makes the output yours rather than generic: your audience, your goal, the reference material and the boundaries the model must respect. Constraints are the limits inside that context, such as a word count, a banned topic, a region or a required level of formality. This is the C in PTCF, and Gemini's large context window means you can paste a brief, a transcript or a whole document and tell the model to work from that source.

A worked example: using only the product notes pasted below, write a 90 word description for an Australian audience. Do not invent features, do not mention price, and keep it suitable for a year-eight reading level. The more relevant context you supply and the tighter the constraints, the less the model fills gaps with assumptions. When an answer comes back vague, a missing constraint is usually the cause.

Specifying the output format

Format prompting controls the shape of the response: a table, a numbered list, a short email, a JSON object, a specific tone or an exact word count. This is the F in PTCF, and it is the cheapest way to make an output usable straight away. If you do not state a format, Gemini picks one for you, and it rarely picks the one your workflow needed.

A worked example: list five SEO title ideas for a Parramatta plumber. Return them as a markdown table with two columns, Title and Character count, and keep every title under sixty characters. Stating the structure up front saves a round of reformatting and lets you drop the result directly into a spreadsheet or a brief. For structured data work, asking explicitly for valid JSON with named keys is far more reliable than hoping the model guesses your schema.

Chain-of-thought and step-by-step reasoning

Chain-of-thought prompting asks Gemini to work through a problem step by step before giving the final answer. Showing the reasoning improves accuracy on multi-step tasks such as maths, logic puzzles, planning and analysis, where a model that jumps straight to an answer often slips. The trigger can be as simple as adding let us think step by step, or asking the model to show its working.

A worked example: a cafe sells 240 coffees a day at $4.50 and wants to lift daily revenue by 15 percent through price alone. Work through the new price step by step, showing each calculation, then state the final price rounded to the nearest five cents. Asking for the steps both improves the result and lets you check the logic rather than trusting a bare number. Google Prompting Essentials, Google's own short course on prompting, teaches this break-it-into-steps habit as a core skill (Google Prompting Essentials).

Prompt chaining and multi-turn prompting

Prompt chaining means splitting a larger job into stages and feeding the output of one prompt into the next, rather than overloading a single request. Because Gemini keeps earlier turns in context, you can do this naturally across a conversation, which is why the technique is also called multi-turn or conversational prompting.

Here is the exact scenario people search for, stated plainly. When you first ask Gemini to list items, and then in a follow-up prompt in the same chat you ask Gemini to use that list to draft something, the technique you are using is prompt chaining (also called multi-turn or conversational prompting). You are using the result of the first prompt as the input to the second. A worked example over two turns:

Turn 1: List six common questions new customers ask
        about a coffee subscription service.

Turn 2: Using the list above, write a short FAQ section.
        Answer each question in two sentences, friendly tone.

Each turn does one clear job, the second builds on the first, and you can correct the list before it becomes the FAQ. Chaining beats cramming because a single prompt with ten nested instructions buries the priority and gives you nowhere to intervene if a step goes wrong.

Iterative refinement

Iterative refinement is the habit of treating the first answer as a draft, then improving it one change at a time. Read what Gemini returns, adjust a single element, the tone, the length, one constraint, and run it again so you can see exactly which change moved the output. Changing five things at once tells you nothing about what worked.

A worked example of a refining follow-up: good start, now cut it to 80 words, make the opening line a question, and remove the word innovative. Refinement pairs naturally with chaining, because each follow-up is just another turn in the conversation. Across these techniques, the same principle holds: name the persona, task, context and format, add examples or reasoning when the task is hard, and refine deliberately. If you also work in other models, our top prompts for ChatGPT show the same ideas applied elsewhere, and our guide to Gemini models helps you pick the right model for the job. For the wider toolkit, see our complete guide to Google AI tools.

Gemini prompting techniques at a glance

Use this as a quick reference: each technique, when to reach for it, and a short example you can adapt.

Working from a course that teaches PARTS rather than PTCF? We set out what PARTS stands for in Gemini prompts, including the two versions of the acronym in circulation.