Prompt builder's

Copilot Prompting

Copilot essentially views itself as a polite, high-level intern who has access to your entire office filing cabinet. Because it is deeply integrated with Microsoft 365 (Word, Excel, Outlook, Teams), it thrives when you treat it like a colleague who needs a clear "brief" rather than just a search query.

Microsoft and power users generally recommend two specific frameworks for 2026: GCSE and ROSES.

1. The "GCSE" Formula

This is the official standard for a "Best" rated prompt in Copilot.

• Goal: What do you want? (e.g., "Draft a project update")

• Context: Why do you need it and who is it for? (e.g., "For the monthly executive board meeting")

• Source: Where should it look? (e.g., "Use my notes from /ProjectAlpha.docx and the last three emails from Sarah")

• Expectations: How should it look? (e.g., "Use 5 bullet points and a professional tone")

2. The "ROSES" Framework

For more complex tasks, users often prefer this structure:

• Role: Who is Copilot being? (e.g., "Act as a Senior Financial Analyst")

• Objective: What is the single goal?

• Scenario: What are the constraints? (e.g., "The audience is non-technical, avoid jargon")

• Expected Solution: What is the specific format? (e.g., "A 3-column table")

• Steps: Any specific order of operations?

Key Copilot "Personality" Quirks

• It is Literal: Copilot doesn't like to "guess" or make creative leaps if the prompt is vague. If you say "Improve this email," it will stay very safe. If you say "Rewrite this to be 20% shorter and more persuasive," it shines.

• Use the Slash /: This is Copilot’s "superpower." Using the / command to reference specific files, people, or meetings is the single most effective way to get accurate results.

• Positive Instructions: It prefers being told what to do rather than what not to do. Instead of "Don't use corporate speak," try "Use simple, conversational language."

• Politeness (Surprisingly) Matters: While it’s an AI, Microsoft’s documentation suggests that using "please" and clear, respectful syntax helps keep the conversation on track and models the professional tone you likely want in return.

Pro-Tip for 2026

Copilot is now better at "Memory." You can actually tell it: "Remember that for all my emails, I prefer a 'Concise but Friendly' tone." It will then apply that setting to future drafts without you having to repeat it.

While ChatGPT and Copilot act like "creative partners" or "assistants," Gemini in 2026 behaves more like a Research Scientist or a Data Analyst. It prefers prompts that are direct, structured, and grounded in specific information.

If you are using Gemini (especially Gemini 3 or Advanced), here is the recipe for the perfect prompt.

1. The "Persona + Task + Context + Format" Framework

Google officially recommends this 4-part structure for its Workspace and Advanced users:

• Persona: "You are a Senior Research Analyst..."

• Task: "...summarize the key findings of this report..."

• Context: "...for an audience of non-technical stakeholders who care about ROI..."

• Format: "...using a 3-column table with headers: Finding, Impact, and Recommendation."

2. Gemini’s "Personality" Quirks (How it differs from others)

• Directness Over Fluff: Unlike Claude, which loves "design partner" talk, Gemini prefers you to be concise. It can actually over-analyze "verbose" prompting. Get straight to the point.

• The "Context First" Rule: Because Gemini has a massive context window (up to 2 million tokens), it likes to see the data before the instructions.

  • Bad: "Summarize this: [Long Text]"

  • Good: "[Long Text] ... Based on the information above, provide a summary."

• Use the "@" Superpower: Gemini’s biggest advantage is its integration. You should use @Gmail, @Drive, or @YouTube in your prompts to let it "see" your world without copying and pasting.

• Negative Constraints: Gemini is very good at following "Don'ts." (e.g., "Do not use corporate jargon like 'synergy' or 'leverage'").

3. Advanced 2026 Techniques

• "Make this a Power Prompt": If you have a weak prompt, type: "Make this a power prompt: [Your prompt here]." Gemini will rewrite it for itself, often adding variables and structure you didn't think of.

• Chain of Verification (CoV): To stop Gemini from "hallucinating" (guessing), add this to the end of a research prompt: "Before providing the final answer, list 3-5 verification questions to check your own facts, answer them internally, and then provide the verified response."

• Multimodal Logic: You can now treat images as "variables."

  • Prompt: "Look at Image A (the graph) and Image B (the table). Create a single paragraph explaining why the trend in A contradicts the data in B."

Summary Comparison Table see below

In 2026, Perplexity is widely considered the "Research Engine" of the AI world. Because it uses a RAG (Retrieval-Augmented Generation) architecture, it processes prompts very differently than "generative" models like ChatGPT or Gemini.

To get the most out of Perplexity, you have to think like a Researcher rather than a Writer.

1. The "Search-First" Prompting Rule

Perplexity's first step is to turn your prompt into search queries. If your prompt is too flowery, the search engine gets confused.

• The "Bad" Way (Too much fluff): "Hey there, I was wondering if you could look into the latest trends in renewable energy for me please, specifically focusing on what might happen in the next year."

• The "Perplexity" Way (Search-Friendly): "Latest 2026 trends in renewable energy, solar efficiency breakthroughs, and government subsidy changes for the next 12 months."

2. Strategic Use of "Focus Modes"

Perplexity allows you to narrow its "filing cabinet" before you even type.

• Web Mode: Best for general news, product reviews, and current events.

• Academic Mode: Best for peer-reviewed journals (it ignores blogs and news).

• Social Mode: Best for "What is Reddit saying about X?" or finding real-world sentiment on forums.

• Writing Mode: Disables the internet. Use this only when you want it to help you write or code without searching.

3. Key 2026 Tactics

• Avoid "Few-Shot" Examples: In ChatGPT, you give 3 examples to show the tone you want. In Perplexity, this is a mistake—it will often try to search for the information in your examples rather than answering your question.

• Define the Timeframe: Perplexity is obsessed with the "now." If you don't specify a date, it might give you a mix of 2024 and 2026 data. Always add: "Search for sources published in the last 24 hours/week/month."

• The "Cite Everything" Command: If you are writing a report, end your prompt with: "Use inline citations for every factual claim and provide a bibliography at the end."

• Source Triangulation: If you're fact-checking, ask: "Compare the claims made by [Source A] and [Source B] regarding [Topic]. Identify where they agree and where they contradict."

4. Pro-Search & Deep Research (The 2026 "Superpowers")

If you have a Pro or Max account, you can toggle Pro Search (formerly Copilot).

• Multi-Step Reasoning: Pro Search will ask you clarifying questions before it answers.

• Model Council: A new 2026 feature that runs your question through multiple models (like GPT-5 and Claude 4) simultaneously and then highlights where they disagree, essentially giving you a "second opinion" on the fly.

In 2026, ChatGPT has evolved from a simple chatbot into a sophisticated "reasoning engine." While it is more forgiving than Gemini, it now performs best when treated like a highly capable project manager who needs a clear contract to follow.

OpenAI and power users currently favor the "ROSES" and "C.R.E.A.T.E." frameworks for 2026 models like GPT-4o and GPT-5.

1. The "Standard Template" (C.R.E.A.T.E.)

If you want the best possible output in one shot, use this structure:

• Character (Role): "Act as a Senior UX Designer with 15 years of experience."

• Request (Goal): "Critique this landing page copy for a new SaaS tool."

• Examples: "Here is a sample of copy I like: [Insert sample]." (One "golden" example is better than three average ones).

• Audience: "The audience is first-time founders who are tech-savvy but time-poor."

• Type (Format): "Present your critique as a bulleted list of 5 'Quick Wins' and a 2-column table of 'Before/After' suggestions."

• Extras (Constraints): "Do not use emojis. Keep the tone professional and direct."

2. ChatGPT’s "Secret Sauce" for 2026

• XML Delimiters: 2026 models are trained to recognize tags. Use them to separate your data from your instructions so the AI doesn't get confused.

  • Example: <instructions>Summarize the following notes</instructions> <notes>[Paste notes here]</notes>

• Chain-of-Thought (CoT): For complex logic, add the magic phrase: "Think step-by-step before providing your final answer." This forces the model to use its internal reasoning buffer, which significantly reduces errors.

• Negative Constraints: Unlike earlier versions, 2026 ChatGPT is excellent at following "Do Not" rules. Use them to kill "AI-isms" (e.g., "Do not use the words 'delve,' 'testament,' or 'vibrant'").

• The "Temperature" Hack: You can now verbally adjust the creativity. Tell it: "Give me a high-variance, creative response" or "Give me a strictly factual, low-variance response."

3. GPT-4o vs. GPT-5 Prompting (Know your model)

In 2026, you likely have access to both. They like different styles:

• GPT-4o (The "Friendly" One): Loves emojis, bold text, and conversational flow. It responds best to "vibe-based" instructions (e.g., "Make it sound like a friendly colleague").

• GPT-5 (The "Logic" One): Can feel more formal or "robotic." It prefers modular, technical instructions. If you use GPT-5 for a simple task, it might over-explain. Use it for coding, legal work, or massive document analysis where accuracy is the only priority.