This is a reference sheet, not a lecture. Bookmark it, and when a ChatGPT answer misses, scan the tables for the one modifier that will fix it. Every phrase below is copy-paste ready, and the starter prompts at the bottom combine them.
The rule underneath all of it: be specific about the output you want. On GPT-5.6 (July 2026), vague prompts still get vague answers. Give it a role, a job, the context, and a format, and it will use web browsing, Python, and your saved memory to deliver.
The universal prompt formula
Every good prompt is some mix of five parts: Role + Task + Context + Format + Constraints. You don't always need all five, but naming each one removes guesswork from the answer.
| Part | What it sets | Example |
|---|---|---|
| Role | Expertise and vocabulary | "You are a senior product marketer." |
| Task | The one job to do | "Write a launch announcement." |
| Context | The facts it needs | "Product: a budgeting app for freelancers." |
| Format | Structure of the output | "120 words, plus 3 subject-line options." |
| Constraints | Limits and rules | "No jargon. Friendly tone. US spelling." |
You are a senior product marketer. Write a launch announcement email.
Context: We're launching [PRODUCT], a [ONE-LINE DESCRIPTION] for [AUDIENCE]. The main benefit is [BENEFIT]. Launch date is [DATE].
Format: Under 130 words, one clear call to action, plus 3 subject-line options.
Constraints: Friendly but professional. No hype words. US spelling.For a fuller walkthrough of building prompts this way, see our guide on how to prompt ChatGPT for business writing.
Models & reasoning modes (GPT-5.6)
Since the GPT-5.6 launch (July 2026), the model picker matters as much as the prompt. The family splits into three tiers, and the harder tiers add reasoning modes you can turn on per task. Match the model to the job before you tune the wording.
| Model | Best for | Notes |
|---|---|---|
| Sol | Hardest reasoning, long reports, agentic coding, deep research | Flagship tier; slowest and priciest. Home of Ultra mode. |
| Terra | Everyday office work — writing, analysis, most prompts on this page | The balanced default. Free and Go users get Terra. |
| Luna | Quick rewrites, short answers, high-volume tasks | Fastest and cheapest; light on deep reasoning. |
| Reasoning mode | What it does | When to use it |
|---|---|---|
| Standard | Default speed and depth | Most prompts — email, summaries, drafts. |
| Max mode | Allocates extra compute for a single hard problem | Tricky analysis, thorny logic, careful edits. |
| Ultra mode | Coordinates four agents in parallel on a demanding, multi-step job | Big research-then-build workflows on the Sol tier. |
| Effort controls | Dial reasoning up or down per task | Available on Plus, Pro, Business and Enterprise — turn effort up when accuracy beats speed. |
Rule of thumb: stay on Terra for daily work, switch to Sol (with Max or Ultra) when a task is long, high-stakes, or spans several steps, and drop to Luna for throwaway rewrites. For multi-step jobs that run across your files and connected apps, hand the whole thing to the Work agent — the best ChatGPT prompts for work roundup shows how to brief it.
Context window: all three tiers hold 272,000 tokens — roughly a 500-page document — so you can paste an entire report, contract, or codebase into one prompt and ask questions across the whole thing. When you do, name the structure you want back (a table, a decision list, a summary per section) or the answer sprawls. There's a starter for exactly this below.
Talking to it: the GPT-Live voice model (July 2026) listens and speaks at the same time, so you can interrupt and steer mid-answer instead of waiting for it to finish. The same modifiers work out loud — say the role, the format, and "keep it to three points" and it obeys. Voice is best for brainstorming and quick rewrites; switch to text when you need a table or code you'll copy.
Role modifiers
A role sets who ChatGPT "is," which changes its vocabulary, priorities, and level of detail. Put it in the first line.
| Modifier | What it does | Example snippet |
|---|---|---|
| Act as a… | Assigns a job function fast | "Act as a hiring manager reviewing my resume." |
| You are a senior… | Raises expertise and rigor | "You are a senior copy editor." |
| Play the role of… | Sets up a persona to interact with | "Play the role of a skeptical customer." |
| Respond as if you were… | Borrows a known voice or style | "Respond as if you were a plain-English explainer." |
| You specialize in… | Narrows to a niche domain | "You specialize in B2B SaaS onboarding." |
| Assume the perspective of… | Reframes for an audience or critic | "Assume the perspective of my CFO." |
| You are an expert with 15 years in… | Signals depth and confidence | "…15 years in email deliverability." |
Roles pair well with any task. Our roundup of the best ChatGPT prompts for work leans on them heavily.
Format modifiers
Naming the exact output structure is the highest-leverage change you can make. Add one of these to almost any prompt.
| Modifier | Effect | Example |
|---|---|---|
| As a table | Comparable rows and columns | "Compare these 3 tools as a table." |
| As a bullet list | Scannable, no prose | "Give me the key points as bullets." |
| Step by step / numbered | Ordered, actionable steps | "Explain the setup as numbered steps." |
| As JSON | Machine-readable output | "Return the result as valid JSON with keys name, price." |
| In exactly N words | Enforces length | "Summarize this in exactly 50 words." |
| In markdown with headings | Structured document | "Format as markdown with H2 headings." |
| One paragraph, no lists | Tight prose block | "Answer in one paragraph, no bullets." |
Tone & style modifiers
Tone words steer voice without you rewriting anything. Add one after the task, or set a default in custom instructions.
| Tone word | When to use it |
|---|---|
| Professional | Client, exec, or formal work communication |
| Friendly / warm | Teammates, customers, onboarding messages |
| Concise / punchy | Slack, subject lines, ads, headlines |
| Confident | Sales copy, pitches, proposals |
| Neutral / objective | Reports, documentation, comparisons |
| Empathetic | Apologies, bad news, support replies |
| Plain / simple | Explaining to a non-expert audience |
| Persuasive | Calls to action, marketing, fundraising |
For applying tone to inboxes specifically, see our ChatGPT prompts for writing work emails.
Reasoning & quality boosters
These phrases make ChatGPT slow down, check itself, or show its options. Use them when the answer matters and a first draft won't do.
| Booster | What it does | Add this line |
|---|---|---|
| Think step by step | Reduces skipped logic on hard tasks | "Think step by step before answering." |
| Ask me questions first | Fills gaps before it guesses | "Ask me any clarifying questions before you start." |
| Cite sources | Grounds claims, enables checking | "Browse the web and cite your sources with links." |
| Self-critique, then improve | Catches weak spots in a draft | "Write a draft, critique it, then give a better version." |
| Give 3 options | Surfaces range instead of one answer | "Give me 3 distinct options, each with a one-line tradeoff." |
| Show your assumptions | Exposes what it inferred | "List the assumptions you made at the end." |
Quick shortcut modifiers
Short prefixes you can type at the start of any prompt to switch mode instantly. They aren't official commands — ChatGPT just reads them as plain instructions — but they're fast to type and consistent.
| Shortcut | Reads as | Use it for |
|---|---|---|
| ELI5: | "Explain like I'm 5" | Stripping jargon out of a hard concept |
| TL;DR: | "Give me the short version" | Compressing a long answer or pasted text |
| Steelman: | "Argue the strongest case for this" | Pressure-testing an idea before you commit |
| Devil's advocate: | "Argue against this" | Finding the holes in your own plan |
| Rewrite: | "Same meaning, better words" | Polishing a sentence or paragraph fast |
| Fact-check: | "Browse and verify each claim" | Grounding anything you'll publish |
Refinement / follow-up phrases
Once ChatGPT replies, you rarely need a new prompt. Drop one of these follow-ups and it edits its own last answer.
| Follow-up | What it changes |
|---|---|
| "Make it shorter" / "cut it by half" | Trims length, keeps the point |
| "Expand section 2" | Adds depth to one part only |
| "More formal" / "less formal" | Shifts tone up or down |
| "Simplify for a non-expert" | Removes jargon, adds plain terms |
| "Same, but friendlier" | Adjusts voice, keeps structure |
| "Give me 3 more like #2" | Iterates on a favorite option |
| "Now put it in a table" | Reformats the same content |
A note on GPT-5.6 specifics
GPT-5.6 chains tools in a single prompt, so you can say "browse for the latest figures, run the math in Python, and put the result in a table" and it will do all three. If a first answer feels shallow on a hard task, re-run it on the Sol tier with Max or Ultra mode rather than rewording the prompt. Save recurring preferences (your role, default tone, banned words) in custom instructions or a Project so they apply automatically and you don't repeat them. Memory carries context across chats, which is handy but worth reviewing so stale facts don't leak in. Claude Opus 4.8 and Gemini 3.x behave similarly here, but the modifier phrasing on this page transfers cleanly across all of them.
Copy-paste starter prompts
Each of these stacks a role, a format, a tone, and a reasoning booster. Fill the brackets and paste.
1. Draft any work email
You are a clear, concise business writer. Write an email to [RECIPIENT] about [TOPIC].
Context: [2-3 key facts]. My goal is [OUTCOME].
Format: Under 120 words, one call to action, professional but friendly tone. Give me 2 subject-line options.
Before you write, ask me anything you need to make it accurate.Best for: the daily inbox. See more in our fill-in-the-blank work email templates.
2. Summarize a long document
You are a sharp executive assistant. Summarize the text below.
Format: A 3-sentence overview, then 5 bullet points of the key decisions, then a "Next steps" list of action items with owners.
Constraints: Neutral tone. Flag anything unclear at the end.
Text:
[PASTE TEXT]Why it works: it fixes the structure so long inputs always come back scannable.
3. Compare options in a table
Act as a decision analyst. Compare [OPTION A], [OPTION B], and [OPTION C] for [USE CASE].
Format: A table with columns Option, Best for, Cost, Main tradeoff. Below the table, give a one-sentence recommendation and the assumption behind it.
Browse the web for current pricing and cite your sources.Best for: buying decisions and vendor picks.
4. Improve your own writing
You are a senior copy editor. Improve the draft below without changing its meaning.
Do this in 3 passes: (1) fix clarity and flow, (2) cut filler and jargon, (3) tighten to under [WORD COUNT] words.
Return the final version, then a short bullet list of what you changed and why.
Draft:
[PASTE DRAFT]Why it works: the self-critique pass catches weak spots a single rewrite misses.
5. Brainstorm with options
You are a creative strategist. I need ideas for [GOAL].
Give me 3 distinct directions. For each: a one-line concept, why it could work, and the biggest risk.
Keep it punchy. After the list, ask me one question that would help you go deeper on my favorite.Best for: escaping a blank page fast.
6. Interrogate a long document
You are a meticulous analyst. I'm pasting a long document below (it may be dozens of pages).
Do three things:
1. A 5-bullet executive summary.
2. A table of every commitment, deadline, or number, with the section it came from.
3. The 3 things a careful reader should question or push back on.
Only use what's in the document. If something isn't stated, say "not specified."
Document:
[PASTE THE WHOLE DOCUMENT]Why it works: the 272K-token window holds a whole report at once, and pinning each fact to its section keeps the model honest instead of paraphrasing from memory.
Frequently Asked Questions
What is the universal ChatGPT prompt formula?
Role + Task + Context + Format + Constraints. Tell ChatGPT who to be, what to do, the background it needs, how to structure the output, and any limits like word count or tone.
Do I still need role prompts like "act as a" in GPT-5.6?
Yes. Roles remain the fastest way to set expertise, vocabulary, and priorities. GPT-5.6 follows them reliably and combines them well with tool use and custom instructions.
Which GPT-5.6 model should I pick — Sol, Terra or Luna?
Use Terra for everyday work; it is the balanced default and what free and Go users get. Switch to Sol for long reports, agentic coding and deep research, and drop to Luna for quick rewrites and high-volume tasks. The modifier phrasing on this page works the same across all three.
What are Max mode and Ultra mode?
They are reasoning modes that make the model work harder on tough jobs. Max mode gives a single hard problem extra compute; Ultra mode, on the Sol tier, coordinates four agents in parallel for demanding multi-step workflows. Paid tiers also expose effort controls to dial reasoning up or down per task. Leave them off for routine prompts.
What format modifiers work best?
Ask for a table, numbered steps, a bullet list, JSON, or a strict word limit. Naming the exact structure produces far more consistent output than leaving format open.
How do I make ChatGPT think more carefully?
Add reasoning boosters like "think step by step," "ask me clarifying questions first," "give three options," or "critique your draft, then improve it."
How do I refine an answer without rewriting my prompt?
Use quick follow-ups: "make it shorter," "expand section 2," "more formal," "simplify for a non-expert," or "same but in a friendlier tone."
Should I put modifiers in custom instructions or in the prompt?
Put durable preferences (tone, format defaults, who you are) in custom instructions or a project. Put task-specific modifiers in the prompt itself so they only apply when needed.
How long should a ChatGPT prompt be?
As long as it needs to be to include role, task, context, and format. A clear five-line prompt beats a vague one-liner. Cut filler, not detail.
Can I stack multiple modifiers in one prompt?
Yes. Most strong prompts combine a role, a format, a tone, and one reasoning booster. The starter prompts on this page show how to layer them cleanly.
How much text can I paste into one ChatGPT prompt?
GPT-5.6's Sol, Terra, and Luna tiers all hold a 272,000-token context window — roughly a 500-page document — so you can paste an entire report or contract and ask across it. Tell it to only use what's in the text and to pin each fact to its section, or it will start paraphrasing from memory.
Do these modifiers work in voice mode?
Yes. The GPT-Live voice model (July 2026) listens and speaks at the same time, so you can state the role, format, and length out loud and interrupt to steer. Use voice for brainstorming and quick rewrites; switch to text when you need a table or code to copy.