20 ChatGPT Prompt Tips for Better Results (2026)

ChatGPT crossed 900 million weekly users in early 2026, according to OpenAI's own figures reported by TechCrunch. Most of them type a few words, take the first answer, and never find out how much better it could have been. The gap between a flat reply and a genuinely useful one is rarely about clever wording. It's about a handful of habits.

These are 20 tips that close that gap, organized into four groups: sharper prompts, better conversations, the setup features most people never touch, and the tools that do real work. This isn't a from-scratch lesson on building a prompt; our guide to writing prompts covers that. These are the tactical moves that pay off once you have the basics down.

Everything here is current for GPT-5.5, the default model as of June 2026, with notes where a 2026 change matters. Grab the tips that fit how you work.

Write a sharper prompt

Most weak answers trace back to a weak prompt, not a weak model. These five fix the input, and they're the ones to reach for first because they cost nothing and change the most.

1. Be specific about what you want

This is the single biggest lever, and it's the reason a lot of people think ChatGPT is "just okay." A vague request forces it to aim at the average of everything it has read, which is exactly what generic output is. Vague in, vague out:

Write a cover letter.

Spell out the goal, the audience, the format, the length, and the tone, and the same model hands you something you can almost use as-is:

Write a 200-word cover letter for a junior data analyst role at a fintech startup. I have one internship and a stats degree. Confident but not arrogant, plain language, and no cliches like 'I am writing to express my interest'.

The more of those details you pin down, the less it has to guess, and guessing is where bland answers come from.

2. Give it a role

A role sets the expertise, the vocabulary, and the point of view before it writes a word. "You are an experienced hiring manager" produces sharper, more skeptical feedback than the same request with no role, because the model answers from that vantage point instead of a neutral middle. It's the fastest way to raise the level of an answer in a single line.

You are an experienced hiring manager at a fintech startup. Review my cover letter below and tell me whether you'd shortlist it, what's missing, and the two changes that would help most. Letter: [paste].

3. Show the format you want

If the shape of the answer matters, describe it or hand over a template. Asking for a table, a five-step checklist, or a strict word count saves you from reformatting a wall of text afterwards. It often improves the thinking too, because a structure quietly forces the model to be complete rather than trailing off.

Summarise this article as a table with three columns: the claim, the evidence given, and how strong that evidence is (weak, mixed, strong). One row per main claim. Article: [paste].

4. Show an example, don't just describe one

Pasting a sample of the style you want beats describing it almost every time. Prompt people call this "few-shot" prompting, and it works because the model is far better at copying a pattern it can see than at interpreting adjectives like "punchy" or "professional," which mean different things to everyone.

Here are two of my past LinkedIn posts so you can match my voice: [paste two]. Now write a new post on [topic] in the same voice, length, and rhythm.

5. Say what you don't want

Telling it what to avoid is as useful as telling it what to do. Most of the tells that make text read as machine-written are habits you can ban outright, and a short "don't" list cleans up a draft faster than any amount of editing after the fact. It's the quickest route to output that doesn't sound like a robot.

Rewrite this to sound human: no em dashes, no 'In today's fast-paced world', no rule-of-three lists, and no words like 'delve', 'leverage', or 'robust'. Keep it plain and direct. Text: [paste].

Get more from the conversation

The first answer is a starting point, not the finish line. These five lift quality after you hit enter, where ChatGPT goes from a vending machine to something closer to a collaborator.

6. Ask it to question you first

A quick way to stop it guessing is to make it gather context before it commits to an answer. Its questions often surface things you hadn't thought to include, and answering five short ones takes less time than spotting and fixing a wrong draft. It turns a one-way request into a short briefing.

I want help planning a 30-day content calendar for my bakery. Before you write anything, ask me up to five questions you need to do this well.

7. Refine in follow-ups instead of restarting

Treat it as a back-and-forth. A run of small corrections almost always beats one enormous prompt, because you can react to what it actually produced rather than trying to predict it in advance. Keep the thread going and steer, the way you would with a real assistant.

Good start. Now cut it by a third, make the opening punchier, move the example to the top, and lose the last paragraph.

8. Ask it to work through the steps

For anything with reasoning, having it show its working lets you catch a wrong turn instead of trusting a confident final number. It also tends to reason more carefully when it knows it has to explain each step. On genuinely hard problems, switch to GPT-5.5 Thinking, which reasons longer before it answers.

Work through this step by step before giving the answer, and show your reasoning at each step so I can check it: [problem].

9. Make it grade and fix its own answer

One of the most underused moves is to turn the model into its own editor. It's often better at spotting the flaws in an answer than at avoiding them in the first place, so asking it to score and rewrite its own work gives you a stronger second draft for free. Do this whenever the first reply is close but not quite there.

Grade your last answer out of 10 against my original request, list its two biggest weaknesses, then rewrite it to fix them.

10. Ask for the other angles

Forcing different viewpoints surfaces blind spots, which is exactly what you want before a decision. A single-perspective answer tends to quietly agree with the way you framed the question, so asking for the case against, or how a different person would see it, breaks that and shows you what you were missing.

I'm thinking of [decision]. Give me the strongest case for it, the strongest case against, and how it looks to [a customer / my finance team / me in two years].

Set ChatGPT up once

Here's what separates casual users from power users, and it isn't clever wording: most people never touch the features that carry context for them. Set these up once and every conversation starts ahead instead of from scratch.

11. Set your Custom Instructions

Tell ChatGPT who you are and how you like your answers, once, in Settings, and it applies that to every conversation automatically. It's the difference between re-explaining your job, your audience, and your preferences in each new chat and never doing it again. A good set of instructions quietly improves output you haven't even asked for yet.

(Paste into Settings, under Custom Instructions) I'm a small business owner writing for non-technical customers. Default to plain English, short paragraphs, and concrete examples. Skip the preamble and get to the point. If a request is ambiguous, ask before assuming.

12. Let it remember the things you repeat

Memory keeps useful context across separate chats, so it holds on to your business, your projects, and your preferences without you pasting them in each time. You stay in control of it: you can see what it has stored, add to it on purpose, or clear anything you'd rather it forgot.

Remember that I run a two-person coffee roastery called North Lane, our tone is warm and a little playful, and our main customers are local cafes. Use this in future unless I say otherwise.

13. Use Projects for anything ongoing

A Project is a workspace with its own instructions, files, and chats, so context builds up instead of resetting each time you start. Drop a brand guide or a recurring brief in once, and every chat inside the Project already knows the rules without you repeating them. It's the cleanest way to keep a long piece of work consistent.

(As a Project instruction) Everything in this Project is for the North Lane newsletter. Match the brand voice in the attached style guide, keep issues under 250 words, and always end with one clear call to action.

14. Hand it the real document, don't describe it

GPT-5.5 takes large inputs, so attach or paste the actual file instead of summarising it from memory. You get answers grounded in the real document, which is both more accurate and less work than trying to describe what's in it. This is where the big context window earns its keep.

I've attached our 40-page Q2 report. Pull out the five numbers that matter most for a board update, each with one line on why it matters and the page it's on.

15. Edit in writing and code blocks

In GPT-5.5, drafts and code now appear as editable blocks right in the chat, after OpenAI retired the separate Canvas panel in 2026. You can edit a block directly or ask it to revise only that part, which makes reworking a long piece far less painful than scrolling back through a chat thread to find the bit you meant.

Put this draft in a writing block so I can edit it, then change only the second paragraph to be more formal and leave the rest exactly as it is.

Put its tools to work

ChatGPT is more than a text box. These five use what it can actually do, and they're where it stops being a writing aid and starts being a tool.

16. Have it search the web for current facts

It can't know recent events from memory, so anything from the last few months needs a live look-up. Tell it to search and cite, and you turn a confident guess into something you can check against a real source. Skip this step on anything time-sensitive and you're trusting a model's memory of a world that has moved on.

Search the web for the latest on [topic] from the past month. Give me a short briefing with the three most important updates, each with a source link, and flag anything sources disagree on.

17. Analyze a spreadsheet or file

Upload data and let it do the number work: trends, outliers, a quick chart. It runs actual analysis on the file rather than eyeballing it, so you can ask follow-up questions and trust the maths more than a guessed-at estimate. It's a genuine analyst for the boring first pass.

Here's a CSV of my monthly sales: [attach]. Find the three clearest trends, anything that looks off, and make one chart worth sharing. Explain each finding in plain language.

18. Generate an image, then refine it

Describe a scene the way you'd brief a designer, then adjust it in follow-ups. The first image is rarely the final one, and small directions like "wider shot, softer light, more empty space" get you there faster than starting over with a brand new prompt each time.

Create a warm, minimal product photo of a bag of coffee beans on a wooden counter, soft morning light, lots of empty space for text. No words in the image.

19. Chain a few tasks in one prompt

It can run several steps in sequence from a single prompt, which saves the manual back-and-forth. Search, then summarise, then draft is a common chain, and stating the whole pipeline up front often produces a more coherent result than three separate asks that each forget the last.

Search for the top three articles on [topic] this week, summarise what each one argues, then draft a short LinkedIn post that pulls the common thread together in my voice.

20. Always verify what matters

The most important tip is the one all the others depend on. ChatGPT can state something wrong with total confidence, invent a source that looks real, or get a number subtly off, and it tends to do this exactly when you're least likely to notice. Treat it like a fast, capable assistant whose work you still check, especially facts, figures, names, dates, and anything you'll act on or publish. The web-search prompts help because they show their sources, so open the links and read the line being quoted rather than trusting the summary. The shortcuts here are real, but the judgment stays yours.

What this post does not cover

These are tips for getting better output, not a guarantee of a correct or complete answer, and nothing here is legal, financial, or professional advice. Features, model names, and limits change quickly, and some tools differ between the free and paid plans, so confirm the current details in ChatGPT. If you want the foundations rather than the shortcuts, start with our guide to writing prompts and the prompt engineering basics explainer, browse the free prompt library for ready-made prompts, or compare tools with our best prompts for Gemini roundup.

Sources

1. TechCrunch: ChatGPT reaches 900M weekly active users
2. OpenAI Help Center: Prompt engineering best practices for ChatGPT
3. OpenAI Help Center: Working with writing blocks and code blocks in ChatGPT