ChatGPT tips and tricks: 25 habits and features that improve results

The ChatGPT tips that make the biggest difference are setting up custom instructions and projects once, giving full context in every request, and editing your prompt instead of arguing with a bad answer. This guide collects 25 tips: settings most people skip, prompting habits, lesser-known features and shortcuts, and quick ways to check answers before you rely on them.

Which ChatGPT tips and tricks make the biggest difference?

Three habits improve ChatGPT results more than any trick: set up custom instructions once, give full context in each request, and edit your prompt when an answer misses instead of arguing in follow-ups. Everything else in this list builds on those.

The 25 tips below are grouped so you can skim:

Features and limits differ by plan and change often. Everything here was checked against OpenAI's help center and documentation in October 2026. If you are brand new, start with the guide on how to use ChatGPT and come back.

What ChatGPT settings should you change first?

Set up custom instructions, review memory, organize work into projects and check your data controls. These take about 15 minutes once and improve every later chat.

1. Write real custom instructions

Custom instructions are guidance you write once that ChatGPT applies to your chats. On Plus, Pro, Business, Enterprise and Education plans the limit is 5,000 characters, which is room for a proper profile. Include:

About me: Product manager at a 40-person B2B SaaS company (logistics software).I use ChatGPT for: specs, stakeholder emails, data questions, research.How to answer:- Lead with the answer, then the reasoning. No preamble.- Use plain US English. No filler phrases or motivational tone.- When I ask for writing, match a direct, friendly, concise voice.- If my request is ambiguous, ask up to 3 questions before answering.- Say "I'm not sure" when you are not, and mark assumptions.- For numbers, show the calculation.

2. Review what memory knows about you

ChatGPT's memory has two parts, per OpenAI's memory help page: saved memories and the ability to reference past chat history. Open the memory settings, delete anything wrong or outdated, and decide whether you want chat history referenced. Turning off Reference chat history does not delete saved memories; those are managed separately and are not removed when you delete the chat they came from.

3. Put ongoing work in projects

Projects keep related chats, files and instructions together. Files in a project's sources are available across its chats, and project instructions apply to every chat in it and override your global custom instructions, according to OpenAI's docs. Make one project per client, course or long-running piece of work.

4. Check your data controls

On personal plans, OpenAI may use your conversations for model training while Improve the model for everyone is on. You can switch it off in Data controls. Business, Enterprise and Edu workspaces are excluded by default. The guide on whether it is safe to use AI at work covers what else to check before pasting work material.

5. Pick the right model for the job

The model picker matters. Faster models are fine for quick rewrites and simple questions; reasoning models are better for analysis, planning, math and code. If an answer feels shallow, switch to a reasoning model before rewriting the prompt. Which models you see depends on your plan; Plus at $20 a month includes OpenAI's reasoning models with usage caps, per the plans page.

What prompting habits get better ChatGPT answers?

Give context, show examples, specify the format and ask ChatGPT to question you first. Most bad answers come from prompts that leave the model guessing.

6. Say who it is for and what it is for

"Summarize this report" gets a generic summary. "Summarize this report for our CFO, who will decide on Friday whether to renew the contract; focus on cost and risk" gets a useful one.

7. Show an example of what good looks like

Paste a paragraph, an email or a table you liked and say "match this." One example often does more than a paragraph of instructions about tone.

8. Ask it to ask you questions first

I need a project kickoff email for a new client. Before writing anything,ask me the 5 questions you need answered to make it specific. Wait for my answers.

This one habit fixes most vague output, because the model stops filling gaps with generic content.

9. Specify the output format

Name the structure: "a table with columns X, Y, Z," "five bullets, each under 15 words," "a numbered checklist," "JSON with these keys." Format instructions are followed far more reliably than tone instructions.

10. Edit your message instead of piling on corrections

When an answer goes wrong, click edit on your original message, fix the prompt and resend. Long threads full of "no, not like that" carry the confusion forward. A clean prompt with the missing detail added usually works on the next try.

11. Start a new chat for a new topic

Each chat carries its context. Mixing a marketing plan, a tax question and a recipe in one thread dilutes all three. New topic, new chat, or a chat inside the right project.

12. Ask for options, then choose

"Give me three different approaches with trade-offs" beats "what should I do?" You get to apply judgment instead of accepting the first idea.

13. Ask it to critique its own answer

After a draft: "List the three weakest points in this answer and fix them." Or for decisions: "Argue against this plan as a skeptical CFO." Models catch many of their own gaps when asked directly.

For deeper techniques, see the prompt engineering guide, and for ready-made templates by role, ChatGPT prompts for work.

Which ChatGPT features do most people miss?

Temporary chats, keyboard shortcuts, quick chat, file analysis, deep research, scheduled tasks and custom GPTs are the features most people never try. Each saves time on a specific kind of task.

14. Use temporary chats for one-offs

Temporary chats do not create or update memories, per OpenAI's help page. Use them for sensitive one-off questions or when you do not want a topic shaping future answers. OpenAI says a copy may be kept for up to 30 days for safety purposes, so they are not a way to share confidential data.

15. Learn five keyboard shortcuts

In the ChatGPT desktop app, these save the most time (full list):

You can customize them under Settings, Keyboard Shortcuts.

16. Analyze files instead of pasting text

Upload spreadsheets, PDFs and documents directly. For data, ask ChatGPT to run the analysis and show the code or formulas it used, so you can check the numbers. Uploading the file also avoids formatting loss from copy and paste.

17. Use deep research for questions that need many sources

Deep research browses many pages and writes a report with citations, which suits market scans, literature overviews and "what are my options" questions. It takes longer than a normal answer, so give it a precise brief: scope, time period, what to exclude and the format you want.

18. Schedule recurring tasks

Paid plans include scheduled tasks: ask ChatGPT to run something at a set time, such as a Monday summary of news in your industry or a daily reminder with a checklist. Phrase the task as you would for an assistant who cannot ask follow-up questions.

19. Build a custom GPT for repeated jobs

If you paste the same long instructions every week (a voice guide, a review checklist, a support style), turn them into a custom GPT. Teammates can use it without learning the prompt. Use projects for your own ongoing work and GPTs for tools other people use.

20. Use voice for thinking out loud

Voice conversations work well for brainstorming while walking, practicing a presentation or rehearsing an interview. Afterward, ask for a written summary of what you discussed.

How do you check ChatGPT's answers?

Ask for sources, check the ones that matter, recompute numbers and watch for invented specifics. ChatGPT is fluent even when it is wrong, so fluency is not evidence.

21. Ask for sources and open them

For factual claims, ask for links and open the ones behind your key conclusions. A link that does not say what the answer claims is a red flag for the whole answer.

22. Recompute anything that matters

For calculations, ask ChatGPT to run them as code and show the steps, then sanity-check the result with a rough estimate of your own.

23. Be suspicious of exact quotes, dates and statistics

These are where models invent most often. If a quote or number will appear in your work, verify it in the original source. The post on AI hallucinations explains why this happens.

24. Ask "what would change your answer?"

This surfaces assumptions. If the answer depends on something you know is different (your country, your company size, a newer version of a tool), you can correct it.

25. Get a second opinion

For important answers, paste the question into a fresh chat, or into another assistant, and compare. Disagreement tells you where to look closer. The ChatGPT vs Claude vs Gemini comparison covers where each one is strongest.

How to put these ChatGPT tips into practice today

Do the one-time setup now and pick two habits to practice this week.

1. Write custom instructions using the template in tip 1.
2. Review memory and delete anything wrong.
3. Check Data controls and set training to match your comfort level.
4. Create one project for your biggest ongoing piece of work and add its key files.
5. Practice two habits this week: "ask me questions first" (tip 8) and editing instead of correcting (tip 10).
6. Try one new feature a week from tips 14 to 20.

If you use ChatGPT's Codex or another coding agent for long programming tasks, Orkio runs agents such as Claude Code and Codex on an always-on cloud box so they keep working after you close your laptop.

Frequently asked questions