Advanced ChatGPT Tips — getting real work out of it
Most of the gap between people who find ChatGPT useful and those who don't comes down to four things: how they brief it, what standing instructions it carries, where its context lives, and whether it can run without them. Facts here come from OpenAI's help center and are linked. The prompting patterns are ours, marked as such.
1. Brief ChatGPT Work like a contractor
OpenAI's own instruction for Work is to "describe the outcome you want" and add "any files, constraints, and review criteria" (Work and Codex). In practice:
- Name the deliverable: "a 10-slide deck", "a spreadsheet with one row per supplier", "a two-page brief".
- Say what done looks like: "every figure cited to a source", "under 600 words", "ready to send".
- Give it the inputs: attach files, or start the task from a Project that already holds them.
- Iterate in the same chat. Work keeps the task context, so "tighten section 2 and add a risks column" is enough.
2. Custom instructions that actually change behaviour
- Mind the limit: 1,500 characters on Free and Go, 5,000 on paid plans (OpenAI). Spend it on rules that matter every time, not on biography.
- Prefer "do X" over "don't do Y". OpenAI's GPT-writing guidance says the same: positive, concrete instructions work better than long lists of prohibitions (OpenAI).
- Use structure: headings or bullets for priorities, and "when X happens → do Y" for workflows.
- Include one example of an output you liked. It often does more than three rules.
- Edits apply immediately to all chats, including ongoing ones.
3. Let context live in the right place
| Put it in… | When |
|---|---|
| Custom instructions | Rules and style that apply to everything |
| A Project | Files and instructions for one client, product or long job. Start Work tasks from it |
| Memory | Background it can pick up over time. Audit it; see Memory |
| A plugin | A reusable workflow with its own skill and connected apps |
4. Make it run without you
- Scheduled tasks repeat a job or watch for changes. Paid plans can run hourly; Free no more than daily (OpenAI).
- Event triggers (Plus and above) run a Work task on new Gmail messages, Slack channel messages or GitHub pull-request activity. Our tip: write the condition narrowly ("only messages from @client.com that mention an invoice") so you don't burn allowance on noise.
- Active-task caps are small (5 on Plus), so consolidate related checks into one task.
5. Stay inside your limits
- "Unlimited" covers everyday text chats only. Reasoning, Pro models, Work, Codex, image generation and uploads have their own allowances (OpenAI).
- Choose the model on purpose in Work. GPT-6 Astra uses the shared Work/Codex allowance faster than GPT-5.6 Sol. Our tip: use Default or Sol for routine tasks and switch to Astra for hard ones.
- With Instant selected, ChatGPT can still add reasoning automatically, which draws on your reasoning allowance. Adjust Higher intelligence in Settings → General if that matters to you.
- There are no per-tool-call charges inside a ChatGPT plan. Per-call and per-session tool pricing applies only to the API.
6. Plugins, carefully
- Read what a plugin includes (skills, apps) before installing, and connect only the apps it needs (OpenAI).
- An OpenAI Verified badge is a quality review, not a privacy or security sign-off for your data.
- If you relied on a custom GPT, migrate it before Dec 11, 2026. See Custom GPTs → Plugins.
← Back to ChatGPT hub · See also: Setup · OpenClaw skills for the self-hosted equivalent
Official sources
Checked against OpenAI’s help center and chatgpt.com/pricing (US prices) on 2026-09-27. If anything here disagrees with these pages, trust the vendor docs.
← Back to ChatGPT hub · See also: OpenClaw Skills Guide for similar patterns