🕹️ 8 ChatGPT cheat codes nobody uses

Most people use ChatGPT like a search box and get search-box results. These eight techniques are the difference between a toy and a tool — no plugins, no paid tier required for most.

1. Custom instructions: set it once

In settings, tell it who you are and how you want answers ("I run a small business in Nebraska; be concise; no filler; ask before assuming"). Every chat starts smarter. This is the highest-leverage 5 minutes in the app.

2. Manage the memory deliberately

ChatGPT remembers things across chats. Curate it: "remember that my business is [X]" for things that should stick, and review/delete stale memories in settings so old context stops contaminating new answers.

3. Temporary chat for clean-room questions

When you want an answer uncolored by your history (or you're asking something one-off), use a temporary chat. No memory in, none saved out.

4. Voice mode for thinking, not just asking

Talk through a problem on a drive: "let me think out loud about pricing — interrupt me with questions." Rambling aloud plus a summary at the end beats staring at a blank page.

5. Photos as input

Snap the thing instead of describing it: a jobsite, an error on a screen, a competitor's flyer, your handwritten notes. "What's wrong here?" with a photo outperforms three paragraphs of description.

6. Force the format

Unformatted answers are where good information goes to be ignored.

7. The re-roll with feedback

Don't accept draft one, and don't just hit regenerate. Say what's wrong: "too formal, cut it by half, lose the buzzwords." Three directed revisions beat ten random re-rolls.

8. Projects for anything ongoing

Group related chats with shared files and instructions per project (your marketing, a renovation, a hire). Context stays put; you stop re-explaining.

Bonus: 4 mode prompts to layer on top

Beyond features, a few one-paragraph "modes" change the answers themselves. Say the mode name to keep it active for the chat:

• Straight-shooter mode: "Stop defaulting to agreement for this chat. If my idea is weak, say so and why. No flattery openers, no both-sides hedging — talk to me like a smart peer who wants me to win."
• Forecast mode: "Using what you know about me from this chat and memory, project how [decision] likely plays out over 12 months — what I'll do, what happens, what I'm underweighting. The read, not the pep talk." (Sharper the longer your memory's been on.)
• Plain-English mode: "Explain this like I'm smart but new: no jargon, define any technical word in the same breath, use an everyday analogy." Then: "now give me the next layer deeper."
• Natural-voice mode: "Rewrite so a real person could have written it: contractions, varied sentence length, no 'in conclusion,' no perfectly balanced lists, cut the filler openers." Feed it two samples of your writing and it becomes YOUR voice.

Stack them — straight-shooter + forecast on a big decision, plain-English + "next layer" for learning anything. Same stacking logic as the Claude shortcuts system.