I remember the exact moment I realized I was using ChatGPT wrong. I'd been typing out the same three paragraphs of context who I am, what my business does, how I like my writing at the start of nearly every single conversation, for months. My partner glanced over my shoulder one evening, watched me retype it for what must've been the hundredth time, and just asked, "why are you doing that every time?"
I didn't have a good answer. I just hadn't looked into what the tool could actually do beyond the basic chat box.
That question sent me down a rabbit hole, and what I found genuinely changed how much value I get out of ChatGPT day to day. This isn't a "10 magic prompts" listicle. It's the stuff I've actually built into my routine after using this thing almost daily for two years, including the features I ignored for way too long and the mistakes that wasted my time.
Stop repeating yourself set up Memory and Custom Instructions properly
Here's the thing that fixed my "retyping context every time" problem: ChatGPT has two separate features that handle this, and most people either don't know about them or mix them up.
Custom Instructions are the rules you set once, in Settings, that apply to every new conversation your tone preferences, your profession, formatting rules you always want followed. I have mine set to always write in a conversational tone, avoid corporate buzzwords, and skip the "I hope this finds you well" openers. Set it once, forget about it.
Memory is different it's ChatGPT quietly picking up facts about you as you chat and carrying them forward automatically. Your job, your ongoing projects, things you've mentioned you care about. You can actually see everything it's stored under Settings > Personalization > Memory, and you can delete individual entries if something feels off or outdated. I do a quick cleanup of mine every couple of months, honestly, because it holds onto things longer than you'd expect.
Step-by-step to set this up:
- Go to Settings > Personalization
- Turn on Memory if it isn't already
- Add your Custom Instructions be specific ("write like a knowledgeable friend, not a textbook" works better than "be casual")
- After a week of normal use, check what's landed in your Memory list and prune anything wrong or irrelevant
Once I did this, probably 80% of the redundant context-setting in my daily use just disappeared.
Use Projects for anything that runs longer than one conversation
This one embarrassed me a little once I found it, because I'd been scrolling through a messy sidebar of a hundred loose chats trying to find "that one conversation about the client proposal" for months.
Projects let you group related chats together, with their own uploaded files and their own instructions, kept separate from your main chat memory. I now have a Project just for client email drafts, with instructions like "audience is B2B decision-makers, keep it under 150 words, never open with 'I hope this email finds you well.'" Every conversation inside that Project automatically follows those rules I don't retype them.
I also keep a Project specifically for a long-running research piece I've been working on, with all my source PDFs uploaded once. Instead of re-uploading a document every time I want to ask something new about it, it's just sitting there, available in every chat inside that Project.
Where this helps most: ongoing client work, long research projects, anything you return to across multiple sessions rather than a single one-off question.
Voice mode is genuinely underrated for thinking out loud
I was skeptical of voice mode for a long time it felt gimmicky. Then I started using it on walks when I was stuck on how to structure an article, and it changed how I brainstorm entirely.
Instead of typing out a stiff, organized question, I just talk through a half-formed idea the way I would with a colleague. "Okay so I'm trying to figure out how to explain this concept without it sounding like a textbook, here's what I'm thinking..." and it responds conversationally, asks a clarifying question back, and somehow the messiness of talking it out gets me to a better answer faster than typing ever did.
It works in the mobile app, and if you drive, it's also available through CarPlay now, which honestly might be its best use case turning commute time into actual progress on something instead of just sitting in traffic.
Deep Research: use it for the stuff that actually needs digging, not everything
This is the feature people either overuse or completely ignore. Deep Research sends ChatGPT off to actually browse multiple sources, read them, and pull together a structured report with citations it can take anywhere from five to thirty minutes depending on the topic.
I made the mistake early on of using it for simple questions I could've just asked normally, and wasting time waiting on something that didn't need that level of digging. Now I save it specifically for things like competitor research, "what are the actual current best practices for X," or pulling together a first-pass literature review before I start writing something myself.
Quick rule I use: if I can answer the question myself with two Google searches, I don't need Deep Research. If it's the kind of thing where I'd normally open fifteen browser tabs and lose an hour, that's exactly when Deep Research earns its keep.
Real examples of how I use it week to week
- Editing my own writing. I paste in a rough draft and ask it to flag anything that sounds stiff or over-explained, without rewriting my voice entirely. This is different from asking it to "make this better," which tends to flatten your personal style into generic AI-speak.
- Turning meeting notes into action items. I dump messy, half-typed notes from a call and ask for a clean list of follow-ups with owners attached. Saves me the ten minutes I used to spend organizing this manually after every call.
- Debugging a spreadsheet formula. I described what I wanted a formula to do, pasted the broken version, and it spotted a misplaced parenthesis in about four seconds that I'd been staring at for twenty minutes.
- Image edits, not just generation. You can now upload an existing image and ask it to change something specific swap a background, adjust text on a graphic instead of generating from scratch every time. I use this for quick social media graphic tweaks when I don't want to open a design tool for something minor.
Common mistakes I see people make (and made myself)
Writing vague prompts and blaming the tool. "Write me a blog post about marketing" gets you generic mush every time. "Write a blog post for small business owners who've never run paid ads, explaining the difference between Facebook and Google ads, in a friendly non-technical tone" gets you something usable. The gap between a bad prompt and a good one is enormous, and it's almost always the actual issue when people say ChatGPT "isn't that good."
Trusting output without checking it. I once had it summarize some financial figures for a client deck and didn't verify one number it was close but not accurate. Now everything factual gets a second check before it goes anywhere important. Treat it like a very fast, very capable assistant who occasionally gets confident about something wrong.
Never touching the settings. Most of the friction people complain about repeating themselves, losing context, generic-sounding output comes down to never opening Custom Instructions or Memory settings even once. Fifteen minutes of setup saves you months of repetition.
Using one giant chat for everything. If your sidebar looks like an unsorted junk drawer, Projects will fix that in about ten minutes, and future-you will be grateful.
Ignoring the model picker. On paid plans, you can actually choose which model handles your request instead of letting it auto-select. For anything that needs real reasoning untangling a complex problem, working through logic step by step picking the "Thinking" model rather than the fast default genuinely gives you a better answer. I didn't realize this existed for the first several months I was paying for Plus.
Where I've landed after two years of this
The honest truth is that most people are using maybe a third of what ChatGPT can actually do, myself included until pretty recently. The chat box is just the entry point. The stuff that actually saves real time Memory, Projects, Custom Instructions, knowing when to reach for Deep Research versus a quick question lives one settings menu away from where most people stop looking.
None of this requires being technical. It just requires spending fifteen minutes poking around the settings once, instead of retyping the same context for the hundredth time like I did.
){kind=link}
0 Comments