ChatGPT can recognise task complexity automatically. If you enable Auto-switch to Thinking in the Configure settings, it will silently decide when to use Thinking Mode based on your request — no manual switching, same chat box.

Ask something simple — "rewrite this sentence so it sounds more confident" — and it answers instantly, same as always. Ask something that needs real reasoning — "build me a 90-day content strategy with a weekly calendar" — and it quietly switches to Thinking Mode behind the scenes, takes a little longer, and comes back with a far more thorough answer.

But if you want to guarantee deep processing for a specific high-stakes task — a financial model, a legal-style argument, a full business plan — manually selecting Thinking Mode ensures full power every time, regardless of how the prompt happens to be phrased.

Web Search is the feature that turns ChatGPT from "a very smart textbook with a cutoff date" into something that knows what happened this morning. Without it, every answer is built from training data that stops at a fixed point in time. With it, ChatGPT goes out, reads current pages, and brings the answer back — with sources attached so you can check its work.

That distinction matters more than it sounds. Ask "what's Canva's latest pricing?" without Web Search and you might get a confident answer that's a year out of date. Turn Web Search on, and ChatGPT checks the actual current page before answering — and tells you where it looked.

Web Search pulls live information from the internet — news, company updates, pricing pages, and recent developments — and cites where each fact came from so you can verify it yourself.

It's perfect for: client research, industry news, competitor updates, recent product launches, and anything where "current" matters more than "comprehensive."

Let's make this real: you're pitching to Canva as a potential brand partner and need their latest news and positioning. Try it.

Add Photos & Files lets you upload images (PNG, JPG, PDF) and documents (Word, Excel, text files) for ChatGPT to analyse, summarise, or reference — directly in the conversation, with no manual retyping.

This is the feature that turns ChatGPT from a writing partner into a document analyst. Upload a 40-page PDF report and ask for the three key takeaways. Drop in a messy spreadsheet and ask it to spot the pattern. Screenshot a confusing chart from a client deck and ask what it actually shows.

Examples: upload a content brief to refine it; upload a competitor's PDF report to extract insights; upload a screenshot of analytics to interpret data; upload a contract to flag risky clauses; or upload a logo file and ask for variations.

Real example: you're handed a 12-page client brand guideline PDF the night before a pitch. Instead of skimming it manually, upload it and ask: "Summarise the brand voice, colour palette, and any messaging rules I need to follow." Minutes, not hours.

Most people treat ChatGPT like a vending machine: insert question, receive answer, walk away. The people getting 10x the value treat it more like a collaborator with a memory problem — brilliant and fast, but needing the right setup every time. The ten habits below are how you give it that setup without thinking about it.

None of these require a paid plan, a browser extension, or a "secret prompt" copied from somewhere. They're operating habits — small adjustments that, once automatic, change the shape of every conversation you have with the model.

Pick two or three of these to try this week. You don't need all ten on day one — you need one that becomes automatic, then a second, then a third. That's how leverage compounds.

Every plan — free included — now runs on a version of GPT-4o-class models with access to web search, file uploads, and image generation. The difference between plans isn't "what ChatGPT can do" — it's how often you can do it, how fast, and with how much room to think.

If you're a casual user — a few questions a week, the occasional draft — the free plan is genuinely enough. The moment ChatGPT becomes part of your daily workflow (research, content, client work, planning), the limits on the free plan start to cost more time than the subscription would cost in money.

ChatGPT's core model is trained on a fixed snapshot of data — it doesn't know about anything that happened after that point unless it looks it up. Web Search is the "looking it up" part. When it's relevant, the model quietly searches the web, reads a handful of current pages, and folds what it finds into its answer — with links so you can verify.

Ask something time-sensitive — "what's the weather in Lisbon this weekend," "who won last night's game," "what's the latest on [current event]" — and ChatGPT will usually search without being asked. You'll see it browsing, then citing sources in the response.

For anything where freshness matters and you're not sure it'll search automatically, just say so: "Search the web for this" or "Look up current pricing for X." You can also select Web Search directly from the tools menu before you type, which is the most reliable way to guarantee it.

One habit worth building: when an answer feels like it might be out of date — pricing, availability, "is this tool still around" — add "search for the current..." to your prompt. It costs a few seconds and removes an entire category of stale-information mistakes.

Every time you give ChatGPT context — "you're a copywriter for my brand," "here's our style guide," "always format responses this way" — that setup disappears when the chat ends. A Custom GPT is what happens when you save that setup permanently. It's still ChatGPT underneath, just pre-loaded with the instructions, files, and tone you'd otherwise type out every time.

From the sidebar, choose Explore GPTs → Create. The builder asks what you want it to do, then generates a configuration: a name, a description, custom instructions, and optionally reference files it can pull from. You can keep refining it in plain language — "make it more concise" or "always ask for the target audience first" — until it behaves the way you want by default.

Start small: pick one repetitive task you do with ChatGPT at least weekly, and build a single Custom GPT around it. If it saves you even thirty seconds of context-setting per use, it pays for itself within days.

Without memory, every new chat starts from zero — ChatGPT has no idea what you do, what you've already discussed, or how you like things formatted. With memory turned on, it can quietly hold onto useful facts — "works in real estate," "prefers short answers," "is building a side project called X" — and bring them into future chats automatically.

"Memory" (saved facts it carries forward) is different from chat history (the model referencing earlier conversations you've had, even ones you didn't ask it to "remember"). Both can be switched on or off independently in Settings → Personalization, and both can be reviewed.

A quick monthly habit: open Settings → Personalization → Memory, skim what's stored, and delete anything outdated. A clean memory means more relevant answers — a cluttered one can pull in context you no longer want.

"Generate an image of a coffee shop" will give you a coffee shop — but probably not the one you pictured. ChatGPT's image generation is highly capable, but it fills in every detail you don't specify with its own default. The more of those defaults you replace with your own choices, the closer the result lands to what you actually wanted.

Strong image prompts tend to stack the same five ingredients: subject (what/who), style (photo, illustration, 3D render, flat design), composition (close-up, wide shot, from above), lighting/mood (warm, moody, bright and clean), and context (where, what's in the background). You don't need all five every time — but the more you specify, the less the model has to guess.

When an image is close but not quite right, resist the urge to rewrite the whole prompt from scratch. Point at the one thing that's off — "the text is unreadable, remove it" or "make the lighting brighter" — and let the model adjust just that part.