8 Hidden ChatGPT Features in 2026 Most People Don’t Know

Best for: students, freelancers, anyone juggling multiple ongoing topics in one account

Projects are folders that hold chats, uploaded files, and their own custom instructions. The part almost nobody knows: since late 2025, Projects are free for everyone, and each project can have project-only memory — meaning what ChatGPT learns inside that project stays inside it and does not leak into your other conversations.

Original use case: If you are preparing for UPSC or a bank exam alongside a day job, make one project called “Mains Answer Practice”, upload the syllabus PDF and two model answer sheets (free plan allows 5 files; Plus allows 25; Pro/Business 40), and set the project instruction to “Evaluate my answers out of 10 like a strict examiner, in under 150 words.” Every practice session lands in one place, and your office-related chats never get polluted with exam context.

Best for: writers, marketers, developers comparing two directions from the same starting point

Added in September 2025 after years of user requests, branching lets you fork a conversation at any message. The original chat stays untouched; the branch carries all context up to that point and goes its own way. It is web-only and works on every plan, including free — yet most users have never noticed the menu option.

Original use case: You have spent 20 minutes briefing ChatGPT on your D2C brand’s Diwali campaign. Now you want one version of the ad copy in formal English and one in casual Hinglish. Instead of confusing one thread with both, branch at the briefing message twice — one branch per tone — and compare the outputs side by side in two tabs. The 20 minutes of briefing is preserved in both.

Best for: daily briefings, recurring reminders, repeating research jobs

Tasks let ChatGPT run a prompt for you at a future time — once or on a schedule — and notify you when it is done. It started as a Plus/Pro/Team beta in early 2025 and is now also bundled into the ₹399 Go plan in India. You are limited to 10 active tasks at a time, and tasks run on standard models (not the Pro reasoning models).

Original use case: A mutual-fund-curious reader of ours set a Monday 8 am task: “Explain one mutual fund concept I have not asked about before, in 200 words, with one Indian example.” Fifty-two weeks later, that is a free personal finance course delivered before breakfast — no app, no newsletter signup, no spam.

Best for: troubleshooting anything physical or on-screen, hands-free learning, language practice

Everyone knows ChatGPT can talk. Far fewer know that inside voice mode on the mobile app, you can turn on your camera or share your phone screen, and ChatGPT responds to what it sees in real time. Standard voice is available on all plans; live video and screen sharing require Plus, Pro, Business, Enterprise, or Edu, with daily usage limits.

Original use case: Your parents get a suspicious “KYC update” SMS with a link. On a video call you cannot see their phone — but they can open ChatGPT voice mode, share their screen, and ask “Is this message a scam?” ChatGPT reads the screen and explains, in Hindi if they prefer, why a bank never asks for KYC via SMS links. It is genuinely one of the best digital-safety tools for non-technical family members.

Best for: sensitive questions, one-off queries you do not want influencing your memory

A temporary chat does not appear in your history, does not read or create memories, and is not used to train models. OpenAI deletes temporary chats from its systems after 30 days. It is the difference between asking a question and asking a question on the record — and it sits one click away on every plan.

Original use case: You want a second opinion on a medical report, a salary negotiation, or a family dispute — questions you would rather not have echoed back at you in unrelated chats for the next six months. Temporary chat keeps the conversation out of memory entirely. It is also the clean way to test prompts for work without your personal custom instructions colouring the output.

Best for: everyone — this is the page most users have never opened

ChatGPT quietly saves “memories” about you from your chats — your profession, your city, your preferences. In June 2026, OpenAI upgraded memory to keep context fresher and reduce stale or contradictory entries, and doubled memory capacity for Plus and Pro users. What stayed the same: you can read, delete, or wipe every memory, and most people never have.

Original use case: Before a big project, do a five-minute memory audit. One freelancer we know found ChatGPT had memorised an old day rate from a 2025 chat and kept anchoring negotiation advice to it. Deleting that single memory immediately changed the quality of its pricing suggestions. Stale memories silently distort answers — pruning them is the fix, and you can also simply tell ChatGPT in chat: “Forget my old rate, remember my new rate is X.”

Best for: anyone tired of ChatGPT’s cheerleader tone

ChatGPT now ships with selectable personality presets — Default, Friendly, Efficient, Professional, Candid, Quirky, Cynical, Nerdy — that apply across all your chats, including old ones. Underneath that sits the older, more precise tool: custom instructions, where you tell ChatGPT who you are and how to answer. The two stack, and saved memories sit on top of both.

Original use case: Set the personality to Efficient and add one custom instruction: “I am in India. Use Indian examples, INR for money, and Indian English spellings. Never pad answers with disclaimers I did not ask for.” That single line removes 80% of the US-centric examples and dollar pricing that make generic ChatGPT answers feel half-relevant here. Writers can keep a second preset workflow: switch to Candid when you want honest critique of a draft instead of polite applause.

Best for: CAs, lawyers, founders, anyone pasting confidential documents into ChatGPT

Rolled out from 4 June 2026 to all logged-in users — Free, Go, Plus, Pro, and self-serve Business — Lockdown Mode is an opt-in setting that restricts ChatGPT’s network-connected capabilities (live web browsing, Deep Research, Agent Mode, file downloads, connectors) to reduce the risk of prompt injection: attacks where malicious instructions hidden in a webpage or document trick the AI into leaking your data. You can still upload files manually for analysis.

Original use case: A CA reviewing a client’s financials with ChatGPT during tax season should work in Lockdown Mode by default. The analysis features still work on uploaded files — what gets blocked is the pathway by which a poisoned document or webpage could exfiltrate that data outward. OpenAI is explicit that this reduces risk rather than eliminating it, but it is the strongest built-in control available today.

Best for: anyone who has used ChatGPT for over a year and never backed it up

Two separate switches live in this menu and both matter. First, “Improve the model for everyone” — turn it off and your chats stop being used for model training. Second, Export Data — a full ZIP archive of every conversation you have ever had, emailed to you with a download link that expires in 24 hours.

Original use case: Your chat history is an asset. Export it once a quarter and you get a searchable record of every prompt that ever worked for you — the JSON inside the ZIP can be opened in any text editor. One content freelancer we spoke to mines her export file for her ten best-performing prompt patterns and reuses them as templates, instead of reinventing prompts from scratch each time.

Best for: anyone using ChatGPT on a laptop for more than 30 minutes a day

ChatGPT on the web has had a full shortcut set for years, and the discovery mechanism is itself a shortcut — which is presumably why nobody discovers it. Press Ctrl + / (Cmd + / on Mac) in any chat and the complete cheat sheet pops up.

Original use case: If you use ChatGPT for coding practice or college assignments, Ctrl + Shift + C alone saves hundreds of mouse trips per week — it copies the most recent code block without scrolling and hunting for the copy button. Combine with Ctrl + Shift + O and you can run a “new question, get code, copy code” loop without touching the mouse at all.

Best for: small business owners, students, anyone with a spreadsheet and a question

The + button in the composer accepts PDFs, Word files, Excel sheets, CSVs, and images — and for data files, ChatGPT does not just read them, it writes and runs actual Python code to compute answers (this is the “advanced data analysis” capability). Free users get a limited number of uploads; the ₹399 Go plan extends both file uploads and data analysis significantly; Plus and Pro extend further.

Original use case: A kirana or boutique owner exports six months of sales from their billing app as CSV and asks: “Which 10 products give 80% of revenue, and which weekday is slowest?” That is a Pareto analysis a consultant would bill for, done in two minutes. Students can upload a question-bank PDF and ask for the five most repeated topics across years — far more useful than asking generic “important topics” questions. Of all the hidden ChatGPT features here, this one converts the most sceptics.

Best for: getting noticeably better answers on hard questions without paying more

Most users never touch the model name at the top-left of the chat — but that menu decides how much “thinking” your answer gets. In 2026’s lineup (GPT-5.5 Instant became the default in May 2026), Instant answers fast and can automatically escalate complex requests to Thinking, which reasons before replying. Paid plans can pin Thinking manually and configure auto-switching. Retiring models get a sunset window — o3, for instance, leaves ChatGPT in August 2026 — so the menu also tells you what is going away.

Original use case: The trick is matching the mode to the stakes. Drafting a WhatsApp message? Instant. Reviewing a rent agreement clause, debugging code, or comparing two health insurance policies? Switch to Thinking first — the same question routinely produces a materially more careful answer. The mid-conversation switch matters too: you can change models partway through a chat when the conversation turns serious.