4 Powerful ChatGPT Features You Might’ve Missed

This Week

Over the past month, I’ve been tracking the major announcements from the big players in AI—Google, Microsoft, and most recently, Apple. Each brought slick presentations and a lot of fanfare, but one of the most widely used AI tools today is a little more understated.

There’s no flashy keynote or million-dollar intro video, just a casual roundtable of employees explaining their latest feature—sometimes brilliant, sometimes oddly awkward. And unless you’re watching closely (or decoding their deeply technical release notes), you might not even know what’s changed in your account.

So this week, after covering all the conference headliners, I’m turning the spotlight on OpenAI’s ChatGPT. What's new, what’s helpful, and what you might have missed.

Here we go.

4 Recent ChatGPT Features You Should Be Using

OpenAI is constantly rolling out new features, which makes it tough to keep up with what’s available. Here are four recent additions worth adding to your workflow:

1. Deep Research
Need a quick but credible answer to a complex question? Deep Research uses the o3 model to browse the web and generate rich, cited reports in minutes. Think of it as ChatGPT with a research assistant built in. Just select the “run deep research” option in the chat bar.

2. Connectors
ChatGPT can now plug directly into tools like Google Drive, Dropbox, Outlook, and GitHub. Instead of uploading files manually, just say “Find my Q2 planning doc” and let it pull what it needs—just make sure you check your privacy settings first. Want the full walk through? Watch the OpenAI explainer video here.

3. Advanced Voice Mode
The new voice experience in ChatGPT is surprisingly natural, featuring realistic tone, pacing, and expression, (voice character Vale has become a big part of my day). It allows you to speak directly with ChatGPT and get spoken responses back, making interactions feel more like a real conversation than a typed exchange.

Voice conversations are available on ChatGPT’s mobile and desktop apps. Plus, Team and Enterprise users can access advanced voice capabilities with daily usage limits. Free-tier users can try a daily preview powered by 4o-mini. To start, just tap the voice icon in the chat bar.

4. Record Mode (Team/Enterprise only)
Running an in-person meeting or workshop? Record Mode (Mac desktop only) lets ChatGPT capture the conversation, transcribe it, and generate a summary automatically. Great for working sessions where you don’t want to stop and take notes. Only available in Pro accounts for now. Visit the help page for the how-to guide.

Projects or Custom GPTs? Here’s How to Choose.

If you’re still starting every ChatGPT session from scratch, it’s time to try Projects.

Projects give you a dedicated workspace where you can save chats, upload reference files, set instructions, and choose your model—all in one place. It’s a simple way to keep your work organized and easy to build on.

Not sure how Custom GPTs are different from Projects? It’s easy to get confused—both let you upload files, set instructions, and shape how the model works. But here’s the key: Custom GPTs are designed to be shared. You’re building a standalone assistant others can use, complete with its own personality and app connections. Projects, on the other hand, are private. They’re like your own work files—organized, contained, and just for you.

Alicia from Enovair put together a great walkthrough that shows how to structure Projects for focus and consistency. Watch the video here.

A Guide to OpenAI’s Latest Models

Confused by all the ChatGPT model names? You’re not alone. Between GPT-4o, o3, 4.1-mini, and now 4.5, it’s hard to know what does what, or which one you should be using.

OpenAI hasn’t quite worked out a consistent naming system yet, which makes it harder for the rest of us to keep track. Unlike Apple’s M1 → M2 → M3 chip lineup, there’s no clear progression or logic you can follow. Some names are internal codenames, others hint at features, and a few just seem...random.

So if you’ve been wondering what actually sets these models apart, here’s a quick breakdown to help you make sense of it all.

GPT Models (like GPT-4o)

These are your all-purpose, fast, and flexible tools—great for day-to-day productivity and creative tasks. They’re also multimodal, meaning they can process text, images, and audio.

Best for:

• Drafting emails, blogs, or social posts

• Translating content between languages

• Summarizing long articles or documents

• Creating visuals from prompts (if enabled)

• Voice chat, image analysis, and casual conversation
  

“O” Models (like o3)

These are your deep thinkers. While they may take a little longer to respond, they excel at precision and structured reasoning—ideal when accuracy really matters.

Best for:

• Solving math or physics problems

• Writing and debugging code

• Planning multi-step processes or workflows

• Answering technical or scientific questions

• Evaluating logic puzzles or decision trees
  

In short:

• Use GPT-4o when you want speed, versatility, or multimodal input

• Use o3 when you need accuracy, logic, or step-by-step analysis
  

Here’s a handy comparison chart I made (with a little help from ChatGPT o3):

Which LLM is Right For Your Privacy Needs?

Using a large language model (LLM) always comes with trade-offs. Who sees your data? Where does it go? Will it train the model? The answers vary widely depending on the platform—and the defaults aren’t always in your favor.

Some LLMs offer clear, customizable privacy controls. Others bury the settings or skip them altogether. If you're serious about minimizing risk, your safest options still sit behind an enterprise paywall. ChatGPT Enterprise and Microsoft Copilot for Business are your best bet. Both licenses ban model training and protect your data. OpenAI deletes enterprise chats within 30 days of your request, and Microsoft keeps Copilot traffic inside the Microsoft 365 boundary, with configurable retention settings.

For individual users, ChatGPT and Perplexity are the most transparent. ChatGPT lets you toggle training off with a single switch. Perplexity doesn’t train on user data at all, and deleting history takes one click.

Claude and consumer Copilot land in the middle. Claude defaults to private unless you opt to share feedback. Copilot limits training to certain contexts, but customization is locked behind preset settings.

Then there’s Gemini. Workspace chats are private, but prompts in the public-facing app can be stored for up to 18 months. Some settings are adjustable, but they’re buried deep in your Google account. 

This handy chart and scoring table comes from Chase Ballard over at Section. It’s super useful for understanding how different platforms handle data control.

If you want to see his full breakdown head over to Section.