Get Your Money's Worth from ChatGPT (2026 Guide)

ChatGPT used to be just a huge artificial brain you had discussions with. Now it is packed with features most subscribers never touch, which means most people are wasting a chunk of the twenty dollars they spend every month. In this walkthrough I go feature by feature with very precise use cases: Deep Research for jobs and leads, Agent Mode, Canvas, Projects, Custom GPTs, and the privacy settings you should change right now.

What you need

• A paid ChatGPT subscription
• A concrete problem to throw at each feature: a purchase, a prospect list, a dataset

1. Web browsing and image generation (0:36)

Web browsing is so well known now that you do not even feel it anymore, so I only mention it. Image generation deserves honesty: ChatGPT's is not very good. The best model at the time of the video is Nano Banana Pro, which belongs to Google and is hosted on Gemini. ChatGPT's version is fun for dramatic images or turning your dog into a human, but it tends to modify the subjects quite a bit, and for business use cases it is still not quite there.

2. Deep Research: the most underused feature (2:08)

Deep Research is one of the most underused features of ChatGPT and one of the most powerful. You get about thirty to sixty deep researches every month, so use them well because they are precious. Here are the use cases I keep in a dedicated folder:

• Job opportunities (2:32). A friend wanted to give piano classes but did not know how to find students. I ran Deep Research reports and found her a list of 30 places looking for piano teachers. Monetizing a hobby is not that hard anymore.
• Business leads (3:17). I asked for the names and LinkedIn profiles of CEOs of consulting firms based in Paris with at most 100 people. It delivers, and it can even hunt down the LinkedIn profile when it is missing.
• Researching one person (4:13). I ran Deep Research on myself and got my full studies and professional career, which makes finding hooks for a first conversation very easy.
• Market analysis (4:55). It finds market research for any industry and presents the trends to you.
• Personalized tutorials (5:12). I generated a step-by-step tutorial for solving a Rubik's Cube, and even re-taught myself differential equations. The sources sit at the bottom of every report so you can verify them.
• Similar songs (6:07). Give it a song you love and ask for others with similar lyrics. Several of them ended up in my playlist.

One tip: sometimes ChatGPT will say "I am not allowed to do this" when researching people. If you say it is only for professional matters, it will allow you to.

3. Shopping research (6:55)

For purchases, skip Deep Research and use the dedicated shopping research. It finds the best deals for any product, and it can hunt for discount coupons:

Could you find a promotional code for this item on this website?

4. Agent Mode (7:29)

Agent Mode gives ChatGPT its own computer and its own internet browser. My test prompt:

I'm going to Copenhagen next month for four days and want to eat
as many good bakeries as I can.
Can you research what are the best bakeries?

Unlike Deep Research, it actually browses: it opens Google Maps on a computer and reports back. It is still kind of slow and there are not a lot of business use cases yet, but it is great when the task needs clicking through a catalog or a map. Cannot decide what to buy on Amazon? Hand it the research and get a full report back.

5. Study and Learn (8:41)

Study and Learn is a hidden mode where ChatGPT asks you questions about a subject to make sure you really understand the core before moving forward. As for the web search toggle, you rarely need to click it anymore; ChatGPT knows when to search, though clicking it does no harm if you feel safer.

6. Canvas: deterministic data work (9:26)

Canvas lets ChatGPT write code, and you should care even if you are not a developer. A lot of daily work is deterministic: action A should always produce output A. LLMs are probabilistic; they behave like humans and change their answers. Code only cares about input and output, and you can rely on it one hundred percent. So for data work, activate Canvas and say:

I will give you a database that I want you to analyze and clean.
Here is what cleaning looks like for me.

It writes a script that removes duplicates and rows with no values, and every run gives the same outcome, because it is not ChatGPT doing the work, it is the code. Canvas can also create webpages (11:13): I had it build a landing page for a flower boutique. Not the greatest UI design, but good enough for quick prototyping.

7. Third-party integrations (12:08)

ChatGPT now talks to outside applications like Spotify. Be aware the integrations are limited: I cannot pause and play, for example. And ask yourself whether it is actually smart to route an action through ChatGPT when clicking one button in Spotify would do. Automate the things that make sense automating.

8. Projects (13:15)

A project is a folder. Any conversation you open inside it stays there, and, crucially, you can give it files: text, images, PDFs, code. Every conversation in the folder has those files in memory, so you stop copy-pasting the same file over and over. You can even add an instruction manual to the files so ChatGPT behaves the same way every time.

9. Custom GPTs (14:24)

Projects stay internal; Custom GPTs face the world. They are GPTs specialized in one thing only, like customer support running on your documentation. To build one, go to Explore GPTs, click Create, and just have a conversation about what you want. The Configure tab is for fine-tuning behavior, adding company files, and toggling capabilities like web search, Canvas, image generation, and code interpretation. My Genial Agency Virtual Assistant was built from one small company file plus a conversation; I wrote maybe one or two lines of its instructions myself.

Do not underestimate the lead generation angle (17:56): every custom GPT shows who created it. The girl who popularized Ghibli-style image conversion did it with a custom GPT that reached over 100,000 followers within a few days; even 1% of that reaching out is 1,000 people. I also know lawyers whose custom GPTs bring them 400 to 500 users every month.

10. Privacy settings (18:57)

The most important setting in the whole app:

Profile > Settings > Data controls > Improve the model for everyone > Off

They say they are not training on your data, and we can believe them, but better safe than sorry.

11. Memories, personalization, and schedules (20:08)

You can personalize how ChatGPT communicates, and the Memories feature collects information about you across chats. Key distinction: remembering things about you does not mean the model was trained on your data. It is the model reading the notebook of your conversations, not knowledge inside its brain. You can also schedule recurring tasks, useful if your budget is tight and ChatGPT is your only tool.

Pitfalls and tips

• Deep Research quota is limited to roughly thirty to sixty runs a month. Spend them on research, not shopping; shopping has its own mode.
• Always scroll to the sources at the bottom of a Deep Research report before trusting it.
• Image generation is the weak spot; go to Nano Banana Pro on Gemini for business-grade visuals.

Where to go next

See how the competition compares in Get Your Money's Worth from Gemini and Get Your Money's Worth from Claude.