From the book

What replaced ChatGPT plugins, and how do you do the same jobs now?

· By Robert Scales · 4 min read

Short answer

GPTs replaced plugins. Custom versions of ChatGPT can browse the web, analyze data, make images and, through "actions," call outside services the way plugins did. OpenAI stopped new plugin chats on March 19 and ended existing ones on April 9, so the jobs plugins did now run through GPTs from the GPT Store or ones you build yourself.

A Beginner's Guide to ChatGPT closed with a chapter on plugins, the add-ons that let ChatGPT reach outside itself to book a table, plan travel, run a calculation or connect to other apps. That chapter described real tools, and the ideas still hold. The delivery has changed: OpenAI has wound down the plugins beta. Since March 19 nobody can install a plugin or start a new plugin conversation, and since April 9 old plugin chats can be read but not continued.

The replacement is GPTs. OpenAI says it built them from plugin users' feedback and that they now match everything plugins could do, with more on top. The plugin store held a little more than 1,000 plugins; OpenAI says the GPT Store holds hundreds of thousands of GPTs.

Plugins and GPTs: what changed

A plugin was a connector you switched on inside a normal chat. A GPT is a custom version of ChatGPT with its own instructions, optional reference files, built-in skills such as web browsing, data analysis and image creation, and optional actions. An action connects a GPT to an outside service through that service's API (the doorway one program uses to talk to another). OpenAI designed actions from what it learned in the plugins beta, and developers can reuse an existing plugin's manifest file to define them.

Two practical differences matter to everyday users. First, you pick a GPT for a task rather than toggling plugins in a chat. Second, GPTs launched for paying plans first, starting with ChatGPT Plus and Enterprise, so check what your plan includes.

The book's plugin examples, mapped

Book examplePlugin used thenHow to do the job now
Booking dinner at a restaurantOpenTableSearch the GPT Store for the company's name; confirm the builder is the company
Planning flights and hotelsExpediaSame: search by name, or use a browsing-enabled GPT for research and book on the official site
Automating workflows across appsZapierZapier's AI Actions GPT, one of OpenAI's launch examples
Calculations and dataWolframThe built-in data analysis skill for numbers and charts, or a math GPT from the store

The method from the book doesn't change: give a role, give context, state the task and format, iterate, and verify. A GPT simply lets you save the first parts so you don't retype them.

A worked replacement: your own restaurant-shortlist GPT

The book's restaurant example asked ChatGPT to find and book a table. Here is a version you can build today without code, using browsing for research while you make the booking yourself.

  1. In ChatGPT, open Explore GPTs, then Create. You can describe what you want in conversation, or use the Configure tab to fill fields directly.
  2. Name it "Dinner Shortlist."
  3. Paste instructions that follow the book's method:
Prompt
Role: You are a careful local dining researcher.
Context: The user gives a city or neighborhood, a date, party size, budget per person and any dietary needs.
Task: Search the web and propose three restaurants that fit. For each, check the restaurant's own website or a reliable listing for opening hours on that day.
Format: A table with name, cuisine, price range, why it fits, opening hours and a link to the official site. Below it, one line on anything you could not confirm.
Rules: Never claim a table is available or booked. Tell the user to book directly with the restaurant.
  1. Under Capabilities, turn on web browsing. Leave image creation off; it isn't needed.
  2. Test it with a real request, read the sources it links, and adjust the instructions if it skips the hours check.
  3. Save it as private, or share it by link.

A developer can go further and add an action that calls a booking service's API, but that needs the service's permission and an API key, which is a job for its makers or your developer.

Try it: Think of one plugin task you used to do, or one from the book. Build a private GPT for it with role, context, task, format and rules in its instructions, then test it three times with different requests and note where it needed correction.

Which GPTs cover what the old plugins did?

Start with OpenAI's own suggestion: search the GPT Store for the name of the plugin you used. Many companies that made plugins also publish GPTs, and Zapier and Canva were among the first examples OpenAI showed. When you find one, check who built it. The store shows the builder's name or website, and a GPT published by the company itself is safer than a look-alike made by someone else.

For general jobs, you often don't need a third-party GPT at all. Web research, file and data analysis, and image creation are built-in skills, so a GPT you configure yourself, or ChatGPT on a paid plan, can cover what the browsing and code interpreter plugins once did.

Keep the book's last step whenever a GPT touches the real world: verify. Prices, availability and opening hours change, and a GPT can be wrong in a confident tone. When a GPT uses an outside service, OpenAI says you choose whether your data can be sent to it, so read what will be shared before you allow it.

Sources

  1. Winding down the ChatGPT plugins beta — OpenAI Help Center, March 19, 2024
  2. Introducing GPTs — OpenAI, November 6, 2023
  3. A Beginner's Guide to ChatGPT — Holistic Hybrid, April 18, 2023

Get the free Guide to Artificial Intelligence

Sign up for the weekly newsletter and receive the free guide: what changed in AI this week, one lesson, one prompt to try. Unsubscribe in one click.

How we handle your address

Back to top

Display

Your choices are saved in this browser only. No data is sent.

Text size
100%
Colours
Reading
More settings
Accessibility statement