How do ChatGPT, Bing Chat and Bard compare?

For the first time, three major chatbots are open to the public at once. ChatGPT came first, in late 2022. Microsoft added a chat mode to Bing on February 7. And on March 21, Google began letting people in the United States and the United Kingdom into Bard through a waitlist. They look alike: a text box, a conversation, an AI that answers in full sentences. Underneath, they are built for different jobs.

How each one works

ChatGPT comes from OpenAI. Subscribers to ChatGPT Plus can use GPT-4, OpenAI's newest model, within a usage cap; the free version does not include it. Both versions write from patterns learned during training. OpenAI says GPT-4 generally lacks knowledge of events after September 2021, and the model does not search the web while you chat. Its strength is fluent, flexible writing and explanation.

Bing Chat is part of Microsoft's search engine. It uses an OpenAI model wrapped in Microsoft's own technology, which the company calls Prometheus, so it can pull in current search results before answering. That means it can handle questions about recent events, and it lists the pages it used in a "learn more" section at the end of each answer. You need to join a waitlist for full access.

Bard runs on a lightweight, optimized version of LaMDA, a large language model Google has been researching for some time. Google describes it as an early experiment and a complement to Search rather than a replacement. Bard often shows a few different drafts of its answer so you can pick a starting point, and a "Google it" button opens Search in a new tab to check the response. Google limits the number of exchanges in one conversation to help keep chats on topic, and it says coding help, more languages and image features will come later.

All three make mistakes. Google's own announcement shows Bard getting the scientific name of the ZZ plant wrong. OpenAI says GPT-4 still invents facts and makes reasoning errors, though less often than earlier models.

A side-by-side test on three tasks

The fairest comparison is to send the same three prompts to each chatbot, each in a new conversation:

1. Facts: What is the boiling point of water at the top of Mount Everest, and why is it different from sea level?
2. Writing: Write a friendly 100-word email inviting my neighbors to a street cleanup on Saturday morning.
3. Current events: What are the three biggest technology news stories from the past week?

Then score each answer. The table below shows what to look for, based on how each tool is designed. It is illustrative, not a set of real test results:

For the facts question, check the boiling point in a trusted reference: a chatbot can explain the right idea and still give a wrong number. For the email, judge tone and length against your request. For the news question, open the sources and confirm that the stories really are from the past week.

Which one should you use for which task?

Use Bing Chat when the answer depends on current information or when you need to show where a fact came from: news, prices, recent product releases, or anything you plan to quote. Open the links before you rely on the answer.

Use ChatGPT for writing and thinking tasks where the facts come from you or do not change: drafting emails, rewriting a paragraph in plainer words, brainstorming, outlining, or explaining a stable concept. If you have Plus, GPT-4 is worth trying for harder reasoning and longer, more careful instructions.

Use Bard if you are in the US or the UK and want to see alternatives side by side, or if you already do most of your checking in Google Search. Treat it as Google intends: an experiment to try, with the "Google it" button close at hand.

For anything important, use two of them. When they agree, check once in a trusted source. When they disagree, you have found exactly the claim you need to verify.