ChatGPT Canvas and Claude Artifact: Good Writing Assistants

ChatGPT Canvas or Claude Artifact separates the work from your conversations with the LLM.

For example, if you are researching something or generating content, it will put it in a separate block: 

How ChatGPT Canvas works

This allows you to work with GPT-4o and Claude 3.5 Sonnet on the job3 separate from the conversation. The chatbot will treat the output more objectively than its exchange with you, enabling more targeted and collaborative editing.

In this chapter, let’s understand the capabilities of the ChatGPT Canvas and Claude Artifact to recognize where they can help content writers.

What can you do with ChatGPT Canvas and Claude Artifact

Pooling all of the research in one place.

Recently, I was looking into how Liam Lawson performed in the 2023 F1 season after Daniel Ricciardo broke his wrist in the Dutch Grand Prix. GPT-4o went through various articles on the internet and gave me a brief rundown: 

ChatGPT Canvas is great for research

You can’t do the same with Artifact as Claude 3.5 doesn’t have internet access. However, you can still use this capability to bring information from various topics on different depths into one place for consumption or analysis.

Another thing you can do with these two features is to edit specific portions of what GPT generated without much prompting effort. The fact that you can edit in the Canvas or Artifact directly enhances the overall process further.

As a content writer, I constantly use it to rewrite particular bullets in a list or recreate specific sections in an outline. In regular use, you can use Canvas or Artifact to craft emails or messages at work. After the LLM produces the first draft, you can make the finer changes yourself before sending it.

Additionally, you can do multi-step patterned work. It’s tasks like generating five outlines from five titles, one at a time, in a particular sequence. As GPT-4o or Claude 3.5 treat the job objectively, the quality of output won’t suck even in longer chats.

You will be scrolling less than before and focus on collaborating with the machine to produce useful outputs.

Applications in content creation

A lot of the above capabilities translate well into the content production workflow. You can run multiple steps at once. For instance, you can generate title ideas and convert them to outlines in a single chat.

No more scrolling up and down to fetch the titles to create the outlines. No need for an intermediate scratchpad or notepad to edit the titles before the outline is created. Everything can be done in a single place.

However, there are certain limitations.

I’ve discovered, in the case of ChatGPT, Canvas doesn’t use custom GPTs. It does in the first reply but fails to in the subsequent questions. This can be annoying as you have to do the extra prompting manually, especially in professional work.

Consequently, the hallucination problem gets worse. Long prompts are more likely to get misunderstood which yields trash outputs. The small 128K tokens worth of context window doesn’t help either.

My recommendation is to just use it for research and ideation. Anything beyond that requires you to manually make tweaks to the LLM-generated content on a text editor. Yes, it’s possible on the Canvas or Artifact itself, but you can’t add anything instructional here.

For instance, while converting an outline to a draft with an AI-powered chatbot, I include commands like “give a simple example here” or “write in 150 words” for each section. You can’t do that within the Canvas or Artifact, as you are supposed to prompt directly to GPT.

In my day-to-day content production needs, the ChatGPT Canvas is a phenomenal research assistant. It can collate information from multiple sources in a desired format and makes it easier for me to prune out irrelevant stuff. Plus, I don’t have to scroll a lot!

Conclusion: Personal notetaker

For most professionals that leverage LLMs to create content (not code) the Canvas and Artifact are efficient information aggregators. They can pull in information, enable collaborative editing, and simplify targeted editing.

At the same time, complex tasks like generating drafts are done better in the regular GPT-4o internet-enabled and GPT-enabled chats.

—

Next Chapter: The Ideal LLM for Content Writing

Previous Chapter: ChatGPT o1-Preview: Great for Content Planning

Index (with Prologue): Content Writing With AI