Choosing sides seems to be a common response to any multiple-choice offering. People have been this way for a very long time; car makers have capitalized on brand preference since 1915 when Chevy’s Model 490 took on Ford’s Model T. Decades later, in the mid-aughts, Apple played on brand loyalty with their Mac vs. PC ads.

There are differences between Apple and PC, Chevy and Ford, but the utility is the same. The purpose of a car doesn’t change with the brand. The purpose of a desktop computer doesn’t change if you have one mouse button or two. Computer or auto, you could buy either brand to accomplish your core need. But when preference becomes fandom, group loyalty isn’t too far behind.

Take a moment and listen to how people talk about AI chatbots, and very quickly, it’s easy to see that groupthink behavior is alive and well in consumer technology. Gather three or more consumers who use AI chatbots, and very soon, the conversation devolves into some variation of “my chatbot is better than yours.” Pretty harmless stuff on the surface, but AI chatbots aren’t cars or computers. There are some very real differences under the hood that make dedication to one brand of bot limiting at the very least, and risky at the extreme.

Different Brands, Different Behaviors

I’m going to limit this discussion to consumer-facing chatbots from the most used brands:

• ChatGPT
• Gemini
• Microsoft’s Copilot (Web)
• Claude
• Perplexity

Perplexity is different than the other chatbots, and we will dig into it later, so hold that thought. The other four chatbots are based on large language models (LLM). Each LLM was trained on a massive database accumulated by crawling and scraping the public internet and proprietary data provided by the chatbot’s owner.

Based solely on LLM + datasets, the chatbots are very much alike. Way back in the beginning (c. 2022), your choice between chatbots was much simpler, a more apples-to-apples comparison. You might have chosen one chatbot over another because you preferred how it delivered its responses: friendly vs. technical, detailed vs. summaries.

Having Favorites Can Bite You

A lot has changed in the AI chatbot world since 2022. For this article, I’m only going to cite one added feature, known as retrieval augmented generation (RAG):

• Pre-RAG, a chatbot could only generate a response based on its database of scraped internet content, any additional information the brand owner uploaded, and data you supplied as an upload (all static content).
• With RAG, chatbots can reach out to the open web and search for fresh, relevant content. The chatbot could Retrieve live web information to Augment the answer it Generated.

ChatGPT, Claude, Gemini, and Copilot all use RAG, but each does so differently. This is where brand loyalty can spell trouble.

Gemini is RAG-integrated, which makes sense coming from a company founded on internet search. But Gemini isn’t just searching; it’s using decades’ worth of Google’s knowledge graph AND search. This is why Gemini answers can look structured. They are often pulled from a database and not website prose.

ChatGPT is RAG-capable, currently based on Bing’s knowledge graph, which is not as extensive as Google’s. ChatGPT uses RAG to summarize the top links from internet searches, when it uses RAG! ChatGPT favors using its database over the open web, which also means there may be a ‘freshness date’ to its answers.

Claude is RAG-emerging. This tool was first developed to apply AI in processing large data sets that the user uploads. Claude has scraped the internet to create a database to train its LLM tool, but it applies that training to summarize your data. Claude can and will search the internet.

Copilot (Web) and Perplexity are RAG-first, these two tools search first and then use their LLM models to synthesize answers from the top search results.

• Copilot (365) is a paid version like Claude, but tied directly into your Microsoft subscription products (Office, SharePoint, Teams). Unlike Claude, you usually don’t need to upload data for Copilot (365) since it already exists as a file within a Microsoft product.

Differences Encourage Specializations

As I mentioned, RAG is only one of many features added to AI chatbots as they have evolved. And as these features have multiplied, so have the differences between the tools. And this is where it can be dangerous to always favor one over another. In a way, these tools exist as separate options because they are distinct from one another.

ChatGPT is primarily a chatbot, with its core strength in its conversational abilities. It can function as a powerful answer engine, though its answers can be dated. Because of its language prowess, ChatGPT is a fantastic creative partner.

Gemini is your research and knowledge assistant, able to easily access live and static knowledge. Gemini excels at comparisons of online content or data. Where ChatGPT has focused on language, Gemini adds imagery. Gemini’s link to Google Veo makes it a powerful visual creator.

Claude is the number-cruncher of the group. Designed with a massive context window (meaning you can upload LOTS of data), Claude is the go-to choice for dataset analysis. While Claude can use RAG to capture live data from the web, its strong suit is coding and summarizing large data sets.

Perplexity and Copilot (Web) summarize live search, so they could be said to be RAG-forward. These are best used as research and citation platforms. One way to think of Perplexity and Copilot (Web) is as summarization tools that help you find your answers faster than clicking through the top 10 blue links on a search results page.

• Copilot (365) is the preferred AI tool for internal data because it doesn’t upload information to the web. The data you work with in Copilot (365) is governed by the same privacy and sharing policies outlined by your company.
• In short, with the other four tools, if you need to analyze a list of data that contains personally identifiable information, you are sharing that list with a third party, and it may be used to train the AI tool. That same analysis performed with Copilot (365) doesn’t share any information with Microsoft, nor is it exposed outside of your company’s security/firewall.

Risk in the Differences

There are risks that you might get an incorrect answer and risk that you make decisions based on a faulty analysis. The danger may be to your reputation or your job. I’d suggest that for most uses, any of the chatbots can get the job done. But if you need a specific kind of information or output, use the tool that can do the job best.

Look at the variety of tasks you need to do and choose the right AI assistant where it makes sense. For our digital ad team, that might mean:

• Using ChatGPT to brainstorm ad copy or landing page content.
• Substantiate new strategic ideas with Gemini by finding references online and summarizing projected outcomes.
• Use Copilot to organize the team’s daily calendars and help prepare for meetings by locating all relevant files, emails, and chats.
• Turn to Claude to understand how Paid and Non-Paid traffic interacts and to find correlations in mountains of data.

You might have warm and fuzzy feelings for ChatGPT, or Claude might be your jam, but don’t let that preference stand between you and the best possible answer. The AI landscape changes fast, and the right tool today might be replaced by a better option tomorrow. So, stay informed, stay curious, and stay flexible. There’s more than one way to slice and dice your options!