The four AI assistants that matter when people buy

The four AI assistants most worth testing when people buy are ChatGPT, Claude, Gemini and Perplexity. They are among the most widely used for research and recommendation questions, and each finds and weighs sources differently, so the same question can produce a different shortlist on each. Measuring all four gives a fairer picture than relying on any one.

This is a judgement about where to begin, not a claim that no other assistant matters. Your own customers may favour a different tool, and you should add it if so.

Why these four?

They cover the main ways buyers meet AI answers today:

• ChatGPT is a general-purpose assistant used by a very wide audience for everything from writing to product research.
• Claude is a general-purpose assistant from Anthropic, used for research, writing and analysis.
• Gemini is Google's assistant, connected to Google Search for grounding, and present across Google's products.
• Perplexity is built as an answer engine, with search and citations at its core.

Together they span general chat, a search-company assistant and a search-first answer engine. That range is the reason to test across them.

How do they differ in practice?

Each assistant's documentation describes its own approach, and you can observe much of it directly.

Treat the table as a map of tendencies. Exact behaviour depends on the product, the mode and the settings, so test the version your customers use.

The differences have consequences. An assistant that grounds in Google Search may lean on pages that do well there. One that uses a different search system may surface a different set of pages. That is why the same brand can be recommended by one assistant and absent from another.

What do you learn by testing all four?

Three things.

Where you are strong. If you appear consistently on one assistant and rarely on another, the sources behind each are different, and you can find out which.

Which sources recur. Pages that show up across several assistants are the ones shaping the category. They are your highest-value targets.

Whether you have a general problem or a specific one. Absent everywhere suggests a foundation issue: unclear pages, little third-party presence. Absent on one suggests an access or sources issue particular to that assistant.

How to compare answers across ChatGPT, Claude, Gemini and Perplexity shows how to lay the results side by side.

Should you treat them equally?

For measurement, treating them equally is a sensible default, because you cannot know in advance which one a given buyer will use. Weighting by audience is reasonable if you have evidence, for example from referral data or customer surveys, about which assistants your customers prefer.

Whatever you choose, state it. A combined figure that quietly favours one assistant is hard to interpret. Citewake's method asks the same questions of all four, with live search on, and shows results per assistant as well as overall.

How do you decide which to prioritise?

Use evidence about your own buyers. Check your analytics for visits that come from assistants, where available. Ask new customers which tools they used during research. Listen to sales calls for mentions of "I asked ChatGPT" or similar phrases. These clues are imperfect but better than guessing.

If you have no evidence, test all four equally and let the results show where you are weakest. The weakest assistant is not always the one to fix first, though. If one assistant is plainly more used by your buyers, a gap there costs more.

Whatever you decide, keep the testing method identical across assistants. Different questions, different settings or different counting rules make comparisons meaningless. The point of testing four is to learn from the differences, and that only works if everything except the assistant is held constant.

Does the list stay fixed?

Not necessarily. Products change, usage shifts and new assistants appear. Review the list from time to time against what your customers say they use, and keep the test method steady so that older results stay comparable when you add a new assistant.

What else might matter?

Other assistants and AI features exist and are used. Search engines add AI-written answers to results pages. Productivity tools embed assistants. Voice assistants answer questions aloud. They may matter to your audience, and they can behave differently from the four above.

The practical approach is to start with the four, learn the habits of testing, and extend to others that your customers use. The habits transfer: clean sessions, repeated asks, kept answers. See how to test AI answers properly.

How do you begin?

Pick ten questions your customers ask. Leave your brand name out. Ask each of the four, more than once, and note who is named and which sources are cited. A quick way to see a first picture is the Citewake checker. For more on what ChatGPT and Perplexity do under the hood, read how ChatGPT decides which brands to recommend and how Perplexity chooses the sources it cites.

What is the short version?

Start with ChatGPT, Claude, Gemini and Perplexity because they are widely used for research and differ in how they search. Test them identically, compare the results, and add other assistants if your buyers use them. The differences between assistants are not an obstacle. They are where much of the insight comes from.