For business use, in my opinion after years of use and tens of thousands of experiments, prompts and ‘chats’, there isn’t one AI tool that’s best at everything. Gemini, Perplexity, ChatGPT and Claude each have different strengths, from research and source checking to brainstorming, content development and writing refinement. Using more than one can make sense when AI is part of your everyday workflow, especially if you want separate tools for research, thinking, writing and final checks.
Here’s why I use ChatGPT, Claude, Gemini and Perplexity for different parts of research, brainstorming, writing, checking and business content.
As someone who spends their days building things (clever online stuff, websites that do things instead of collecting dust as static websites, etc) and creating loads of content, I get asked fairly regularly which AI tool I’d recommend.
ChatGPT? Gemini? Claude? Perplexity?
The slightly annoying answer is that I use all four.
And, yes, I pay for all four.
I’m sure that sounds excessive if you’ve only ever opened the freebie ChatGPT occasionally to write a facebook post, an email or ask it a question. But I don’t really see them as four versions of the same thing anymore.
They’ve become different tools in the toolbox.
On a normal working day I might use two or three of them on the same project. Sometimes I’ll use all four.
I’d happily have fewer subscriptions, but I keep paying for these because they’re useful for different jobs.
Gemini – research and the Google world
Gemini is one of the tools I’ll often turn to when I’m researching something, particularly when the subject overlaps heavily with Google, search, websites or information Google itself is likely to understand particularly well.
The free version is useful. The paid version gives you substantially more capability and access to stronger models and features.
Like all AI research tools, though, I don’t treat what it tells me as gospel.
AI can be confidently wrong. Hence I use other tools to check someone its output.
That applies to every name in this article.
If I’m going to rely on something in published content, I check it.
Oh, and being able to have a 300 source Notebook attached to each project is well worth the monthly subscription.
Gemini is my go-to for creative images too – eye-candy for a blog post for example. Despite what anyone says about ChatGPT images, I usually find images made in Gemini far better if you know how to prompt it well.
Perplexity – research where I want to see the sources
Perplexity gets more of my research work than anything else. This, for me, has replaced 90% of my use of Google as a search engine.
The main reason is fairly simple: it makes the sources very visible.
If it tells me something, I can usually see where that information came from and go and read the original source myself.
That’s particularly useful when I’m researching something that’ll eventually become published content. Especially something I’m not particularly knowledgeable about.
I wouldn’t describe Perplexity as “unbiased”, because no AI system deserves the ‘unbiased’ label. It can still select weak sources, misunderstand them or reach questionable conclusions.
But it’s very good for source-led research and for quickly getting an overview of a subject before digging deeper.
The paid version also gives access to a selection of models from OpenAI, Anthropic and Google.
That doesn’t replace having the full versions of ChatGPT, Claude or Gemini, but it can be useful.
ChatGPT – where most of the thinking happens
ChatGPT is probably the tool I spend the most time actually ‘talking’ to.
And I do mean talking.
I don’t normally sit down and type:
“Write me a blog post about XYZ.”
I’ll dictate thoughts. Challenge ideas. Ask questions. Throw in another angle. Disagree with something it suggests. Go off on a tangent. Come back again.
A conversation might run to several thousands of words before anything resembling an article appears.
That’s where I think ChatGPT is particularly strong.
Brainstorming. Exploring an idea. Finding gaps in an argument. Organising a pile of thoughts that started life as a fairly chaotic voice note.
For me, it’s the best overall collaborative thinking tool of the four.
Again, a paid plan makes a substantial difference if you use it heavily.
Claude – where I often take the writing afterwards
Claude tends to come later in my process – unless I’m asking for help with coding but that’s another subject altogether.
Once I know what I’m trying to say, I’ll often use it to refine writing, check whether instructions have been followed properly and look at the piece with a slightly different pair of artificial eyes.
In my own experience, Claude is particularly good at following detailed writing instructions.
That’s useful because I’m quite fussy. I do go through about a dozen or more prompts for every bit of content I write, after-all. Yes, very fussy and I’m not in a rush to publish something that an LLM spits out.
I don’t want generic corporate waffle (if I’d allowed an LLM to write that line it would have said ‘prose’ instead of ‘waffle’).
I don’t want everything turned into neat little marketing slogans or the same sentence construction repeated six times.
And I certainly don’t want something that reads as though a machine swallowed LinkedIn and then vomited it back onto the page. Yuk!
Claude tends to be very good at that final refinement work. After thousands of uses it has become quite familiar with how I work and how picky I am.
That’s what I’ve found from using it. Someone else might have a completely different experience, and these models change too quickly for any league table to stay accurate for long.
However…
The tool matters less than what you give it
You can subscribe to the most expensive AI system on the market and still produce complete rubbish.
If it knows virtually nothing about you or your business, it has to fill in the gaps which means it will guess and churn out rubbish.
When an AI fills in gaps, you tend to get generic output.
Imagine asking somebody you met five minutes ago to write your website.
They know almost nothing about your history, your customers, why people choose you, what those customers worry about, who you’re competing with, how you normally speak or what you stand for.
Yet people give an AI one sentence of context and expect it to somehow know all that.
Then they complain that AI content sounds generic.
Well, of course it does. It looks naff as a social media post, blog post or ‘another AI generated graphic’.
Context beats clever prompting
There’s an entire industry built around supposedly magical AI prompts. Most libraries of prompts are a complete waste of time.
Some prompting techniques are useful.
But I’d take a well-informed AI with a fairly ordinary instruction over an uninformed AI with a 700-word “ultimate prompt” any day of the week.
- Give it background.
- Give it examples.
- Explain the business.
- Explain the audience.
- Explain what people are usually looking for when they find you.
- Tell it what you’re trying to achieve.
- Tell it what you hate about typical industry content.
- Tell it when it gets something wrong.
- Correct it.
- Challenge it.
Treat the process as a conversation. Don’t just press a button marked GENERATE CONTENT and expect something brilliant to fall out. It won’t.
I also don’t let one AI check its own work
There’s another reason I use several systems.
I don’t like having one AI help develop a piece of content and then ask the same AI whether the finished piece is accurate, balanced and well written.
It already has a relationship with the work. So, no, I’m not going to ask the one tool to ‘mark its own homework’.
It helped build the argument.
It knows why various decisions were made.
I want another system that wasn’t involved in the conversation to look at it cold.
That might mean researching with Perplexity, developing the thinking in ChatGPT, checking something against Gemini and then giving the writing to Claude for another pass.
Not every job needs all that. But some jobs do.
Some jobs need one tool. Some need two. Occasionally something important gets the full tour.
And I’m not exactly done yet. I do the final edits when I’m using the output and there’s another hour’s work but it’s essential. Well, I believe so anyway.
You don’t need four LLM/AI subscriptions
I should probably make that clear.
Most businesses absolutely don’t need to subscribe to four AI services.
If you’re just starting, pick one. Learn how to use it properly. Give it good information about your business. Learn what it’s good at and, more importantly, learn where it goes wrong.
The free versions of these tools are astonishingly capable compared with what was available only a few years ago. Remember the hype of 2022 and 2023? They’ve come on a lot since then, even the freebies.
But if you start using AI heavily in your work, or you want more features and power, you may eventually discover what I did.
I’ve ended up preferring different tools for different parts of the job as I’ve explained here.
- Perplexity often gets the research,
- Gemini gets different research projects and when I’m using up to 300 hand selected sources, and particularly when I want another view or I’m dealing with something that sits heavily in the Google world.
- ChatGPT gets much of the random ideas, ‘head all over the place’ thinking and brainstorming (with context, remember), and
- Claude often comes in when I’m refining some writing – my own, a client’s or the result of what came from Gemini, ChatGPT and Perplexity.
Writing a Facebook post hardly requires four AI subscriptions. They earn their keep because I use them for different parts of the work.
Yes, they can all still get things wrong!
Can you trust AI? No. Not entirely. They lack the ability to say “I’m not really sure about that, mate but here’s some bits I did find and a few links you might want to check for yourself”.
They are programmed to be helpful, to keep you using them, to seem like an incredibly confident assistant. Sometimes they can create more confusion than they solve, sometimes they can give you a load of outdated, irrelevant and hallucinated ********.
Oh, and those AI summaries at the top of search results?
Yes, those summaries that now appear at the top of Google or Bing after you search for something?
I tend to ignore them.
There’s always a chance the AI has misunderstood something, pulled together weak sources or simply filled in a gap with a confident guess. There are stats about the guess-rate of AI summaries and I can’t work with guesses (hallucinations).
That’s my own personal view, and I’m sticking to it.
Anyway, I hope this post helped.
