How to Get the Best Results from AI Assistants in 2025

This article was updated June 2025

AI assistants aren’t shiny novelties any more. They’re an integral part of how many of us work, think, write and research every single day. And the reality is, they’ve got unbelievably good since they first came about a couple of years ago.

When I first wrote a version of this guide in 2023, it was mainly focused on ChatGPT because that was the AI assistant at the time. Fast-forward to mid-2025, and the market is much broader and a lot more capable. We’ve not just got ChatGPT’s latest models, but also Google’s Gemini, Anthropic’s Claude, and even Apple Intelligence starting to make itself useful on your iPhone or Mac (finally… sorry Apple, but it’s true). 

So, whether you’re dabbling or fully embedded, here’s a refresher on how to get the best out of these tools right now.

The biggest shift? They’re getting better at thinking

The main story across almost every major AI assistant is deeper reasoning. It's much less “here’s a random answer I scraped together”, although, to be fair, that was quite entertaining at the time and part of the charm of using AI.

Thankfully, ChatGPT still has this little gem when you ask how many Rs are in strawberry. It’s due to the fact that ChatGPT is converting your question into a chunk of numbers rather than looking at the letters, but that’s just not as funny, is it?

• ChatGPT’s latest models (o3 / GPT- 4o) now handle much longer conversations, juggle multiple files, share screens, and can work on more complex reasoning tasks. You can select the different modes you want it to go into before submitting your query. If you’ve never used the deep research function you should give it a try, it pulls information from all across the web AND gives you the references.
  
  
  One of ChatGPT's best features (and no, I’m not sponsored by OpenAI, I just genuinely rate this) is Custom Instructions. This handy bit lets you tell ChatGPT a bit about yourself, like what you do, what you’re working on, or even just what you’d like it to call you. 
  
  You can also shape how it replies, whether that’s keeping things casual, giving you step-by-step advice, or taking on a particular role. Once you’ve set it up, it remembers all this every time you start a new chat. And if you change your mind, you can update or reset it anytime.
  
  
  
  
• Google’s Gemini (2.5) is increasingly focused on live multimodal input. It can look at images, video, and live screenshares while you talk. It’s rolling out something called "Deep Think", much like ChatGPT’s “Think for longer” which helps it pause and reflect before answering harder questions. You can also now run longer “agent-style” workflows, where Gemini takes over more steps of a task for you.
  
  
• Claude (Anthropic’s model) is quietly making a name for itself as the go-to for serious, sustained reasoning. Its latest model, Claude 4 Opus, can handle extended multi-hour coding sessions or longer writing tasks, all while maintaining accuracy. It’s also one of the safest assistants on the market, with tighter controls around hallucinations and fact-checking.
  
  
• Apple Intelligence Siri on iOS 18 and macOS is focusing on practical, privacy-first personal assistance. It uses GPT-4o for writing, rewriting, and summarising tasks, and adds fun extras like custom Genmoji and improved visual search. Let’s be honest... Siri has always lagged behind fellow smart assistants like Bixby, Cortana (RIP) and Alexa, let alone the likes of ChatGPT, Gemini and Claude, so any upgrade is very welcome!

The bottom line is, you don’t have to cross your fingers and hope for a half-decent reply any more. If you give them good instructions, these tools are now capable of following through with surprising depth.

How you ask matters still matters

If there’s one thing that hasn’t changed, it's that vague prompts still lead to vague answers.

Surreal had billboards back in 2023 that were “Written by AI”, and it illustrates our point pretty well:

[Source: Surreal]

Now, this isn’t exactly what would happen. If you were to ask ChatGPT that same question, it would say something like "Start Your Day Right with our Irresistible and Nutritious Cereal – The Perfect Breakfast Fuel!". But it's a perfect example of what can happen when you don't put effort into your query.

Instead of saying a query like:
“Can you help me plan my holiday?”

Try saying something more detailed:
“Can you help me plan a seven-day itinerary for a girl's trip to Italy, visiting Rome, Florence, and Venice. Include tourist attractions, local food recommendations, and ideas for group activities.”

The more you feed AI with things like audience, tone, length, and purpose, the better the output.

Breaking work into steps beats throwing everything in at once

Another big learning that holds true across all the AI models, it's much better to work in stages.

• Start with an outline.
  
  
• Flesh out sections one by one.
  
  
• Then use the AI to edit, polish, or rephrase once you’ve got a draft.

Most assistants are far better at iterative work than trying to generate a full finished product first time round. And with newer tools like ChatGPT’s Projects or Gemini’s Canvas, you can often keep everything in one place as you work through different parts of a task.

Quick guide: who’s good for what?

Here’s a very simple, very honest cheat sheet for mid-2025:

And finally, no, they won’t do everything for you

Despite all the progress, these tools still aren’t magic, so don't be lazy! The world still needs great thinkers. They don’t understand nuance or strategy the way a human expert does. They won’t replace careful thinking or good judgment. And they absolutely still need you to sanity-check their output, especially if you’re publishing content, or making decisions based on their responses.

But what they are becoming is something far more useful to help you streamline your workflow. Whether you’re planning a campaign, writing a report, brainstorming creative ideas, or digging into data.

The key is learning how to work with them, not expecting them to work instead of you. 

This article was updated June 2025