Since the release of ChatGPT 3.5 in November 2022, the interface has changed a lot. We moved from a very raw interface, able to generate content from a prompt (which was even suggested to us):
To a much more complete ecosystem, packed with features to boost the power of the raw models:
With that in mind, if you did not jump on board along the way, you feel a bit lost… and that is normal :)
The goal, then, is to find the ecosystem best suited to most users.
Choosing the right model for your tasks
To answer the following question: which tool should you use? Claude, ChatGPT, Gemini… or another alternative model?
It is hard to answer without knowing your goals and the kinds of tasks you do, and the quality, speed or confidentiality you expect.
But today there are 4 ecosystems to favor for serious use: ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google) and Mistral (France).
They all include a polished interface, a mobile app, a voice mode, multimodal features (image, audio, documents), the ability to run code, and quality models.
ChatGPT (OpenAI) is the most complete to date. Smooth interface, powerful multimodal interactions, good file handling, and tools like canvas or deep research (I talk about it below) that change the way you work. The ecosystem is coherent, well thought out, and evolves very fast (faster than the others).
Claude (Anthropic) wins many people over with the writing quality of its answers and a stronger respect for confidentiality. But it lacks quite a few features (no image generation, for example).
Gemini (Google) is a good mix of the two: very strong at research and cross-referencing information (text, web, …), the ability to add videos as input is a big plus, and its context window is huge. The integration into the Google ecosystem is also interesting.
Mistral (France) bets on open models (open-weights). I add it mainly for its open source and French angle. Even if the models perform comparably to those above, it lacks quite a few features..
The $20/month subscription is often essential if you want access to the advanced features. You can start with the free version and subscribe once you feel limited.
What about the others?
There are also alternatives (models or agents specialized in certain tasks). A few pointers:
- Grok (xAI) may appeal to some X users, but personally I do not recommend it: little transparency, sometimes biased generation, and a questionable ethical framework…
- Copilot (Microsoft) is often presented as an alternative, but I advise against it too: too dependent on the Windows environment, and performance well below the other models above.
- DeepSeek (China) offers powerful open-source models. Excellent on paper, but no smooth interface or ready-to-use integrated service.
- Perplexity is the most interesting tool for document research. It is fast, structured, sourced, with an 'answer engine' logic. Very good, but only for the 'finding information on the web' part.
- Manus, a web agent able to automate long tasks (collecting, sorting, email, data extraction). A bit technical to get started with, but impressive when well configured.
A closer look at OpenAI: how I choose my models in ChatGPT
ChatGPT offers several models accessible from the drop-down menu. And here too, you can quickly get lost.
I use this tool daily. Here is my feedback, based on real-world use:
In short, mainly the 4o and o3 models. But also the deep research feature (I talk about it right after).
For data security, whatever your choice, do not forget to disable the use of your data to train the models. Go to the top right in: Profile > Settings > Data controls
Then click on 'Improve the model for everyone' and turn it off:
Deep research mode: super-powerful in-depth research
Deep research is a key AI feature, and I find it is underused by most of us (then again, it is a bit hidden).
Deep Research solutions can produce top-quality reports. It is worth experimenting with these reports in your own field to gauge the potential. And it is also very relevant for more personal cases (like planning a trip, getting a thorough opinion on a topic, …)
As with a classic use of the models, there can be hallucinations, but the answers are far more reliable than a simple answer generated on the fly, often because there are citations like here.
Here, I share a quick diagram of what the models do for a deep research (Linkup can be replaced by ChatGPT, for example):
An accessible use with voice mode
The easiest way to get started is to talk directly to the mobile app. On mobile, Gemini and ChatGPT handle this very well. I open the app, I dictate while walking or driving, and that is enough to test plenty of things.
Note that in voice mode there is no web search, so for factual information it can literally say anything.
The real superpower: showing what you see
The most impressive part is screen or camera sharing, which I use a lot personally, for example:
- I take a photo of a plant in a store to find out what it is and whether it suits the season
- the AI sees the same thing as me and answers right away
Creating images, videos, code and other files
AIs can produce all sorts of content. ChatGPT, Gemini, Mistral can generate images on simple request. Claude cannot do it yet.
Overall, ChatGPT is the most precise: you can really control the size, the style, the details.
I recorded this demo video on generation with OpenAI, in case it gives you ideas. You will see classic cases: creating a marketing visual, illustrating a blog post, etc.
For video, Gemini stands out with veo 3 which you may have seen on social media.
Beyond images, the models can generate, as output, interactive code, charts, interfaces (very powerful on Claude), and documents (PDF, PPT, etc.).
An example of an interface generated on Claude (and which uses Claude via API):
But be careful with document generation, it is a bit of a weak spot. As soon as you ask them to format, number, or insert tables into documents, they hallucinate a lot. In short, I advise against trusting them with a final ready-to-use document. It is better to grab the raw text and lay it out by hand.
My tips
As you can probably see, there is plenty to do. You can stay in chatbot mode for basic use, or reach a super-powerful copilot mode if you brief it properly.
I advise you to follow these steps to get the most out of AI on every request:
1️⃣ Choose a powerful model → By default, the app launches a fast model. Change it manually for more complex tasks.
2️⃣ Bring context → Do not start from a vague question. Give concrete information, files, examples, needs. That is what changes everything.
3️⃣ Use voice mode + screenshot → Show what you see: take a screenshot and drag & drop it into the tool, for example
4️⃣ Explore the Deep Research feature → to generate complete, sourced, actionable reports: product comparison, competitive intelligence…
5️⃣ Keep the conversation bouncing → edit your prompts, test variations, ask follow-up questions. That is where the secret lies.
Points to watch
There are a whole lot of common traps when using AI.
First, hallucinations. All models hallucinate, but less and less. Especially if they go on the web, and if you have the citations. Stay alert!
There is no magic, and the models have no memory, nor do they learn continuously. If the AI pulls off the impossible with little context, it is often luck, a bit like an oracle :)
The model enables a conversational logic. You can ask it to question you if it does not have enough context. The real gain comes from usage, from back-and-forth with the tool.
Rather than asking why, verify the information. Either by looking at the chain of thought on reasoning models, or by asking it where it found the information, where it comes from.
Your next step?
You now have the toolbox, all that is left is to open it ;)
To sum up, you can:
- Choose your AI and get the paid version (the real power starts at 20€)
- Take action right away. Advanced model: give it a real challenge you have at work with all the useful files (docs, spreadsheets, charts). Ask for a concrete deliverable, memo, code, diagram, then have it refined as much as possible.Deep Research: launch it on a complete study (such as a competitive analysis)Voice mode in context: I use it to brainstorm, practice a language and solve snags on the fly.
- Explore everything else such as image generation (e.g. marketing visuals) and reading PDF/PowerPoint so it summarizes or reformats. Test these building blocks in your professional projects and personal ones
Most people start out: short question, zero context, default settings. You have the keys to do better: upload your documents, push it to propose 30 ideas, duplicate conversations to compare, alternate text, voice and image. The difference between the average user and a power user is not the magic of a prompt, it is usage.
