OpenAI
ChatGPT
The broad generalist, with a persistent preferences layer most people never turn on.
AI chats
Pick the assistant you use. Get better at it.
Every AI chat is shaped differently — what it can see, how it should be prompted, where it breaks. Choose the one you already have, and read how to get more out of it. This is not a comparison: pick the one you use, not the one we’d pick for you.
Listed alphabetically. We don’t rank these.
Not sure which you’re using?
A few quick tells:
- Signed in with a work Microsoft account, and it appears inside Word, Excel, or Teams? That’s Copilot.
- Built into your Google account and can read your Gmail and Drive? That’s Gemini.
- Answering only from documents you uploaded yourself, with exact-passage citations? That’s NotebookLM.
- Lives inside X, with a casual tone and a feel for what’s trending? That’s Grok.
- Every answer comes with clickable citations to web pages? That’s probably Perplexity.
- Opens documents, code, or diagrams in a side panel you can edit in place? That’s likely Claude.
- Running on your own hardware, or through a developer API, with visible step-by-step reasoning? That’s probably DeepSeek.
- None of the above, and it’s simply called “ChatGPT”? That’s ChatGPT.
Anthropic
Claude
Best for long documents and sustained drafting you iterate on in place.
DeepSeek
DeepSeek
For step-by-step reasoning you can check, with the option to self-host.
Gemini
Best when your work lives in Google Workspace, or your prompt isn't only text.
xAI
Grok
For reading what's being said right now, not for settling what's true.
Microsoft
Microsoft Copilot
Best inside the Microsoft 365 documents and permissions you already have.
NotebookLM
Answers only from the documents you give it, and cites the exact passage.
Perplexity AI
Perplexity
Answers a question from the live web with sources you can click, immediately.
Using more than one?
These have different shapes, not different scores.
The skill is matching the shape to the job, not picking a favourite.
- Discover, then analyse. Use a search-grounded assistant to find primary sources; move the resulting fixed set into a source-grounded one for careful, cited analysis.
- Draft where the document lives. If the output belongs in a Doc or a Word file, drafting inside that ecosystem keeps context you would otherwise have to re-explain.
- Check across grounding types. If two assistants with different sources of truth agree, that is weak corroboration. If one searched the web and another only knows its training data, their agreement means less than it looks.
- Keep one assistant per task, not per question. Switching mid-task discards accumulated context and usually costs more than it saves.