Six guides, each reviewed on 14 October 2026 against the vendors' own documentation. Pick the job you need done, then compare the tools on the same fields: what each is best for, its documented strength, and the limitation most likely to stop you.
The guides
Best AI chatbots
ChatGPT, Claude, Gemini and Perplexity compared on file limits, web search, projects and whether your chats train their models. The deciding facts are often the dull ones: ChatGPT gives Free users 3 uploads a day, while Gemini's context window ranges from 32k tokens without a plan to 1 million on its top plans. Start here if you want one assistant for mixed work.
Best AI coding tools
GitHub Copilot, Cursor and Claude Code compared on where they run, how much they may do without asking, and what happens to your code. The guide includes independent research on developer speed and on security flaws in generated code. Read this before you let an agent edit a real repository.
Best AI writing tools
Grammarly, Claude, QuillBot and DeepL Write compared by how much rewriting you need: fixing errors, rewording a passage or restructuring a draft. Three of the four train on individual users' text by default, so the guide shows where to change that.
Best AI image generators
Adobe Firefly, Midjourney, Canva and Google's image models compared on editing, commercial terms, training data and content credentials. It also explains what the US Copyright Office says about who owns an AI image.
Best AI video generators
Google Veo 3.1, Runway Gen-4.5 and Adobe Firefly Video compared on clip length, resolution, audio and watermarking, with Descript included for editing footage you already have.
Best AI search tools
Perplexity, ChatGPT search and Google's Deep Research and AI Mode compared on citations and source controls. It includes the Tow Center's 2025 test, which found incorrect answers to more than 60% of news-attribution queries across eight AI search tools.
How to use these guides
- Start with the job, not the brand. Each guide opens with a short answer naming a documented starting point and the situations where another tool fits better.
- Check the limitation column first. Upload caps, clip lengths, usage windows and plan restrictions decide more purchases than headline features.
- Change data settings before you upload anything sensitive. Every chatbot and writing guide shows the training default for individual accounts.
- Run your own small test. Each guide ends with a short test you can repeat on your own work in an afternoon.
What "documented starting point" means
None of these guides names a measured winner, because HumanRank has not yet published hands-on tests. A documented starting point is the tool whose own documentation best matches the most common need in that category. When we publish a test, the guide will say so, link to the method and show the result. Until then, the score column reads "Not tested". Our ranking method explains the difference between documented facts, editorial judgment and tested results.
How these pages stay current
AI products change their limits, plans and defaults often. Each guide shows the date it was last checked. When a vendor's documentation changes, we update the affected sentence and the date, and we never move the date without re-checking. If you spot something out of date, tell us. For how commercial relationships are handled, see our disclosure, and for who is behind the guides, read about HumanRank.
More categories are coming
Additional categories will be published as their source checks are completed.