AI chatbots

The best AI chatbots compared: apps, with what each one does, what it costs and who it suits.

Updated 2026-11-22 by the The Review NYU editorial team

AI chatbots explained

AI chatbots have become the front door to artificial intelligence for most people. You type or speak a request, and a large language model answers: explaining a concept, drafting an email, summarising a report, debugging code, planning a trip or role playing a job interview. Their flexibility is their strength. The same assistant can help with homework at noon and a spreadsheet formula at midnight.

That flexibility also makes them hard to evaluate. Chatbots are fluent even when they are wrong, they behave differently depending on how you ask and they change frequently as companies update models and features. This section explains what chatbots are good and bad at, how the main assistants differ, how to get reliable results and how to use them responsibly. Our guide to AI chatbot apps compares app features in more detail. Descriptions of products are based on companies' own materials, and features change often.

How widely chatbots are used

Pew Research Center's survey of 5,119 U.S. adults in February 2026 found that about half now use AI chatbots, up from a third in 2024, with about a quarter using them daily. ChatGPT was by far the most used chatbot in that survey, reported by 44 percent of adults, followed by Gemini at 24 percent. The most common reason for using chatbots was searching for information, and among employed adults, work tasks were nearly as common. A year earlier, Pew had found that 58 percent of adults under 30 had used ChatGPT. Chatbots are no longer a niche tool, which makes knowing how to use them well a basic digital skill.

What chatbots are good at

Strength Examples
Drafting Emails, outlines, cover letters, social posts, first drafts
Explaining Concepts, jargon, code, legal or technical language in plain terms
Transforming Summarising, translating, changing tone, reformatting
Brainstorming Ideas, angles, names, questions to ask
Analysing provided material Documents, spreadsheets, transcripts you upload
Practising Interview role plays, language conversation, quizzes
Coding help Snippets, explanations, debugging, tests

What chatbots are bad at

Weakness What it looks like
Confident errors Plausible but false facts, numbers or quotes
Invented sources Citations, links or papers that do not exist
Outdated knowledge Answers that ignore recent events unless the tool searches the web
Hidden assumptions Filling gaps in your request with guesses
Sycophancy Agreeing with you rather than correcting you
Arithmetic and counting Errors in calculations unless the tool uses code or a calculator
Uneven performance Excellent on one task, poor on a similar one

Research on the jagged technological frontier helps explain the last point. In a preregistered experiment with 758 Boston Consulting Group consultants, Fabrizio Dell'Acqua and colleagues found large gains from GPT-4 on tasks within its capabilities, but on a task designed to fall outside them, consultants using AI were 19 percentage points less likely to reach the correct answer. The practical implication is to test a chatbot on your own tasks and to stay alert even after it has performed well.

How the main assistants differ

The leading general assistants share a core: a capable language model, a chat interface, file uploads and, increasingly, web search, voice, image understanding and tools that can run code or take actions. Differences tend to show up in:

Area What varies
Model quality on your tasks Writing style, reasoning, coding, languages other than English
Context length How much text or how many files it can handle at once
Web search and citations Whether answers link to sources you can check
Integrations Email, documents, calendars, coding tools
Voice and multimodal features Speaking, listening, images, video
Privacy controls Whether conversations are used for training, retention settings
Price and limits What free users get; caps on advanced models

Because the leaders are close in many areas and change often, we recommend choosing based on a short trial with your own tasks rather than on benchmark headlines.

Getting reliable results

  1. Give context. Say who you are, what the output is for and who will read it.
  2. Provide sources. Paste or upload the material you want used instead of relying on the model's memory.
  3. Be specific about format. Length, structure, tone and what to include or leave out.
  4. Ask it to show uncertainty. Request that it flag anything it is unsure about.
  5. Check facts and citations. Open every source before relying on it.
  6. Iterate. Say precisely what to change rather than starting over.

A one week test for choosing a chatbot

Benchmarks and launch announcements tell you little about how a chatbot will handle your work. A short structured test tells you much more.

Day Test What to look for
1 Explain a concept you know well Accuracy, clarity, whether it admits uncertainty
2 Summarise a long document you upload Faithfulness to the source, missed points
3 Draft something you write often Tone, how much editing it needs
4 A factual research question Sources, whether citations are real and relevant
5 A task with numbers or data Correct calculations, use of code or tools
6 A multi step task Whether it follows all instructions
7 Review your notes Which assistant saved time after checking

Run the same tests on two or three assistants. The winner is the one that needed the least fixing on your real work.

Chatbots that search the web

Many assistants can now search the web and cite sources. This reduces outdated answers and makes checking easier, because you can open the links. It does not remove errors: assistants can misread pages, rely on poor sources or summarise inaccurately. Treat cited answers as a good starting point for research rather than a finished one, and read the most important sources yourself.

Writing better prompts

Good prompts are usually longer than people expect. A useful pattern is role, task, context, constraints and format. For example: you are a careful editor; tighten this cover letter; it is for a junior data analyst role at a nonprofit; keep it under 300 words and keep my examples; return the revised letter and a short list of changes. Saving prompts that work for recurring tasks turns a chatbot into a dependable assistant.

Using chatbots responsibly

The U.S. National Institute of Standards and Technology's AI Risk Management Framework sets out ways organisations can manage AI risks, from accuracy to privacy and fairness. For individual users, a few rules cover most situations:

  • Protect data. Do not paste confidential, personal or unpublished information unless you know how the tool stores and uses it and you are permitted to.
  • Follow your institution's rules. Schools, universities and employers increasingly have AI policies, including on disclosure.
  • Check before you rely. Treat answers on health, law, money and safety as starting points, and confirm with qualified sources.
  • Be honest about AI use where it matters, such as academic work and published writing.

Free versus paid

Free tiers of the major assistants are capable enough for many people. Paid plans typically offer more use of the most capable models, longer context, more file uploads, advanced voice or research features and sometimes stronger privacy options. Pay when you regularly hit free limits on work that matters. Check first whether your school or employer provides an institutional licence, which often includes better data protection.

Chatbots for students

Students are among the heaviest users of chatbots. Used well, they are excellent tutors: explaining concepts in different ways, generating practice questions, checking understanding and giving feedback on drafts. Used badly, they replace the thinking that coursework is meant to build and risk academic misconduct. Read your course and university policies, use chatbots to learn rather than to produce work you submit as your own, and keep notes of how you used them in case you need to explain.

Memory, projects and custom instructions

Many assistants now let you save custom instructions, organise conversations into projects with shared files and remember details across chats. These features can save a lot of repetition: your preferred tone, your role, the course you are taking or the style guide you follow. They also mean the assistant stores more about you. Review what is remembered, delete what you do not want kept and avoid saving sensitive personal details. If you share an account or device, be careful about what memory features reveal.

Agents and actions

The newest assistants can do more than chat. They can browse websites, fill in forms, run code, edit documents and connect to your email or calendar. These agent features can handle multi step tasks, but they also raise the stakes of errors: an agent that misunderstands can send the wrong message or change the wrong file. Start with low risk tasks, review actions before confirming them and keep agents away from payments and sensitive accounts until you trust how they behave.

Example users

These illustrative examples are not real people.

A first year undergraduate uses a chatbot to explain difficult readings, generate quiz questions before exams and get feedback on essay structure, while writing every submitted word herself.

A marketing coordinator drafts campaign ideas and email variations with an assistant, then edits and fact checks before anything goes out.

A researcher uploads her own notes and asks the assistant to find gaps in her argument, but never uploads unpublished participant data.

Common mistakes

  • Treating chatbot answers as facts without checking.
  • Using invented citations.
  • Asking vague questions and blaming the tool for vague answers.
  • Pasting confidential information into a personal account.
  • Assuming one good result means it will be right next time.
  • Ignoring course or workplace AI policies.
  • Letting agent features take actions without reviewing them first.
  • Never reviewing what a memory feature has saved about you.

What users commonly say

These are recurring themes from user discussions, not quotations from individuals.

  • It explained things my textbook made confusing.
  • It gave me a citation that did not exist.
  • Better instructions made a bigger difference than switching chatbots.
  • Uploading the actual document stopped it making things up.
  • The free version was fine until I needed it every day.

How we write these guides

The Review NYU is an independent publication. We base these guides on hands-on testing, companies' own documentation and published research. We label company claims as such, do not accept payment for placement and never invent statistics, reviews or credentials. Read about our editorial standards.

In this section

  • AI chatbot apps: Guides to the best AI chatbot apps: what each one does, what it costs, and who it suits best.

Frequently asked questions

Which AI chatbot is best?

The leading assistants are close in many areas. Try two or three on your own tasks for a week and choose the one that suits your work, devices and budget.

Are chatbot answers reliable?

Often, but not always. Chatbots can produce confident errors and invented sources, so check facts that matter.

Is it safe to share documents with a chatbot?

Only if you understand how the service stores and uses data and you are allowed to share the material. Use institutional accounts with data protections where available.

Can students use chatbots for coursework?

It depends on course and university policies. Use them to learn and get feedback, follow the rules and disclose use where required.

Should I pay for a chatbot subscription?

Pay if you regularly hit free limits on work that matters, and check whether your school or employer already provides access.

How can I get better answers?

Give context, provide source material, specify the format, ask the chatbot to flag uncertainty and iterate with precise feedback. Save prompts that work for tasks you repeat.

Are chatbots that search the web more accurate?

They are more current and easier to check because they cite sources, but they can still misread pages or rely on weak sources. Open and read the most important links yourself.

What are AI agents?

Agent features let an assistant take actions such as browsing, editing files or sending messages. They can handle multi step tasks, but review actions before confirming them and start with low risk tasks.

Should I turn on chatbot memory?

It saves repetition, but it also stores more about you. Review what is remembered, delete what you do not want kept and avoid saving sensitive personal details.

Sources

  1. Americans and AI 2026: Chatbots, Smart Devices and Views on Impact, Pew Research Center, June 2026
  2. 34% of U.S. adults have used ChatGPT, about double the share in 2023, Pew Research Center, June 2025
  3. Navigating the Jagged Technological Frontier, Dell'Acqua and colleagues, Harvard Business School Working Paper 24-013, September 2023
  4. AI Risk Management Framework, National Institute of Standards and Technology, accessed November 2026

This guide is independent and editorially reviewed. Company claims are labelled as such, and we never invent statistics, reviews or credentials. Read about The Review, our editorial standards, or contact us.