The best AI chatbots in 2024

ChatGPT was the first widely used AI chatbot, but now the competition is getting fierce. Other models are joining the scene, offering longer conversational memory, empathetic responses, and grounding in your own data—among many other possibilities.

Allow me to save you a dozen Google searches (or ChatGPT prompts). I spent time talking to some of the best AI chatbots to see how they measure up. You’ll find a bit of everything here, including ChatGPT alternatives that’ll help you create content, AI chatbots that can search the web, and a few just-for-fun options. You’ll even see how you can build your own AI chatbot if you don’t find what you’re looking for here.

It’s likely that between the time I write this and the time you read it, there will be even more AI chatbots on the market, but for now, here are the most interesting ones to watch.

The best AI chatbots

What to look for in an AI chatbot

Before we dive in, let’s talk a bit about how ChatGPT works. This will help you understand what’s interesting about each AI chatbot and use it to your advantage.

ChatGPT is an app created by OpenAI that lets users interact with its AI models, including its two newest ones: GPT-4o (available on paid tiers) and GPT-4o mini. The app takes the prompts you write and passes them to the AI model. This model runs the prompt through its systems and returns the results back to the app, so you can read them in a conversational chatbot style.

While the app takes care of the features—for example, saving your conversation history—the AI model takes care of the actual interpretation of your input and the calculations to provide an answer. For more context, take a look at our breakdown of ChatGPT vs. GPT.

Many of the apps on this list are also powered by OpenAI’s GPT models. But even when that’s the case, the app developers can pass additional commands to configure how the model replies, so you may see different results when you try each chatbot, even though they’re running on the same engine. (Some apps on this list use non-GPT models, which are often proprietary to the company that built the app.)

With that in mind, here’s what I was keeping an eye out for as I tested each AI chatbot online:

• AI model quality. The engine behind the app needs to produce consistently good answers, with fluency and decent accuracy.

• Conversational experience. The combination of the app and the AI model needs to create an experience that feels good to use, from the kind of output you’ll receive to the flexibility in completing the tasks you put forward.

• Chat app tools. Does the app offer more than just a message back-and-forth? Can you search the web, generate images, create charts, or build mini-dashboards?

• Integration with a core app or other apps. Some chatbots exist to add more features on top of a platform; others focus on connecting with multiple third-party apps to improve their data sources or functionality. I made sure to include the best across both worlds.

• Extra features. Anything on top of entering your prompt and receiving the output is welcome, from multi-language support to connecting directly to the internet.

For this update, I decided to only include chatbots that are actively being worked on and improving at a good pace. Since this category is evolving so fast, it’s natural to select the tools that are adapting to the changes as fast as possible.

Based on my research and experiences interacting with them, here are the best AI chatbots for you to try. Have fun—I know you will.

---

The original AI chatbot

Model: OpenAI GPT-4, GPT-4o-mini, GPT-4o, DALL·E 3

ChatGPT is the original AI chatbot. After its late 2022 release, it took the world by storm in 2023 and put AI back in the news. It’s very simple to use: start a conversation by writing your prompt at the bottom of the screen. As the output comes back, a new entry is created on the left-side menu, so you can keep all your threads separate and come back to them later. If any of these is particularly interesting, you can share a link to it and show it to others (and they can start where you left off).

Even though it sometimes puts out factual errors while displaying total confidence—what experts call hallucinations—ChatGPT is still the industry leader for now. It remembers what you’ve said within each conversation, using it as context to provide more accurate output as it moves forward. It can accept text commands, helping you format and customize the output. And it’s extremely flexible, tackling tasks in any discipline with an acceptable level of accuracy—just be sure you fact-check. You can share your conversations with others and add custom instructions to customize the bot even further.

With the release of GPT-4o, ChatGPT got even more interesting. It’s now using a multimodal model, which means you can do everything from text-to-image prompting, data analysis, and voice conversations—and it’s impressively fast.

The innovation doesn’t stop here: OpenAI is brainstorming its Google Search killer. Currently in prototype status, SearchGPT will use GPT technology combined with online searches to surface the best results for any question you have. It’s still unclear when (or if?) it will be released, but if you’re curious, jump on the waitlist to learn more.

You can even build your own custom bot with unique instructions and data using GPTs. These bots can carry out straightforward tasks, simplify processes, or just do zany things for the fun of it. When you add Zapier’s AI Actions to it, you can access thousands of other apps from your GPT too.

You can also do the opposite, building ChatGPT into your existing workflows with Zapier’s direct ChatGPT integration. No matter where you are, you can use ChatGPT to summarize, generate replies, or anything else you can dream up. Here are a few examples.

Want to build with the GPT models? Brainstorm your prompts in the OpenAI Playground, and follow Zapier’s guide to the OpenAI API.

Creating interfaces with Artifacts

Model: Claude 3 Haiku, 3.5 Sonnet, and 3 Opus

Meet Claude. Anthropic’s chatbot aims to be helpful, harmless, and honest. The conversation flows naturally, with responses that are straight to the point, without lengthy introductions and conclusions like ChatGPT sometimes prefers using.

The context window length is pretty good. Claude can remember up to 150,000 words per conversation, great for uploading PDFs and asking questions or chatting away until the morning sun. Claude 3.5 Sonnet is outperforming GPT-4o in some benchmarks too, a surprising turn in the AI LLM leaderboard. 

But the best feature is Artifacts, hands down. With just a few prompts, you can create a dashboard page to display your data, with buttons to change it if needed—like this personal budget planner. Some users were even able to create animations and simple games (check out this physics game created with Claude). These creations are accessible as you chat, so you can change them with prompts and interact with them in real time. To activate them, go to the main menu on the bottom-left side of the screen, click Feature Preview, and toggle it on.

On top of Artifacts, Claude helps you organize documents, code, or files into Projects, so you can quickly reference that data and keep everything organized in your workspace.

The free plan used to be limited to 25 messages per day, but now it seems this varies depending on demand—the app will let you know when you’re close to the limit. You can unlock more by subscribing to the Pro plan for $20 per month. Once you remove that cap, you can integrate Claude with Zapier to automate your tech stack. Learn more about how to automate Claude, or get started with one of these pre-made workflows.

Want to build with the Claude models? Refine your prompts in the Anthropic Console, and follow Zapier’s guide to the Claude API.

Open license

Model: Llama 3.1

Meta recently released Meta AI, accessible only in the U.S. for now. You can log in with your Facebook or Instagram account and start chatting. You can ask for generated images and even small animations—there’s the option for creating short videos, but the system doesn’t seem to be working well yet.

The response quality is pretty good, but the model hallucinates sources from time to time and isn’t shy to state everything confidently. It doesn’t have a web search tool to enrich the responses, so be sure to fact-check everything to the letter if it’s important. Because of this, it feels like a more vanilla model ready to be tuned and adapted to specific use cases or business needs. Combine it with your data and business logic, and suddenly you have a good reasoning engine using your facts.

Which brings us into the generous open licensing framework.

If you’re a developer, you can use the Llama models to empower or build your apps and pay nothing until you reach a very high revenue level. These terms were designed to keep other big tech companies away from Llama, so most businesses of all sizes could use this AI model without having to pay any access costs—just the cost of having hardware running the calculations.

The Meta AI website suggests this engine may one day be fully integrated with the company’s social media platforms—it’s already there in some of them. If you’re curious about how it stacks up against other models, read the ChatGPT vs. Meta AI comparison to learn more.

Largest conversational memory

Model: Gemini

Google has been in the AI race for a long time, with a set of AI features already implemented across its product lineup. After an epic hiccup during the initial product demo, Gemini (formerly Bard) is really growing on me.

Gemini can connect to the internet to find sources (even offering a handy button that lets you Google it yourself), which is a huge selling point. The search results can show images directly on the chat window. It also lets you edit your prompt after you’ve sent it and offers up to three drafts of each output, so you can pick the best one. It can keep track of your conversation history—it has the longest available context window at the moment—and you can share your conversations with others.

But here’s what I love: it integrates deeply with your Google account and with other Google products such as Hotels, Flights, and YouTube. Want to search your Gmail file jungle with one prompt? Do it. Summarize files inside your Google Drive? Yes, please. Check real-time flight and hotel prices as the AI builds your trip? Schedule that time off: it even gives you a packing list. But no matter what Gemini says, please don’t put glue in your pizza.

When compared with ChatGPT, Gemini feels more conversational and less oriented toward text commands. To read more about their differences, here’s a direct comparison: ChatGPT vs. Google Gemini.

And with Zapier’s Google Vertex AI and Google AI Studio integrations, you can access Gemini from all the apps you use at work. Here are a couple of examples to get you started.

Want to build with Gemini? Start creating your prompts in Google AI Studio, get your free development key, and follow Zapier’s guide to the Gemini API.

Online search, text, and image generation

Model: OpenAI’s GPT and DALL·E models

Early in 2023, Microsoft upped its investment in OpenAI and started developing and rolling out AI features into its products. One of those was Bing, which now has an AI chatting experience that will help you search the web. Once you enter your prompt, it will search the internet for you, process the results, and present you with a reply containing the links it used as a base.

It can now show image results in the chat window, but it doesn’t pick up too well on the intent of image search: it usually prints out a list of image links instead of an image gallery. And if you want to re-read past conversations, you can do so by clicking on each one on the right side of the screen.

Moving away from search, Copilot is now connected to OpenAI’s DALL·E 3, so I can get my daily fix of AI-generated golden retriever images. State that you want to generate an image while chatting, and enter the prompt. Once it’s out, you can ask to regenerate with new instructions, though in my experience, it doesn’t work well 100% of the time.

It’s fully integrated into Microsoft Edge, the native Windows browser. There are a few cool tools there. The best one lets you generate text in the sidebar and paste it with one click into any input field on a web page. This is useful for generating emails and pasting them into the compose window, for example.

Microsoft Copilot is still behaving strangely, sometimes ending conversations abruptly—still, it’s nothing like when it revealed its gaslighting skills. Don’t take it personally if it says it doesn’t want to continue the conversation. Hit New Thread, and keep going.

Here’s a full rundown of ChatGPT vs. Microsoft Copilot.

Automation

Model: OpenAI’s GPT models

Zapier Central is an experimental AI workspace where you can teach AI assistants to work across thousands of apps. But the interface is a chatbot, which means that creating your own AI assistant is as simple as using ChatGPT.

Start by giving your assistant access to your company’s source of truth. Then, you can ask it about that data as well as all the data in your apps—and data sources will sync automatically. You can train your assistants to work while you’re not there: they can write, send and update information, analyze data, and everything in between.

And because it works across all your apps—Slack, Gmail, Facebook Lead Ads, Google Sheets, Airtable, Shopify, you name it—you can trigger behaviors based on anything that happens in your tech stack.

Think of these assistants as mini teammates who help you with specific tasks, like analyzing your spreadsheets, searching the web, or drafting responses to customers. By chatting directly with AI, you can train them by showing them the exact behaviors to follow—without using a line of code.

It’s AI chat, automation, and data analysis all in one place. Learn more about Zapier Central, and then get started for free.

For trying multiple AI models

Model: OpenAI’s GPT and DALL·E models, StableDiffusionXL, Claude, Gemini, Llama, Solar, Mistral

GPT-4o, Claude 3.5, and StableDiffusion XL walk into a bar. That bar is called Poe, which acts as a kind of AI model aggregator. It’s the best if you want to try out the top models on the market right now for a single-pack price of $19.99 per month. There will be message limits for the highest quality models, but it’s still better than subscribing to each individual one if you want to explore.

When you’re done chatting, you might want to try creating your own bot. You can set its system prompt, add a knowledge base, and configure a greeting message. This feels like one of OpenAI’s GPTs, with two key differences: you can pick which AI model you want to use as a base, and you can monetize it by setting up your creator account. As I write this, you get $20 whenever a user subscribes to Poe by using one of your chatbots. If you create a good one, maybe your Poe subscription can pay for itself multiple times over.

For internet deep dives

Model: OpenAI’s GPT models and Claude

Perplexity is another take on AI internet-connected chatbots for handling more information and longer, less organized searches. Here’s why: when getting the output, you’ll see a list of all the sources below it. Then, you can add a new prompt to continue that search, or pick one of the suggested related search terms. All the results will stack on the bottom, so you can scroll up or down to read everything.

You can tick Copilot in the search bar to get some help with product recommendations, best healthy recipes, or travel tips, for example. Once you enter your prompt, Perplexity will ask you a set of qualifying questions to home in on your intent. The resulting output summarizes all the key information, acting as a good starting point for a deep dive.

When you share your chats with others, they can continue the conversation you started without limitations. On your end, you can see the views for shared conversations, likes, and follow-up questions, making the experience more interactive.

A new feature, Discover, rounds up popular searches into one short, snappy article. For example, “the most streamed artist in 2023” brings together the sources that prove that Taylor Swift attained 26.1 billion streams, coupled with a short AI-generated summary unpacking the important information about the topic.

Now better prepared for advanced tasks like solving equations and processing real-time data by leveraging Wolfram Alpha, Perplexity is positioning itself more and more as a powerful and straightforward search tool. Learn more about how it compares to ChatGPT.

Open source

Model: Mistral model family

Ouh là là. The appropriately-named le Chat is out to show Mistral’s high quality models, with a simple conversation experience. The chat app isn’t as powerful as others on the list, but the answers of the AI model consistently hit the right response length. Sometimes it’ll go verbose when explaining a topic in bullet points, other times it opts for a single paragraph to give basic information. This makes the conversation easier to follow—I don’t really like when AI models always give me wall upon wall of text.

When starting a new chat, click the model dropdown to pick the one you’d like to try. There are three available at the time of writing: Large 2 is your typical know-it-all, generalist LLM; Nemo is fast and cost-effective for simpler tasks; and Codestral can be your coding companion, helping you solve complex problems or that semi-colon on line 182 you forgot last night at 3 a.m.

The company behind the models is based out of France, made up by professionals from both Google’s DeepMind and Meta AI. A small company, they’re rising up to proeminence, their model featured in other chatbots due to the high quality results. With a focus of providing open generative models to empower devs and businesses, it’s another big play against the closed-model frameworks of OpenAI.

For building your own chatbot

Model: OpenAI’s GPT models via your own API key

Zapier’s free Chatbots tool lets you create custom AI-powered chatbots with the power of GPT. Creating a chatbot is easy: you just give it a name, an optional greeting and prompt placeholder letting chatters know what to do, and then the directive. In the directive, you’ll tell the chatbot who it is, what it knows, and how it should answer—and Zapier gives you lots of help along the way to be sure your prompt is on point.

You can turn the creativity up or down (like you might in the OpenAI playground) and even customize the look and feel of your bot. And you can train the bot on specific documents, so it can serve as a knowledge source based on your documentation.

Then you can create a nice little landing page for it and give it a unique URL that you can share with anyone. It’s a great way to create a support or lead generation bot, and because it’s built on Zapier, it’s connected to thousands of other apps. Read more about how to build your own AI chatbot with Zapier.

Open source models

Model: Llama, Mistral and Mixtral, Cohere, Nous Hermes, Yi, Microsoft Phi

Presenting HuggingChat, an open source chatbot assembled by Hugging Face. It used to have a cheery model that could hold the conversation for a few dozen turns before breaking down into an incoherent jet of nonsense. That’s the past: the new HuggingChat is now powered by multiple high-quality open source models from Meta, Mistral, and even Microsoft’s Phi.

The image above is the model selection popup. Each has a different number of parameters—which dictates their flexibility, among other things—so be sure to read the description to understand which would be best for the task at hand. You can add a system prompt to change the core behavior of the model too, helping you tune it to your workflow.

Like ChatGPT, you can also use Assistants here. These will change the system instructions, add start messages, and determine whether there’s internet access or not. Unlike GPTs, you can’t upload documents into the knowledge base, which limits the flexibility of these tools.

It feels great to use HuggingChat. It’s something to compare ChatGPT to, revealing a bit about how these models take your inputs and calculate the outputs. You can definitely add it to your brainstorming toolkit, and start testing it for more important workflows over time.

You can also connect Hugging Face to Zapier, so it can talk to all the other apps you use. Here are some examples of how to automate Hugging Face, or you can get started with one of these templates.

Chatting with X’s content

Model: Grok-1.5

I finally got a chance to chat with Grok, X’s resident chatbot. Available on the X Premium plan—with article writing unlockable at the uppermost tier—it promises to be more flexible while discussing sensitive topics.

In terms of safety features, Grok is high up there. It won’t discuss anything it interprets as hate speech, offering a polite note on its boundaries. It won’t even engage with derisive humor or anything that could take on a double meaning.

I started discussing existence, politics, and economy anywhere on the spectrum between nuance and extremism, but Grok never veered too far off the middle ground. There weren’t meaningful differences when treating these topics when compared with a model like GPT-4o. I was expecting something more extreme, a challenge to my way of thinking. Perhaps those were just my assumptions when considering how the model presents itself and the media coverage it got in the past.

Grok is great as an X explorer. When starting a new conversation, you can click recent trend suggestions to understand a bit of what’s going on around the social network. This is super useful to understand at a glance what the new trends are about. Since the chatbot offers posts as sources, you can quickly see top creators and jump into the conversation quickly. With enough time, this entry point could become an alternative feed to engage with the content on the platform.

There are two chat modes, Fun Mode and Regular Mode. The former is a bit off the mark: if you’re up to having a longer, tongue-in-cheek conversation, the bot repeats itself a lot, especially on the first paragraph of each response. Regular Mode is much, much better so it’s probably best that you keep it on by default.

For personal use

Model: Inflection-2.5

Pi: with a short name that stands for personal intelligence, Inflection AI’s new chatbot sets out to be engaging and supportive. It may not be as flexible as ChatGPT—it can’t write articles or search the internet—but it offers a fresh user experience that’s worthy of mention.

The app is minimalistic and filled with loads of cute details and animations. You’ll find that Pi isn’t tuned for long answers. Instead, it prefers shorter bursts of conversation and loves asking questions. It wants you to share your day, mention difficulties you’re having, or talk through problems in your life. It’s friendly, and while vague at times, it always has nice things to say.

In addition to the standard chat mode, you can switch to SupportPi to talk things through, get advice, or just get a sounding board for stuff on your mind. You can combine these models with the Discover section, where you can choose a conversation type, with options such as “practice a big conversation,” “get motivated,” or “just vent.”

No big productivity use cases, sure. But maybe there’s no productivity if you don’t keep your personal stuff in order, and that’s exactly Pi’s angle. 

It’s free to use at the moment, so you can jump right in. To keep track of your conversation history, you’ll have to provide your name and phone number. This way, Pi will be able to text you from time to time to ask how things are going, a nice reminder to check in and catch up.

For searching the web

Model: OpenAI’s GPT models, Claude, Llama, DBRX, and others

You.com, formerly YouChat, used to look like a Google Search results page. Now, it’s resembling more and more Perplexity’s instead, with a greater focus on the chat experience.

Once you enter your prompt and receive the output, you can browse a list of web search results on the right side of the screen. At the bottom, you can also find contextual buttons that open up a collection of Reddit posts about the topic or maps with pins of any places discussed, for example. If you like the simplicity of ChatGPT, this might feel a bit crowded, but it’s great for browsing lots of information faster.

You can adjust the priority that the engine should give to different sources by up- or down-voting them. This feature is called Apps—you can browse a huge list containing names such as Reddit or TechCrunch, and you can set the priorities based on your interests.

If you’re feeling adventurous, you can swap out You.com’s brain: choose one out of seven AI models that’ll read the sources and generate an answer. The results vary a lot: it’s cool seeing how responses vary depending on how the processing is done. Who knows, you may end up ditching the almost-classic GPT models for something more niche.

The content writing powerhouse

Model: OpenAI’s GPT models

Jasper is an all-in-one AI content creation powerhouse, now targeting larger businesses with larger problems. It has AI templates for all kinds of content types—YouTube video scripts, blog posts, LinkedIn profile, about page copy, you name it—and has also rolled out its own Jasper Chat, joining in on the hype as a ChatGPT competitor.

The great part about it is that you can quickly turn a conversation into a document (or more), making ideation and pushing first drafts easy work. When you input a prompt to create an article, Jasper Chat will return the result and suggest follow-up articles on similar topics.

The output quality is more or less the same when compared with ChatGPT—after all, they both use OpenAI’s GPT models—but when reading the output, it feels like Jasper’s developers are tuning it to adapt better for content production. Jasper Chat also connects to the internet, so you’ll be able to fact-check faster with lists of fact sources.

Once you have dozens of fresh pieces to post, you may need images to go along with the text. Jasper also offers an AI image generation add-on, so you don’t have to leave the platform to take care of aesthetics. All these features come with a price, but if you’re on the high-volume content game, it shouldn’t feel too expensive for the power you’ll have at your disposal.

You can connect Jasper to Zapier to automate a lot of your content creation workflows. Discover the top ways to automate Jasper, or get started with one of these pre-made workflows.

For GTM tasks

Model: OpenAI’s GPT models

The Copy.ai team is light on their feet. The platform used to be on the content generation side—and, to be clear, it still retains some of these features—but the packaging changed completely. Now, instead of copywriters and content marketers, Copy.ai is helping sales teams with their go-to-market efforts.

What does this mean? B2B sales teams have a lot of work to do as they look for, qualify, and convert leads. Finding accurate contact information, researching prospects, and keeping all the information organized and actionable takes a lot of time and mental space. Copy.ai offers automated workflows, buildable with AI and powered by it as well. You can, for example, create a flow where you input the URL of the target company, and the app does the rest: it’ll visit the site, scrape the pages, and create a report so you can understand what they’re all about.

The chatbot is one of the main ways you can interact with the data here, which keeps the platform neatly within this list. Since you can add your brand voice, you can generate emails for these leads or quick reply ideas to customer objections. If you explore the prompt templates, you’ll find most of the old templates to write blog posts or ad copy, great for using the data you have here to create content. (Learn more in our Jasper vs. Copy.ai comparison.)

You can do even more with Copy.ai by connecting it to Zapier, so you can access it from wherever you spend you time. Learn more about how to automate Copy.ai, or try one of these pre-made workflows.

For sales, marketing, and exploring your HubSpot CRM data

Model: OpenAI’s GPT models

HubSpot is in on the AI race too: the CRM giant is building ChatSpot, still in beta, and it has a lot of potential. The app can be connected to your HubSpot data, so you can:

• Create, update, and add new tasks to any contacts

• Run reports

• Research companies and leads

• See analytics data for your HubSpot sites

• Do keyword research

All this with natural language prompts instead of a festival of clicks on the HubSpot CRM app. But these features are just the beginning. You can also use ChatSpot to write blog posts and post them straight to your HubSpot website.

To further improve productivity, it connects with Google Drive to speed up document creation: you can send any prompt straight to a Google Doc, a table of companies to a Google Sheet, or a generated image to Google Slides, all with just prompts. It used to offer dad jokes in loading screens, but it’s so fast now that you’ll be spared from this kind of humor. And you can connect HubSpot itself to thousands of other apps using Zapier—read more about how to automate HubSpot, or get started with one of these templates.

For chatting with CRM data

Model: Einstein-1

Salesforce is the biggest CRM on this planet. With such a huge client base and far-reaching features covering so many sales problems, adding AI into the mix is like dropping rocket fuel into the fire. The Einstein platform offers a suite of AI-powered tools to help automate your CRM and provide better visibility of the data you’re collecting every day.

Offered as a separate package, the Einstein Copilot opens on the sidebar to answer questions about your sales data. Any question you ask comes with sources, so you can jump directly to the contact or deal at hand. Less clicking, less wandering around looking for lost data, and the bonus of being able to ask direct questions to correlate or compare data.

But the chatbot is just the tip of the iceberg. The platform has a dedicated prompt engineering screen. Here, you can write prompts to quickly summarize customer cases or deal status, while adding and pulling data variables from anywhere in the platform. You can then add these actions as buttons across the CRM’s interface. You’re basically adding little AI-powered buttons everywhere to quickly get insights so you can move forward faster. The downside? Be ready to code, or at least understand basic systems design.

You can do more by connecting Salesforce to Zapier to automate all of your CRM workflows. Learn more about how to automate Salesforce, or get started with one of these pre-made workflows.

For building a customer support chatbot easily

Model: OpenAI’s GPT models

Hey, what’s that friendly boop I’m hearing on that website? That’s Intercom’s customer support chatbot, now powered by AI. Fin can take care of the repetitive messaging using a combination of methods from knowledge bases to more scripted flows. All to free up human intelligence for the bigger customer support challenges.

Spotify and Slack are already using it. I asked a few questions to their demo bots, and they were all on point, offering sources and extra resources in case I were interested in jumping down the rabbit hole.

The process of building a support chatbot was also the easiest I’ve found: you can simply add your website URL, and Fin starts crawling it, learning everything there is to know about what you do, how you do it, and the frequently-asked questions that pop up when people find you for the first time.

Once it’s trained and deployed, it’ll be ready on the corner of your website’s viewport. It’ll keep the conversation going forward until it runs into the human transfer settings, passing the ball to a real rep so you can solve the problem with the power of human intelligence.

Fin can also take care of email support, and because improvement is important, there are handy analytics on data like conversation time or transfers to reps so you can add a few new behaviors, more knowledge, or adjust the nuts and bolts to make the experience better.

And when you connect Intercom to Zapier, you can automate even more of your support workflows. Learn more about automating Intercom, or get inspiration from these pre-made workflows.

For building a customer support chatbot with multiple data sources

Model: Proprietary

Branded as an AI agent for customer service, Ada can leverage multiple data sources to get things done. You can connect your analytics for context, business systems for actions, communication channels to meet users where they are, and your customer data platforms for personalized answers. Code required? Not a lot. Wire up the APIs, and Ada can work on it.

Because of this breadth, the platform includes a set of builders that help you leverage both the flexibility of generative AI and the structured experience that a customer support chatbot should have. In the Guidance screen, you can advise the chatbot on how to proceed based on certain topics or actions. In Processes, you can structure answers for sensitive topics so the bot doesn’t generate something that doesn’t reflect what your company does or stands for. And with so many possibilities for enriching the knowledge base, Ada can know everything it needs to help your customers solve their problems.

The website chatbot isn’t the only place where the action happens. Ada can take care of your phone calls, WhatsApp requests, and messages on Messenger and Instagram. A real multi-channel giant out to automate most of the low-level customer support work so you can take care of the tough problems with more time and grace.

For building a customer support chatbot flexibly

Model: OpenAI’s GPT models, Claude

Botsonic, a product from the Writesonic family, is a flexible tool for building chatbots for a variety of use cases, from customer support to shopping assistance. It integrates with many models, from OpenAI to Claude, but it will only let you select those that the team believes to be the best at the moment. The advantage here is that you’ll always be using the industry-leading AI model without having to switch platforms or reconfigure everything.

You can train your bots on your website data or by uploading your documents. If you use Notion, OneDrive, or Google Drive, you can pull in data from these sources as well. There are good controls for the appearance and user experience of the bot: you can pick colors that better match your branding or add starter questions to help your visitors know what they can ask.

On top of these tools, you can also create guidelines and agent actions. The former will let you customize how the chatbot responds to the users—for example, how to set the tone or how to respond to certain sentences based on the topic. Agent actions are more powerful: they’ll let the bot run actions on your internal systems based on what the users write. This can be useful to grab support ticket status or to check if a product is available on your shop. A bit trickier to set up, but once you have it, your chatbot will be your website’s master of ceremony.

You can connect Botsonic to the rest of your apps using Zapier, so you can send any form submissions from your bot to the other apps you use most. Here are some templates to get you started.

For streamlining multi-channel communication

Model: OpenAI GPT-4o, Claude 3.5 Sonnet, Llama 3, Upstage Solar

Sendbird is a unique platform. It offers communication features that you can plug into your other software to help people talk to one another. Handling live chats, video calls, and messaging in whichever platform you need, it also offers an AI chatbot builder to help automate conversations.

Once you log in to the platform, click on AI Chatbot on the left side menu. Once there, you’ll be able to create a new bot and customize it just the way you need it. You can change its appearance, fill the knowledge base, and choose which AI model you’d like to use—currently, you can choose OpenAI, Meta, Claude and Upstage models.

The AI chatbot can run actions when you connect other systems to it, such as letting the user search for their deliveries with a tracking code to see where their parcel is. If you need more structured conversations at any point, you can create workflows that are basically like a traditional chatbot: once triggered, the user can choose any of the options that you set up in advance.

If you’d like to perfect your chatbot over time, head over to the Analytics tab. You’ll be able to see statistics on messages, user feedback, and even which resources the bot accessed more frequently. This can give you clues on how to expand and improve, so you can add more knowledge sources or new actions.

Plus, you can connect Sendbird to Zapier so it can talk to all the other app you use too. Here’s an idea for how to get started.

For messaging and training your AI clone

Model: Proprietary GGT-P, OpenAI’s GPT models, Claude

It’s difficult to get GPT-4o to communicate exactly like you do. Personal AI helps fix that. It’s based on a proprietary AI model called GGT-P, and it trains on your data: messages, documents, favorite links, and anything else you send it. Your personal AI will learn what makes you you and will adapt to your communication style.

It doesn’t require a massive amount of data to start giving personalized output. To make each response more flexible, it uses OpenAI’s GPT models to plug in the gaps, creating a mixture between a general and a personal response. You can see how much of each it is by taking a look at the Personal Score percentage.

Personal AI is great for messaging others, writing emails, or creating content in your own voice and starting from your own knowledge. It will suggest replies in your conversations that will start sounding more and more like you over time. Once you’re confident it represents who you are and what you know, you can set it to Autopilot on a conversation-by-conversation basis to let it take over.

You can also connect Personal AI to Zapier, so you can automatically create memories for your chatbot as you’re going about the rest of your day. Discover the top ways to automate Personal AI, or get started with one of these pre-made workflows.

For personal productivity

Model: OpenAI GPT models, Claude, Gemini, Mistral

Merlin AI claims to be a 26-in-1 AI productivity assistant. Here’s why: it understands the most common actions you take when interacting with AI and builds tools and shortcuts to help you do them faster. When you press Ctrl (or command) + M on your keyboard, the available actions pop up, and you can quickly choose what you’d like to do. This beats always having to go back to a browser tab to ask a question or generate an image.

Available as a browser extension, you can chat with web pages, upload a PDF and ask questions to it, and summarize Google searches with a couple of clicks. It’s a Swiss army knife of AI actions, helping you write emails in Gmail, summarize YouTube videos, and quickly comment on tweets as you read them. Another interesting possibility here is creating a double of any public person’s tweets: you’re basically training a chatbot with their past posts so you can ask questions to it.

It integrates with most of the top AI models available, including OpenAI, Gemini, Claude, and Mistral. If you’re not sure which one is best for each task, the Merlin Magic feature can pick the best one for you—which is also good to help you try out new models from time to time.

For personal productivity

Model: Sophos 2, OpenAI’s GPT models, Claude, Llama 3

TextCortex is a content generation app that has a collection of templates to turn your prompt into a first draft quickly. Within this app, you’ll find ZenoChat, the chat experience powered by Sophos 2, an in-house model. OpenAI’s GPT-4, Claude, and Llama are also available if you’re looking for something more familiar.

The proprietary model improved a lot since I last tested it. It can generate good output, leaning on brevity and straightforwardness. You can tune its base personality in the chat box dropdown, enable or disable web search, add a knowledge base to it, or set it to a different language.

It has a good collection of prompt templates in the marketplace. Simply browse and pick the one that best matches the task at hand. If you’re too brief when writing prompts, ZenoChat has a unique feature that expands your prompt with as much detail as possible. This way, when you send it over, you can be sure you covered all the bases to get the best possible answer.

For fun

Model: C1.2

Ever wondered why Elon Musk bought Twitter? While you can’t pick up the phone and ask the real one, you can hop on Character.AI and do the next best thing: ask an AI model set to talk like Musk and brainstorm together about what’s next for the social media platform.

There are plenty of characters to talk to. Click on their profile to see more information about them, and if you’d like to start a conversation, you can do so with a few clicks.

It’s also possible to create characters of your own, with an impressive set of controls. You can then proceed to train them by chatting and rating the responses it gives you. If you’re looking for an AI chatbot for fun, this might be your pick.

AI in social media

Model: OpenAI’s GPT models

Snapchat made a name for itself by introducing disappearing messages into the social media scene. Now it also offers My AI, an AI chatbot that can answer almost anything directly within the app.

It’s powered by OpenAI’s models, so the output isn’t wildly different from the original ChatGPT experience. To access it, open the app, and tap the chat icon, where you’ll find the My AI conversation. You can tap its profile image to change settings and manage your data.

As you might expect, the uses for productivity are limited, but My AI can answer quick questions on things to do around town, offer gift advice, and even comment on snaps you send over to it: “Wow, those eyes are really unique!”

If you’re interested in new chatbots in development for social media, be sure to take a look at TikTok’s Tako too.

For learning

Model: Proprietary

Khan Academy has built a reputation for offering high-quality learning resources for free. As AI opens up new avenues in learning, Khan Labs is working on Khanmigo, an AI-powered tutor to help you master complex topics.

It will offer one-on-one tutoring on topics ranging from history to mathematics, helping you wrap your mind around the core issues. What I like about it is how it doesn’t tell you the answer to an exercise—instead, it asks you a set of questions and provides hints to get you to think your way to it. It never, ever yields the answer, not even if you beg.

As you progress through Khan Academy’s curriculum, you can review topics, see what’s next, and hop on interactive quizzes to keep knowledge fresh. This interactivity is a breath of fresh air in the familiar online course experience, making the material more approachable and fun to engage with.

For coding autocomplete

Model: OpenAI Codex

Technically, GitHub Copilot doesn’t have the chat-like experience you’re used to when using ChatGPT. But since it integrates with your integrated development environment (IDE) and acts as an autocomplete, it sort of feels like you’re having a dialogue with an AI model as you code.

When you start typing a comment or writing a function, Copilot will suggest the code that best accomplishes what you’re setting out to do. You can tap to cycle through all the suggestions, and if you find a fitting one, press tab to paste it.

Since there can be security risks when using generated code, Copilot includes security vulnerability filtering to ensure it doesn’t create more problems than it solves. You’ll still have to audit the code, especially since some suggestions aren’t as efficient as they could be. If you want to take a look at the productivity and happiness impact of using Copilot, be sure to take a look at this study.

You can connect GitHub to all the other apps you use with Zapier, so you can do things like integrate GitHub with Slack. Learn more about how to automate GitHub, or try one of these pre-made templates.

For coding autocomplete

Model: Proprietary

Amazon is in the AI game, too, with Q Developer (fomerly CodeWhisperer), a machine learning solution to help developer productivity. It’s very similar to GitHub Copilot with a few key differences:

• Amazon is running a proprietary AI model trained on open source code and Amazon Web Services (AWS) usage data. It then trains on your own project data to learn about your coding practices to generate more personalized suggestions.

• Since it’s built with AWS in mind, it will help you stay within the code best practices for Amazon services such as EC2, Lambda, and S3.

• It doesn’t support as many languages as Copilot, being limited to C#, Java, JavaScript, Python, and TypeScript. As for IDEs, it supports all Amazon-based ones along with JetBrains and Visual Studio Code.

For coding autocomplete

Model: Proprietary

Tabnine feels like a more enterprise and team-oriented solution. Its AI models are trained in fully open source code, keeping copyleft code outside. For context: copyleft code can be freely used and shared as long as the resulting code is also copyleft. As a company, I’m sure you don’t want to be exposed to potential infringement here, so it’s nice to know that Tabnine’s models will never generate code that can fall within this kind of license.

The AI models can be adapted to your own codebase, combining general coding practices with the ones preferred by your organization. There are plenty of security features to keep your data safe, with deployment options that range from a secure SaaS to on-premise. To top it off, Tabnine Chat beta can answer all your technical questions, grounded on your own data and on the best coding practices.

These features are shiny, especially for larger organizations. Offering all of this is surely expensive, which may explain the limited free plan that only offers two-to-three-word code completion.

For coding autocomplete

Model: Proprietary

Another option with great online reviews and a generous free plan for individuals, Codeium does a bit more than completing your code. It has a chatbot that you can use to scope projects, ask to explain code, and get improvement suggestions. A programming language polyglot supporting more than 70 languages, integrating with over 40 IDEs, Codeium is another solid app to consider if you’re a coder.

Which AI chatbot should you use?

As ChatGPT itself would tell you, “The answer to this question really depends on what you want to use the chatbot for. There are many different AI chatbots available, and each one has its own strengths and weaknesses.” Nailed it.

The biggest thing to remember is that most of these AI chatbots use the same language model as ChatGPT, and the ones that don’t sound pretty similar anyway…at least if you squint. Most of the differences are in how the apps are to interact with, what extra features they offer, and how they connect to the other tools you use. Almost all of these AI chatbots are free to test, so take a day and give them all a spin. At the very least, it’ll be an experience.

Related reading:

This article was originally published in April 2023. The most recent update was in August 2024.