Claude Review: An Easy-to-Use AI Chatbot That Emphasizes Privacy
PCMag editors select and review products independently . If you buy through affiliate links, we may earn commissions, which help support our testing .
Anthropic's Claude stands out for its privacy-first approach in a space full of competitors that rampantly collect and train on your data. But Claude is more than its privacy policy: It has competent complex reasoning, creative writing, file processing, and web searching skills. Claude's sleek user interface and impressive (albeit in development) app maker are also points in its favor. However, missing features (such as image and video generation), short deep research reports, and unexceptional integrations are all drawbacks. Claude is still worth exploring if you want a chatbot that's enjoyable to use and respects your privacy, and it places at the top of our . However, remains our Editors' Choice winner, thanks to its more comprehensive feature set and peerless response quality.
What Is Claude?
Claude is an AI chatbot, like ChatGPT and , that you can chat with over text. You can use Claude to answer questions, do research, write creatively, search the web, solve equations, and many other things. Put simply, Claude is an advanced virtual assistant.
You can use Claude to code, and Anthropic even has an agentic coding tool, , that lives in your terminal and understands your codebase, learning how to help you code faster over time. That tool is outside the scope of this review, but you can still .
I think AI chatbots work best for answering questions and searching the web. Depending on what you're looking for, it can be easier and quicker to simply ask a chatbot like Claude instead of Googling something and scrolling through search results. Claude's deep research might be a better jumping-off point for learning about certain complex topics, but more on that later.
Regardless of what you use a chatbot for, and which chatbot you use, it's important to keep in mind that they are fallible. Chatbots can and will confidently provide incorrect information, so you shouldn't take everything one says at face value. Make sure to check anything you learn from a chatbot against a reputable source if it's important.
How Does Claude Work?
At the most basic level, Claude takes in prompts and puts out responses. It does this via made up of trained on massive amounts of data. These models are complicated algorithms that give Claude the ability to access information on everything from Peruvian cuisine to Soviet architecture and ultimately answer your questions. Claude can also search the internet for knowledge on current events.
Claude isn't like traditional software: Anthropic doesn't need to update it for it to get better. Much like a person, Claude can learn and improve over time. Simply by talking to Claude, you help to train its underlying models, assuming you give Anthropic permission to use your data to do so first. However, don't expect Claude to change from chat to chat—it's a gradual process.
Claude uses two primary lines of models: Opus and Sonnet. Sonnet is Claude's conversational, flagship line, while Opus specializes in complex reasoning, making it a good fit for things related to coding, math, and science. Opus 4 and Sonnet 4 are the latest versions of Claude's primary models, though you can also access Opus 3 and Sonnet 3.7 with a paid subscription. Claude has a third model line, Haiku, that delivers responses faster than Opus or Sonnet at the expense of accuracy and detail. My testing focuses on Sonnet 4 (Claude's default model) and Opus 4.
Pricing: Slightly Cheaper Than Competitors
Claude has a free version, but premium plans increase usage limits and unlock more features.
For free, you can use Claude's Sonnet 4 model to answer questions, generate code, make apps, process files, search the web, or visualize data, among other things. A Pro plan ($17 per month, billed annually) gives you access to Claude in your terminal, deep research, extended thinking for complicated tasks, higher usage limits, integrations with Google Workspace , models beyond Sonnet 4, and unlimited Projects for organizing chats and documents.
The top-tier Max plan gives you priority access to Claude during high-traffic times and even higher usage limits. Max comes in two flavors: 5x more usage than Pro ($100 per month, billed annually) and 20x more usage than Pro ($200 per month, billed annually). You also get early access to new features, but none were available at the time of writing. The vast majority of people interested in premium plans, though, should opt for the Pro plan, which has generous usage limits. For this review, I primarily tested the Free and Max 5x tiers.
For chatbots, $20 per month is the standard fare for a flagship premium plan. Thus, Claude Pro's $17-per-month rate offers some savings. The top plans from ChatGPT ($200 per month) and Gemini ($250 per month) are in line with Claude's most expensive Max plan, but Claude gives you the option to sign up for a cheaper version of Max, which I appreciate. However, Claude's free plan lacks a lot of the functionality you get with ChatGPT and Gemini, including deep research, image and video generation, model switching, and voice chat. In fact, Claude doesn't support image and video generation at all.
Where Is Claude Available?
Claude is accessible as a mobile app (available for Android and iOS) and on the web. It doesn't have an official browser extension, but unofficial ones are available to download. You can also get desktop versions of Claude (available for macOS and Windows), but these apps are currently in beta, so don't expect flawless performance just yet.
Other services and sites use Claude's models, such as Perplexity, but Anthropic doesn't develop, operate, and own these iterations, meaning they aren't part of the official Claude package. Regardless of where you see Claude's models pop up, stick with the official apps and site if you want to access its full set of features.
Ease of Use and Interface: An Elegant Design
To start using Claude, you need to create an account, which requires an email address and a password. Signing in brings you to your dashboard. Claude has the same basic design as pretty much every chatbot: A text field for starting chats sits in the center alongside a left-hand sidebar that shows your conversation history. Buttons below the central text field provide ideas on what exactly you can use Claude to do, from vibe coding to writing a case study.
Claude's core design might not be vastly different from other chatbots, but I enjoy interacting with it more. Its color scheme, font, and slightly blockier aesthetic are pleasing and make it feel less sterile than ChatGPT or Gemini. Even across deep research progress screens and settings pages, Claude's interface is cohesive. It is consistently tidy and readable.
From your dashboard, you can talk to Claude about anything. Responses aren't quite as snappy as with Gemini, but they aren't as consistently slow as with . Once you get a response, you can copy or regenerate it. An easily accessible share button is at the top of the screen. Sometimes, responses hang or encounter errors during generation. However, this isn't a major problem with Claude, and all chatbots trip up occasionally.
, but Claude's tone, by default, is closer to Gemini's. In other words, you can expect more direct, slightly stiffer responses. However, you can customize Claude's personality to your liking, like you can with ChatGPT. Claude calls these styles, and a variety are available to choose from, such as concise or formal. You can also create a custom style simply by describing how you want Claude to communicate or providing an example. That's not the end of the customization options, either. You can tell Claude to keep any manner of personal preferences in mind or to refer to you by a certain nickname, too.
When it comes to memory, Claude's doesn't persist across chats, like it does with ChatGPT and Gemini. As a result, you have to remind it of prior conversations if you want to pick up where you left off. Claude can remember certain things you tell it, like the aforementioned personal preferences, but that's it. The lack of persistent memory makes for an overall less satisfying chatting experience, especially if you talk to Claude about many topics.
Like DeepSeek, Claude doesn't support speech-to-text or voice chat on the web. However, like Gemini, Claude does offer a voice chat mode for its mobile apps. You can choose from a variety of lifelike voices, albeit ones that don't sound quite as natural as Copilot Voice , and converse about all the same topics you can over text, including files you upload.
However, Claude's voice chat mode is uncharacteristically clunky. It isn't free-flowing, like with Gemini, meaning you (typically) need to click send after speaking. Claude sometimes responded to me before I sent my message, however, which made for an inconsistent experience. The microphone is always active in voice mode, too; it doesn't even turn off if you minimize the app.
Web Search: Competent Responses
Chatbots don't have to rely just on their initial training data. Now, every chatbot can search the web, including Claude. Accordingly, Claude, ChatGPT, and Gemini had little trouble answering questions about current events at the time of testing, like what happened in "The Beginning After the End" finale or the latest news on the conflict between Iran and Israel. ChatGPT managed to tell me which Incarnon weapons were in Warframe 's current weekly rotation, whereas Claude and Gemini could not. That said, failing to respond correctly to web search prompts is a rarity with both Claude and Gemini.
It's still possible to trip Claude up sometimes. For example, it told me the current pope was Pope Francis, even though I made sure to toggle the live web search option. I had to explicitly tell it to search the web to get the correct answer. With ChatGPT, you just need to be careful to ask for the current pope or toggle the web search option before you ask your question. Otherwise, ChatGPT tells you it's Pope Francis, courtesy of its outdated training knowledge.
Claude is slightly slower than competitors at searching the web. Its responses aren't quite as detailed as ChatGPT's, and its sourcing isn't as robust, either: ChatGPT highlights the relevant portion of its response when you hover over a source, which makes it easy to connect claims to evidence, and its source icons display the name of the source. Claude doesn't do in-text highlighting, and its source icons show the name of the attached article, which is less useful at a glance.
Gemini can show you an image if you ask for one, while ChatGPT includes images in its responses automatically when relevant. Claude, on the other hand, doesn't display images, even if you ask.
You can ask Claude for buying advice, but it doesn't have a shopping component. ChatGPT and Gemini can provide clickable tiles with links to products and other relevant information.
Deep Research: Speed Over Depth
I love using chatbots' deep research features: Ask a question, and you'll get a comprehensive report complete with dozens of citations. I've used chatbots to research everything from how to translate controller settings between games to which egg brands have the lowest incidence rates of salmonella.
Claude calls its deep research feature 'Research.' Unlike ChatGPT's and Gemini's equivalent features, Claude's requires a paid subscription. Copilot is the same way. ChatGPT and Gemini merely limit how much deep research you can do for free.
Like ChatGPT, Claude asks follow-up questions on research prompts before it begins. Gemini, on the other hand, won't ask questions, instead providing a research plan you can tweak. In general, I prefer ChatGPT's and Claude's approach, as they're easy to answer and often provide essential context to the research at hand that Gemini sometimes struggles without.
Like other chatbots, Claude handles straightforward research queries with ease, such as providing or . However, chatbots can struggle when researching complicated topics with sources that conflict with one another. As such, I tasked , , and to research Valkyr, a freshly updated character in Warframe that the community doesn't fully understand yet.
Claude's research reports routinely have less detail than those from ChatGPT and Gemini, and this was the case here, too. However, Claude did do many more searches (318) than ChatGPT (93) for this query, citing just a handful of them and providing less detail. For the information it does provide, Claude's accuracy is similar to ChatGPT and Gemini. Perhaps due to the length of its reports, Claude often researches topics quicker than ChatGPT and Gemini, coming in at under 10 minutes in most cases.
Claude's deep research interface is pleasant to use. Like ChatGPT, it provides in-text links. These are more readable at a glance than Gemini's source tiles, which it hides behind clickable carets. Also like ChatGPT, Claude displays useful stats, such as the total research time, total searches, and total sources; Gemini doesn't do the same. Claude has a dedicated screen that breaks down its searches by website into a bar graph, too, which is nicer than the simple lists of searches and sources you get with ChatGPT and Gemini. I did miss Gemini's one-click button that exports reports to Google Docs and how ChatGPT displays reports in its central chat window, instead of relegating them to a sidebar, though.
In terms of tone, Claude's reports read similarly to Gemini's: like formal papers. ChatGPT's reports read more like forum posts. I prefer ChatGPT's tone here, just like I prefer ChatGPT's tone in general. However, you might not, depending on your research topic.
File Processing: Better With Documents Than Images
Claude can process files you upload, which means it can analyze and understand documents and images to critique resumes , evaluate fashion choices on video game characters, identify plants and animals, translate text, or do a million other things. Image recognition and complex document processing, though, are tough tasks for chatbots, so I tested them.
To start, I provided an image of my computer (pictured below), which includes ample RGB lighting and reflections that might trip up a chatbot. I asked Claude (Sonnet 4), ChatGPT (GPT-4o), and Gemini (2.5 Flash) to identify as many components as possible with as much specificity as they could muster. I also asked them not to reference prior conversations in their answers since I've run similar tests before.
All of the chatbots got some things wrong and were routinely vague in their identifications. However, Claude had the least detailed response of the bunch. ChatGPT impressively managed to identify my Lian Li case, ASUS ROG Hero motherboard, and D5 pump. Gemini gave a longer and more descriptive response than ChatGPT, but it wasn't able to pick out any components, missing a core part of my prompt.
Images aside, how well does Claude handle processing documents? I uploaded the manuals for my case, CPU waterblock, and fan controller to Claude (Sonnet 4), ChatGPT (GPT-4o), and Gemini (2.5 Flash). Then, I gave them the following prompt: "Based solely on the provided manuals, what RGB connections do my CPU block and case have, and can they connect directly to my Octo fan hub?"
All the chatbots were able to conclude that my water block RGB could work with my fan controller if I had the appropriate adapter, as well as that my case (rather than my fan controller) could control the RGB lighting. Both findings are true. However, Claude stood out by identifying the part numbers for possible adapters I could use, something the others didn't do. Meanwhile, ChatGPT gave me further commentary on how I might connect my case RGB to my fan controller, while Gemini provided more quotes and citations from the manuals than the others.
ChatGPT had the most detailed response in my test, but all three performed well. Nonetheless, I still don't recommend uploading your homework for Claude to do for you: It can sometimes, like ChatGPT and Gemini, make up quotes and misunderstand documents. As always, make sure to verify the answers you get about something important.
Creative Writing: A Strong Performer
Chatbots can do all manner of creative writing, whether that means preparing a speech for a conference, telling a joke, or writing a personalized message. Of course, chatbots can tell stories, too, but as they get more advanced and their underlying technologies mature, evaluating their creative writing abilities goes beyond whether they can generate something coherent.
I focused my creative writing test on poetry, which requires careful attention to detail across not just content but also punctuation and style. I gave the following prompt to Claude (Sonnet 4), ChatGPT (GPT-4o), and Gemini (2.5 Flash): "Without referencing anything in your memory or prior responses, I want you to write me a free verse poem. Pay special attention to capitalization, enjambment, line breaks, and punctuation. Since it's free verse, I don't want a familiar meter or ABAB rhyming scheme, but I want it to have a cohesive style or underlying beat."
Claude's and ChatGPT's poems meet my specifications, opting for fairly cohesive free verse and lowercase styles complete with a variety of punctuation. Their enjambment and line breaks don't feel random, either, and they don't read like prose. I also appreciate how both Claude and ChatGPT include titles for their poems, even though I didn't ask for them. Gemini, on the other hand, doesn't include punctuation beyond commas and periods. Its poem also reads quite similarly to .
Aside from the poem, Claude also included a brief message describing its poem, explaining why it made the choices it did in service of fulfilling the requirements of my prompt. While ChatGPT and Gemini display their poems in simple text, Claude presents its poem in a clickable Artifact tile: This opens up the poem on the side of the screen, allowing me to easily read it and refer to Claude's accompanying message. The poem's formatting remains consistent across different screen resolutions and window sizes, which I absolutely love as a writer.
Complex Reasoning: Extended Thinking Is a Must
Chatbots aren't just creative: They can solve complicated problems, too. To test this ability, I gave Claude, ChatGPT, and Gemini exam questions from undergraduate courses in computer science, math, and physics from , , and . Afterward, I compared their responses to the solutions. I used each chatbot's complex reasoning model: Claude's Opus 4, ChatGPT's o3, and Gemini's 2.5 Pro. If you sign up for a premium Claude subscription, you can enable extended thinking, which is ideal for difficult-to-handle prompts. I tested Claude's complex reasoning with and without the extended thinking mode active.
Claude got nine math questions wrong without extended thinking, which shrunk down to zero with extended thinking. It managed to answer every computer science question correctly, even without extended thinking. Physics was more challenging for Claude, as it answered three questions wrong, even with extended thinking mode active. For comparison, ChatGPT got one math question and two physics questions wrong, while Gemini got a part of a math question, a part of a physics question, and three parts of computer science questions wrong. Altogether, that adds up to three questions wrong for Claude with extended thinking, three questions wrong for ChatGPT, and five parts of questions wrong for Gemini.
Although Claude managed to keep pace with the other chatbots, extended thinking mode is a must, considering it can make significant mistakes without it. Still, this isn't a perfectly consistent feature: Claude (rarely) got some questions wrong with extended thinking active, and then answered them correctly with that feature off. Since all the chatbots I've tested are imperfect when it comes to complex reasoning, I don't recommend using any of them to help you study without verifying the information they provide.
Integrations: Gemini Lite
Like Gemini, you can connect Claude to Google Workspace apps, such as Calendar, Drive, and Gmail. These integrations are currently in beta, so they will likely improve over time. Whereas Gemini pops up in these apps directly, Claude has you access them through its interface. Since Google develops both Gemini and the Google Workspace suite, its integrations are understandably deeper and more robust; Gemini also works with substantially more apps in the suite. However, Claude can do some of the same things as Gemini in Google Workspace apps it supports.
For example, if you connect your Gmail, you can ask Claude to list your unread emails or find a particular email that a keyword search isn't surfacing. Claude can even sometimes do things Gemini is reluctant to do, such as analyze my inbox for emails I should unsubscribe from. On the other hand, Claude's analysis of my email often feels superficial and biased toward newer messages. Gemini doesn't have this issue. In fact, . Claude isn't nearly as invasive, but it's also not as powerful.
A big part of Gemini's effectiveness comes down to its presence directly within Google Workplace apps. Drafting an email in Gmail itself with Gemini is a lot more convenient than doing so with Claude. So, while I can prompt Claude about emails in my Gmail inbox, items in my Google Drive, or my itinerary in Google Calendar, I'm not sure why I would when Gemini makes it so much easier.
Of course, you might prefer Claude's approach to privacy, which I discuss in more detail below. But if you're that concerned about privacy, you're better off with a more privacy-conscious than Gmail in the first place, considering Google's fairly sketchy history.
Artifacts: Great Concept, Limited Implementation
When you use Claude to do deep research or generate creative writing, it presents these responses in the form of Artifacts, which are clickable tiles that open in a sidebar. You can view all your Artifacts from Claude's Artifacts tab on the left side of the screen while you use the chatbot, which is a great user experience innovation. ChatGPT and Gemini don't give you an easy way to view all your generated content at once.
However, Claude's Artifacts are much more than simple user interface elements. Artifacts can be apps, which Claude can build for you, even if you don't know anything about programming. Simply describe your app to Claude, and it will make it. You can browse a library of Artifacts to try out before you create one, too. After you generate an Artifact, you can publish and share it. Artifact apps you can try out include flashcard apps, interactive drum machines, quizzes, simple games, writing editors, and much more.
The apps Claude can make are fairly basic web experiences, so you can't have Claude make you a new desktop program or iPhone app. However, what it can do is still quite impressive, if you're willing to put the necessary time into tweaking them. For example, I asked Claude to make me a build calculator for the Warframe character Oraxia. My app allows me to see all her stats and how they change as I set up a build for her. At the time of writing, this little application has more functionality than Warframe's most popular build calculator website, which doesn't yet fully support Oraxia.
However, creating the app was a major hassle. Claude often ran into its context window limit during testing, so I needed to start fresh chats to continue working on it. The Chrome tab I had Claude open in could balloon to 8GB of memory usage, after which it would crash. Asking for a simple change to an app made Claude delete the rest of the code of the app and only present me with the update, unless I forced it to rewrite the entire app with the change included. Claude also ran into frequent errors that required ground-up rewrites, too.
The above won't be the situation for every app, especially less complicated ones, and this feature is brand new at the time of writing, so I expect it to improve. The inefficiencies and issues of Artifact app creation are frustrating, but it's still cool to design custom apps that you otherwise couldn't.
What Can't Claude Do?
Most importantly, . It can't feel anything, think for itself, or understand like a human being can. In other words, it can't be your friend, romantic partner, or therapist. Claude is a complicated algorithm designed to accept prompts and return responses.
Adult content, illegal activity, and taboo subjects like hate speech, among other things, are against Anthropic's policies and . With enough prompt calibration, you likely can get around these built-in restrictions, but Claude didn't violate its own policies during testing. I can't say the same for DeepSeek and Gemini.
Claude also has practical limitations, like its context window. Think of a context window as short-term memory. Its size determines how much data it can process in a single conversation. For free users, the context window is entirely dynamic depending on server load. Paid users get up to a 200,000-token context window, but this is also dynamic. I was able to hit a context window limit when uploading documents as a free Claude user, and also when iterating on an Artifact app as a paid user.
Alongside the context window size are usage limits, which determine how many messages you can send to Claude within a five-hour period. For free users, this number is dynamic. Pro users can send up to around 45 messages, and Max tiers support up to 225 (5x Pro) or 900 (20x Pro) messages. Anthropic dynamically caps paid tiers, too, though, so you might be able to send more messages, depending on server load.
Is Your Data Safe With Claude?
One of Claude's main selling points is its user-friendly approach to . Accordingly, Claude collects contact data when you sign up for an account, payment information if you choose to purchase something, conversation data when you use Claude, and any feedback you give to Anthropic on its services. Claude also receives certain information automatically, such as cookie data, device and connection data, log and troubleshooting data, and usage information (like time and date data, among other things).
By default, Claude encrypts personal user data, and Anthropic employees can't access it outside special circumstances, such as enforcement of . Claude gives you the ability to , and, in general, deletes conversation history automatically from its back end within 30 days. Unlike other chatbots, Claude doesn't use the data it collects to train its models by default, either.
Even though Claude's approach to data collection and privacy is admirable, Anthropic is no stranger to scandals. In 2024, , accidentally sending customer information to a third party, and for using bots to access its site after Anthropic allegedly stopped its bots from doing so. Beyond that, . Anthropic has also come under fire for .
From a data collection and privacy perspective, Claude is the best chatbot you can use right now, but that doesn't mean it (or the company that makes it) is perfect. There's good reason to feel safer sharing information with Claude than with other chatbots, but I advise caution. It's wise to limit how much you share with any online service.
Disclosure: Ziff Davis, PCMag's parent company, filed a lawsuit against OpenAI in April 2025, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.
