Perplexity AI Review: Unveiling the Pros and Cons for 2025

Rating: ☕️☕️☕️ (3 Shots) - Innovative but inconsistent—great potential, yet not ready for heavy professional use.

In 2025, artificial intelligence is transforming how we search for information, create content, and manage our daily workflows. Perplexity AI emerged as a promising all-in-one solution, bundling access to multiple advanced language models—GPT-4o, Claude 3.7, 03-mini, DeepSeek R1, and others—into one subscription for just $30 AUD a month. But after two months of heavy use, I’ve found that beneath the innovative packaging lie a host of issues that range from context limitations to data privacy concerns. In this review, I’ll delve into my experiences, supported by the feedback of others, and explain why, despite its potential, Perplexity AI might not yet be ready for prime time.

What Is Perplexity AI?

Perplexity AI positions itself as an advanced answer engine that goes beyond traditional search. Rather than returning a list of links, it uses a conversational approach to provide direct, comprehensive answers. Its model-agnostic platform gives users access to a range of sophisticated AI models with a single subscription—a feature that, on paper, is both innovative and cost-effective. However, this approach also means that every query, file upload, or command is processed through Perplexity’s own system before reaching the final model, adding layers that can compromise performance, privacy, and reliability.

The Context Window Challenge

One of the most significant limitations I encountered was Perplexity’s context window. Although the platform once boasted a “1 million token” capacity, this was quietly reduced—an admin even admitted on Discord, saying, “Yep, we reduced that. We definitely shouldn’t have done so without informing everyone first.” For everyday use, a limited context may be acceptable, but for someone like me who relies on combining multiple documents and large datasets, the window closes on me far too quickly. The system routinely forgets earlier input, often prioritising the most recent data, which is a major frustration when precision and consistency are paramount.

Spaces: A Feature With Untapped Potential

Perplexity’s Spaces feature is designed to allow users to create custom, context-rich environments—a bit like having your own mini AI assistant with persistent memory and personalized instructions. In theory, this is a powerful tool. In practice, however, the feature suffers from two key issues:

Perplexity AI’s Spaces feature allows users to create customized, context-rich environments for tailored AI interactions. However, onboarding and file-sharing limitations hinder its full potential.

1. Poor Onboarding and Explanation: I didn’t even know Spaces worked like a custom GPT with persistent context until a friend mentioned it. The platform provides minimal in-app guidance, leaving many users unaware of its full potential.

   
   

2. File Limitations and Link Sharing Woes: Each Space can hold up to 50 files—a generous limit until you hit it and must decide what to keep. Worse still, sharing links (like Google Drive or Dropbox) within Spaces is virtually useless. Every time I tried, Perplexity claimed it couldn’t access the file. The only workaround was to use public sites like Pastebin, which risks exposing sensitive information. Imagine trying to share an internal memo or a confidential note without compromising privacy.

   
   

Additionally, when I attempted to merge content from multiple documents by copy-pasting them into one prompt, Perplexity’s handling of the input was suboptimal. Whereas models like Claude 3.7 directly accessed would blend the inputs seamlessly, Perplexity stubbornly acknowledged only the most recent addition—rendering the effort to consolidate data either painfully manual or error-prone.

Content Moderation and Inference Time

Perplexity’s content moderation filters present another set of issues. For instance, I experienced the rejection of benign images—a shirtless male figure was rejected, and an image of a woman with just a hint of cleavage was also refused. More bafflingly, even AI-generated images of nuclear explosions were flagged, despite the underlying models accepting such content when accessed directly. This hyper-strict moderation adds extra inference time (the delay between sending a request and receiving a response) that further disrupts the workflow. In other words, you’re left waiting not only for your answer but also for the system to filter out content in a manner that feels both overzealous and inefficient. If you already have a preferred model, the extra waiting and hassle might not be worth it.

Data Collection and Privacy Implications

Perplexity AI’s settings panel offers access to various advanced AI models and image generation tools, but users must weigh privacy concerns against customization options.

Like many modern AI platforms, Perplexity offers personalization settings that ask for your preferences to tailor the experience. Both Claude and ChatGPT have similar features, but with Perplexity acting as a third party between you and the AI models, there’s an added layer of data collection. This raises a critical question: How much are we comfortable with Perplexity knowing about us? When you interact directly with a model, you’re aware of which company is handling your data. With Perplexity, however, your information passes through their servers as well as the model’s—potentially doubling the exposure.

Perplexity’s privacy policy indicates that user data is retained and used to improve AI performance. While this is standard practice, it means that every search, uploaded file, and personalized setting becomes part of their dataset. Given the sensitive nature of some queries, it’s a pressing concern that users should consider carefully.

iOS App Experience and Overall Stability

On the iOS app, the experience is marred by frequent stability issues. If you background the app for even a few minutes, your active chat sessions are lost—something that neither ChatGPT nor Claude suffers from. Moreover, the platform’s frequent updates often feel like live tests rather than polished improvements. Downtime, disappearing chats, and misbehaving features have repeatedly disrupted my workflow. When reliability is key, especially for daily productivity, these issues become a significant deterrent.

The Comet Browser and Possible TikTok Acquisition

Perplexity’s ambitions extend beyond just search. The development of the Comet browser—a Chromium-based browser that integrates agentic AI—promises automated web tasks like booking reservations and ordering tickets. However, this innovation brings its own challenges:

• Chromium Limitations: Relying on Chromium means dealing with Manifest V3, which significantly limits the functionality of popular ad blockers like uBlock Origin. Users who value privacy and a clutter-free browsing experience may find this unacceptable.

  
  

• Privacy Concerns: A browser that accesses live data through Perplexity servers can potentially collect sensitive information, from bank details to personal identifiers. The stakes are high when your daily digital life is mediated through such systems.

Comet, Perplexity AI's upcoming browser for agentic search, promises intelligent browsing and automated web tasks. However, its reliance on Chromium raises questions about privacy and ad-blocking limitations.

Additionally, Perplexity’s expressed interest in acquiring TikTok—a platform with over a billion users—raises further questions about data accumulation. While the promise to open-source TikTok’s algorithm is intriguing, it also implies that Perplexity could gain unprecedented access to user data, compounding existing privacy concerns.

Conclusion: Is Perplexity AI Ready for Prime Time?

Perplexity AI is undeniably a powerful tool with innovative ideas. Its bundled access to multiple language models at a competitive price is a strong selling point, and features like Spaces and the potential integration of a browser indicate bold ambitions. However, persistent issues—ranging from restrictive context windows and inefficient content moderation to stability problems and data privacy risks—have made it less reliable for heavy, professional use.

For now, I find myself returning to more stable platforms like ChatGPT and Claude, which, while not without their own shortcomings, offer a smoother, more predictable experience. Perplexity shows promise, but until it resolves these critical issues, it remains more of an intriguing experiment than a dependable daily driver.

If you’re considering Perplexity AI, weigh the innovative potential against the practical challenges. As the platform evolves, I hope to see these issues addressed, paving the way for a tool that can truly deliver on its ambitious promise.