Perplexity

Perplexity is the best-positioned tool in the category for research tasks where current, sourced information matters — but it is not a general assistant, and citation accuracy requires verification rather than implicit trust.

Perplexity's capability is structured around a specific value proposition: every answer is grounded in live web sources with inline citations visible alongside each claim. This is the operating mode, not a feature to enable. The practical effect is that Perplexity stays current on topics that other assistants get wrong after their training cutoff — recent events, current pricing, evolving research. Model flexibility is genuine: users can route queries through the current flagship model, Claude Opus, Gemini, or Perplexity's own Sonar models in the same session. Comet browser — agentic web browsing — is free for all users . The documented gap is citation accuracy: multiple independent reports have found claims misattributed to cited sources. The citations look authoritative but require spot-checking before use in anything consequential.

Perplexity's recent years web scraping controversy and absence of consumer training opt-out create a murkier privacy picture than the clean research-first product image suggests — enterprise customers have clearer protections than individual users.

Perplexity's privacy picture has two layers. The product layer: query data is collected and may be used to improve services, with no consumer opt-out documented. Enterprise Pro excludes query data from training, but standard tiers don't have an equivalent commitment. The reputational layer: recently, Wired and Forbes published reports of robots.txt violations and content aggregation practices that Perplexity disputed but didn't fully resolve publicly. The EU AI Act GPAI compliance statement hasn't been published ahead of the an upcoming regulatory deadline deadline. For individual research use, these concerns are unlikely to affect daily experience. For organisations or journalists using Perplexity for sensitive research, the data collection posture and unresolved publisher disputes are worth factoring in.

Perplexity's trust score reflects ongoing publisher disputes and the absence of a transparency report — the platform works well as a product, but the legal and ethical questions around its content sourcing practices remain open.

Perplexity's trust profile has a specific wrinkle that most AI assistants don't have: the recent years publisher disputes. Multiple publishers filed complaints over content aggregation practices — the accusation being that Perplexity was scraping content, paraphrasing it, and presenting it with citations in ways that bypassed publisher paywalls and reduced direct traffic. Perplexity disputed the characterisation but the legal and ethical questions weren't cleanly resolved. SoftBank is a major investor, which creates alignment considerations with SoftBank's broader portfolio. No transparency report on government data requests has been published. The citation misattribution issue documented independently — claims attributed to sources that don't support them — is a trust concern specific to a platform whose core value proposition is sourced answers.

Perplexity's ecosystem is narrower than ChatGPT's but purpose-built — the Sonar API is genuinely useful for developers who need search-grounded responses, and Comet adds an agentic layer that most competitors don't offer for free.

Perplexity's ecosystem is purpose-built rather than broad. Browser extensions for Chrome and Firefox bring the search-first interface into everyday browsing. The Sonar API is the most distinctive developer offering: search-grounded responses via API at $1/$1 per a very large context window for Sonar Large, priced for high-volume research applications. Comet browser adds agentic task execution — browse, book, multi-step research — free for all users. Spaces provide persistent instructions and file context for Pro users. The ecosystem doesn't try to be ChatGPT's GPT Store equivalent: there's no plugin marketplace and the integration surface is narrower. What exists is specifically designed to support the research and sourced-answer use case rather than general-purpose task handling.

At the standard paid plan, Perplexity Pro delivers real-time sourced answers plus access to multiple frontier models — strong value for users whose primary need is research rather than general-purpose assistance.

At the standard paid plan, Perplexity Pro delivers something distinctive: unlimited access to multiple frontier models (the current flagship model, Claude, Gemini) through one interface, with real-time search grounding on every response. Users who would otherwise pay the standard paid plan for ChatGPT Plus and want to access Claude on the side effectively get both through Perplexity Pro, plus the sourced research capability that neither provides natively. The Max tier at the maximum plan tier — unrestricted frontier model access — is positioned for power users who need maximum context and capability across multiple models without managing separate subscriptions. The free tier is genuinely useful: unlimited Sonar searches with limited Pro model access per day covers casual research needs without a commitment.

Perplexity is reliable for what it's designed to do — real-time research — but citation misattribution means outputs require verification before use in anything consequential, which partially undermines the speed advantage.

Perplexity's operational reliability is good for its core function. The public status page at status.perplexity.ai provides incident history — a transparency measure that matters when real-time data retrieval is your core value. Real-time web retrieval means responses stay current for time-sensitive topics without depending on training data freshness. The documented reliability gaps are quality-related rather than availability-related: citation misattribution is the most significant — claims can be attributed to sources that don't support them, which requires verification before downstream use. Output consistency varies by underlying model selected: switching between Sonar, the current flagship model, and Claude produces meaningfully different response styles and quality levels for the same query.