Perplexity
The best first-pass research layer for cited answers — treat citations as leads not proof, and expect quotas and models to shift under you.
Element scores
Strengths
Citation-first answers remain the fastest way to do sourced first-pass research — independent evaluations consistently rank its citation density and incorrect-citation rate best-in-class among AI search engines. €20/month buys access to multiple frontier models plus in-house Sonar. The free Comet browser (10/2025) and Perplexity Computer (02/2026) show real product velocity, and the developer platform grew into Search, Agent and Embeddings APIs with transparent per-request cost reporting.
Honest dings
Citations are a verification mechanism, not a guarantee: a 500-query independent test found only 83.3% of citations actually supported the claim (health queries: 68.5%), and the Tow Center found answers wrong in roughly 37% of cases. Trust took real hits in 2025–26: the model-substitution incident, Pro search quotas cut 600→200/week and Deep Research 50→20/month without announcement, and heavy legal exposure (Reddit and NYT copyright suits, Amazon v. Perplexity over Comet's agentic shopping). The flagship agent features are locked behind the ~€185/month Max tier.
Sources
Every audit lists the research it rests on — transparency and traceability are the product. Tools evolve: each audit is a snapshot of its audit date, and re-audits supersede older versions (kept below for reference).
- perplexity.ai/hub/blog/introducing-perplexity-compu… — Official Computer launch (25/02/2026): sub-agents, long-running tasks, Max-exclusive, multi-model orchestration (accessed 2026-07-10)
- perplexity.ai/hub/blog/comet-is-now-available-to-ev… — Official: Comet browser free worldwide (10/2025) (accessed 2026-07-10)
- perplexity.ai/changelog — Official changelog: Comet iOS (03/2026), Computer in Microsoft 365 (05/2026), analytics dashboards/APIs (05–06/2026) (accessed 2026-07-10)
- docs.perplexity.ai/docs/getting-started/overview — Official API platform surfaces: Agent API, Search API, Embeddings API (accessed 2026-07-10)
- docs.perplexity.ai/docs/getting-started/pricing — Official API pricing: Sonar $1/M tokens + $5–12/1k requests, Sonar Pro $3/$15 + $6–14/1k; per-request cost breakdown in responses (accessed 2026-07-10)
- vantaige.io/ai-tool/perplexity — Independent review (04/2026): citation fabrication cases, Nov 2025 model-substitution incident, silent quota cuts (600→200 searches/week, Deep Research 50→20/mo), lawsuits, PDF >10 pages failures (accessed 2026-07-10)
- fonzy.ai/blog/does-perplexity-cite-sources — Independent 500-query citation test: 83.3% citation accuracy overall, 68.5% on health; 6% of answers uncited (accessed 2026-07-10)
- techsifted.com/reviews/perplexity-ai-review-2026 — Independent review (04/2026): strengths in sourced research, weakness on niche/thin-source topics and citation mismatches (accessed 2026-07-10)
- techtimes.com/articles/318028/20260608/perplexity-r… — $200M raise at ~$20B valuation (05/06/2026); Comet timeline; Comet Plus $5/mo publisher revenue-share tier (accessed 2026-07-10)
- gradually.ai/en/perplexity-statistics — Third-party stats compilation (04/2026): no official user numbers published; 34M MAU estimate; subscription-only since 02/2026; $22.6B Jan-2026 valuation mark (accessed 2026-07-10)
- adwaitx.com/perplexity-vs-chatgpt-research-accuracy… — Independent comparison (01/2026): Stanford-based hallucination benchmarks (Perplexity Deep Research ~26.6% vs ChatGPT-4o ~39.1% fabricated references); Tow Center ~37% incorrect answers (accessed 2026-07-10)