Perplexity: An Honest User Overview
I’ve been using Perplexity as my primary research tool for about six months, after bouncing between Google, Bing, and ChatGPT for deep-dive queries. Here’s what I’ve learned, without the hype.
What It Does Well (With Real Examples)
Perplexity is an AI-powered search engine that gives you direct answers with sources. Unlike Google, which buries you in links, or ChatGPT, which often makes up citations, Perplexity shows you the exact paragraph it pulled the answer from.
Example 1: Fact-checking a claim.
I asked: “Did the FDA approve a drug for Alzheimer’s that was later withdrawn?”
Perplexity returned: “Yes, Aduhelm (aducanumab) was approved in 2021, but Biogen withdrew it in 2024 after Medicare coverage restrictions and low sales.”
It cited: a Reuters article (2024), an FDA press release, and a JAMA study. Each source was a click away. I could verify within seconds. Google would have given me 10 pages of SEO-optimized blog posts. ChatGPT would have given me a confident answer with no way to check.
Example 2: Technical comparison.
I needed to compare “PostgreSQL vs. MongoDB for time-series data.”
Perplexity gave me a table: scalability differences, indexing limitations, and real-world use cases (e.g., “MongoDB’s time-series collections are better for IoT sensor data, but PostgreSQL’s TimescaleDB extension beats it for SQL-based analytics”). It pulled from official docs, Stack Overflow, and a few engineering blogs. I saved 30 minutes of manual cross-referencing.
Example 3: Research with a messy question.
I typed: “What’s the latest on the EU AI Act’s risk categories for open-source models?”
Perplexity summarized the four-tier system (unacceptable, high, limited, minimal), noted that open-source models fall under “limited risk” unless used in high-risk contexts, and linked to the actual EU legislative text (2024 version). No fluff.
Limitations (The Grit)
Hallucinations still happen.
Perplexity is better than ChatGPT, but it’s not immune. I asked once: “Who invented the first electric car?” It claimed “Thomas Davenport in 1834” (correct) but then added “and the first production electric car was the 1996 GM EV1” (partly correct, but the EV1 was a lease-only, not a production car for sale). The source link was a Wikipedia snippet that actually said “limited production.” The summary overgeneralized. Always double-check.Source quality varies.
For niche topics (e.g., “19th century Japanese textile dyes”), it often pulls from Reddit threads, personal blogs, or outdated forum posts. The algorithm prioritizes any page with relevant text, not necessarily authoritative sources. You have to manually filter.Context window is limited.
Free users get about 3000 tokens per query. If you paste a 20-page PDF or ask a multi-part question like “Compare the economic policies of Sweden, Norway, and Denmark from 2000-2020,” it truncates the answer or misses details. Pro users get 5000 tokens, but it’s still not enough for large documents.No real-time updates for some categories.
Stock prices, sports scores, or breaking news can be stale by a few hours. It says “last updated” but doesn’t always refresh automatically. For live data, use a dedicated news feed.
Key Workflows
- Quick fact-check: Type a question, scan the answer, click the source to verify. Takes 10 seconds.
- Research summaries: For academic papers, I copy-paste the abstract or a paragraph, and ask Perplexity to “explain this in plain English” or “list the key findings.” It works better than ChatGPT for retaining citations.
- Coding help: I use it to debug error messages or compare libraries. Example: “Why does my Python script throw a MemoryError when loading a 10GB CSV?” It gave me chunking strategies (pandas.read_csv with chunksize) and linked to a real Stack Overflow thread.
- Product comparisons: For buying decisions (e.g., “Kindle vs. Kobo vs. Boox for note-taking”), it aggregates reviews and specs from multiple sources. The table format is excellent.
Pricing Reality
- Free tier: Unlimited queries, but you get 5 “Pro” searches per day (which use GPT-4 or Claude 3.5). After that, it defaults to a smaller model (likely GPT-3.5 or a fine-tuned Llama). Speed is fine, but answers are less nuanced. You also can’t upload files (PDFs, images) on free.
- Pro tier ($20/month): Unlimited usage of GPT-4, Claude 3.5, and Perplexity’s own “Sonar” models. You can upload up to 20 files per day (PDFs, images, text). You also get priority access to the “Deep Research” mode (which takes 2-5 minutes to generate a report with 10+ sources).
Is it worth it? If you’re a student, journalist, or researcher doing 10+ fact-checks a day, yes. If you just want to replace Google for casual browsing, the free tier is fine.
Who Should Use It
- Researchers, students, and journalists: For verifying claims, finding sources, and summarizing papers. The citation system alone saves hours.
- Developers and analysts: For technical comparisons, debugging, and exploring documentation.
- Anyone tired of SEO garbage: If you hate wading through listicles and affiliate links, Perplexity gives you direct answers with real sources.
- Not for: Casual browsing (e.g., “what’s the weather?” or “funny memes”), real-time updates (sports, stocks), or creative writing (it’s not a generative tool like ChatGPT; it’s a search engine).
Final Honest Verdict
Perplexity is the best tool I’ve found for verifiable research. It’s not perfect—hallucinations, source quality, and context limits are real issues—but it’s a massive upgrade over traditional search for anyone who needs accurate, cited answers fast. If you’re willing to double-check sources and accept its quirks, it’s worth using. If you want a magic answer machine that never makes mistakes, keep waiting.
