I Tested Perplexity, NotebookLM, and Google Gemini for a Month – Here’s the Brutal Truth

I’ve spent the last 30 days living inside three AI tools: Perplexity, NotebookLM, and Google Gemini. I threw everything at them – research for a book I’m writing, planning a vacation, debugging code, summarizing legal documents, and even just casual Q&A. I wanted to know which one actually helps you get stuff done, and which ones are just flashy demos that fall apart when you push them.

Here’s the comparison table to get the lay of the land:

Feature	Perplexity	NotebookLM	Google Gemini
Pricing	Free tier (limited Pro searches), Pro at $20/month	Free (Google account needed)	Free tier (limited), Advanced at $19.99/month
Core strength	Real-time web search with citations	Deep analysis of your uploaded documents	Multimodal (text, images, video, audio)
Context window	~100k tokens (Pro)	~100k tokens (supports up to 50 sources)	~1 million tokens (Advanced)
Internet access	Yes, always on by default	No (only your uploaded content)	Yes, but only if you enable it manually
Citation style	Inline numbers with source links	Audio “deep dive” and written notes with quotes	Links in responses (when internet is on)
Best for	Quick research, fact-checking, competitive analysis	Understanding complex documents, study prep	Brainstorming, creative tasks, multimedia analysis
Worst for	Deep analysis of personal files	Real-time news or anything outside your uploads	Factual accuracy without internet, long-form writing

I’m not going to sugarcoat it. Each tool has a very specific job it does well, and two of them have gaping holes that make them frustrating to use daily. Let me walk you through my actual experience.

Perplexity: The Research Powerhouse That’s Too Smart for Its Own Good

Perplexity is like having a PhD student who’s glued to Twitter and academic journals 24/7. It’s built for one thing: finding accurate, up-to-date information from the web and presenting it with citations. And for that, it’s scary good.

What I loved

I was researching the history of microplastics for a chapter in my book. I typed “How did microplastics become a mainstream concern? Give me the key scientific papers and timeline.” Within seconds, Perplexity gave me a bullet-point timeline starting from the 1970s, linked directly to Thompson et al.’s 2004 paper in Science, the 2018 UN report, and a 2022 study from Environmental Science & Technology. Every single claim had a number next to it. I clicked number 3, and it opened the exact PDF page. No hallucinations. No made-up sources. This is a godsend for anyone who’s been burned by ChatGPT making up fake court cases.

I also used it for competitive analysis. I asked, “What’s the pricing model for Notion’s enterprise tier as of July 2024?” Perplexity pulled the current pricing page, a Reddit thread from two weeks ago complaining about a price hike, and a blog post from Notion’s CEO. It even summarized the differences between the old and new plans. I saved about an hour of manual browsing.

The “Pro” search feature (you get 5 free per day, unlimited with $20/month) is a game-changer. It chains multiple queries together and digs deeper. I asked it to “Compare the battery life of the latest iPhone, Samsung, and Google Pixel phones,” and it ran separate searches for each, then presented a table with real-world test results from different review sites. That’s not just a summary – it’s original synthesis.

Where it falls apart

Perplexity is terrible at handling your own documents. I uploaded a 50-page PDF of a client contract and asked, “What are the termination clauses?” It gave me a decent summary, but when I asked “Compare the indemnification sections of section 4 and section 12,” it got confused. It kept referencing the wrong paragraphs. The tool is clearly designed for web-first research, not document analysis. I felt like I was forcing a round peg into a square hole.

Also, the free tier is aggressively limited. After 5 Pro searches, you’re stuck with the basic model, which is noticeably dumber. It starts giving you shorter answers and sometimes misses obvious sources. If you’re a heavy researcher, you’ll feel the $20/month pinch quickly.

Pricing: Free tier is usable for casual browsing. Pro at $20/month is worth it if you do any kind of professional research. But there’s no middle ground – you’re either on the free plan with training wheels, or paying full price.

NotebookLM: The Librarian Who Only Reads Your Books (And Reads Them Really Well)

NotebookLM is Google’s weird, wonderful experiment. It’s not a general-purpose chatbot. It’s a tool that only talks about the stuff you upload. No internet access. No opinions. Just your documents, your notes, and your questions. I was skeptical at first, but it turned into my secret weapon for deep work.

What I loved

I uploaded the entire manuscript of my book (about 80,000 words) and a stack of 15 research papers. I asked NotebookLM, “Find all the sections where I contradict myself about the economic impact of AI.” It scanned everything and came back with three specific paragraphs, quoted them verbatim, and pointed out the logical inconsistency. I had missed that in my own editing. That’s not just a search – that’s a second brain.

The “Audio Overview” feature is wild. You click a button, and it generates a podcast-like conversation between two AI hosts discussing your documents. I uploaded a dense legal brief about data privacy laws, and the AI hosts argued about the implications of the GDPR vs. CCPA. They quoted specific sections, debated interpretations, and even paused to say “That’s a good point, but let’s look at the actual text.” It felt like I had two law students arguing in my ear. I used it to absorb complex material during my commute. It’s not perfect – sometimes the hosts get too enthusiastic and miss nuance – but for getting the gist of a 100-page document, it’s unbeatable.

I also used it for studying. I uploaded a textbook chapter on quantum mechanics and asked, “Create a study guide with key equations, their derivations, and common mistakes.” NotebookLM gave me a structured document with direct quotes from the textbook, cross-referenced with other chapters I had uploaded. No hallucinations. No guessing. It only used the source material I gave it. That’s incredibly reassuring when you’re learning something new.

Where it falls apart

NotebookLM is useless for anything outside your uploads. I asked it, “What’s the latest news on the OpenAI board drama?” and it just stared at me blankly. It can’t access the internet. Period. If you need real-time information, you have to manually copy-paste web pages into it. That’s a workflow killer.

Also, the interface is clunky. You have to create a “notebook” for each project, and you can only upload up to 50 sources per notebook. If you’re a power user with hundreds of documents, you’ll hit that limit fast. And there’s no API, no integrations with Google Drive (yet), no way to automate anything. You’re stuck in their web app.

Pricing: Completely free. That’s insane for what it does. Google is clearly using this as a data-gathering experiment, but for now, it’s a steal. I’d pay $20/month for it if they added internet access and more sources.

Google Gemini: The Multimodal Swiss Army Knife That Cuts Both Ways

Gemini is Google’s attempt at an everything app. It does text, images, audio, video, and code. It has a massive 1 million token context window (that’s like reading the entire Harry Potter series in one go). On paper, it should crush the competition. In practice, it’s a frustrating mix of brilliance and incompetence.

What I loved

The multimodal capabilities are legit. I took a photo of a handwritten whiteboard from a brainstorming session, uploaded it to Gemini, and asked, “Turn this into a structured project plan with deadlines and dependencies.” It read my terrible handwriting, understood the messy arrows, and output a clean table with tasks, owners, and timelines. That’s magic.

I also used it to analyze a 45-minute video of a conference talk. I uploaded the MP4 file and asked, “Summarize the key arguments and give me timestamps for each one.” Gemini processed the entire video, pulled out three main points, and gave me exact timestamps. I skipped the boring parts and went straight to the good stuff. Perplexity can’t do that. NotebookLM can’t do that.

The 1 million token context window is not a gimmick. I threw an entire codebase (about 800,000 tokens) at it – a messy Python project with 30 files – and asked, “Find all the unused imports and suggest optimizations.” Gemini scanned everything and gave me a list of 12 unused imports, plus refactoring suggestions. It even remembered a function I had defined in file 27 when I referenced it in file 3. That level of coherence is impressive.

Where it falls apart

Gemini is a liar. Not in a malicious way, but in a “I’m going to sound really confident about something that’s completely wrong” way. I asked it, “What’s the population of France as of 2024?” and it gave me 67 million. That’s the 2021 figure. The actual 2024 estimate is 68.4 million. When I pointed out the error, it apologized and gave me the right number. But why didn’t it get it right the first time? Because by default, Gemini does NOT search the internet. You have to manually enable “Google it” in the settings. Most users won’t know that.

I also tried using it for long-form writing. I asked it to write a 2,000-word essay on the history of the printing press. The result was a generic, boring, Wikipedia-style summary with no original insight. It felt like it was trying to please a teacher, not write something interesting. Perplexity’s answers have more personality. NotebookLM’s audio summaries have more depth.

The user interface is a mess. There are multiple versions: Gemini in the web app, Gemini in Google Docs, Gemini in Gmail, Gemini on your phone. They all behave slightly differently. The web app has a “Gemini Advanced” toggle that’s easy to miss. The mobile app has a different set of features. I kept getting confused about which version I was using and what it could do.

Pricing: Free tier is decent for casual use. Advanced at $19.99/month unlocks the 1 million context window and better multimodal support. But honestly, the free tier is so limited (only text and basic image analysis) that you’ll feel pressured to upgrade fast.

Head-to-Head Comparisons

Research: Perplexity wins, hands down

I needed to find the latest studies on gut microbiome and depression. Perplexity gave me 8 recent papers, all with direct links to PubMed. NotebookLM couldn’t help (no internet). Gemini gave me 3 papers, two of which were from 2020, and one was a blog post, not a study. Perplexity is the only tool here that treats research as a serious task.

Document Analysis: NotebookLM wins

I uploaded a 200-page legal contract. NotebookLM created a study guide, an audio summary, and answered specific questions with exact quotes. Perplexity struggled with the length and got confused. Gemini could handle the length but gave me generic answers that didn’t reference the actual text. NotebookLM is the only tool that respects your source material.

Creative Brainstorming: Gemini wins

I needed ideas for a marketing campaign for a new coffee brand. Gemini gave me 10 creative concepts, complete with slogans, visual ideas, and target demographics. Perplexity gave me a list of existing campaigns (boring). NotebookLM refused to answer because I hadn’t uploaded any marketing documents. Gemini’s multimodal capability (I uploaded a mood board image) made the ideas more specific and visual.

Fact-Checking: Perplexity wins

I asked all three, “Is it true that drinking coffee reduces the risk of Alzheimer’s?” Perplexity said yes, with a 2023 meta-analysis link. Gemini said yes, but cited a 2016 study. NotebookLM said “I don’t have information about that.” Perplexity’s real-time citation system is unbeatable for truth-seeking.

Learning a New Topic: NotebookLM wins

I wanted to learn about the basics of blockchain. I uploaded a textbook chapter and a few articles. NotebookLM created a personalized study plan, an audio deep-dive, and a Q&A session that only used my sources. Gemini gave me a generic explanation that included a mistake about proof-of-work. Perplexity gave me a list of links, but no structured learning path. NotebookLM is the only tool that feels like a tutor.

The Verdict: There Is No One Winner, But There’s a Clear Best Setup

I’m not going to pretend one tool is the best for everything. That’s a lie. Each tool has a specific superpower, and trying to use one for everything will leave you frustrated.

Perplexity is the best for research and fact-checking. If you’re a journalist, a student writing a paper, or a professional who needs to stay current, Perplexity is your daily driver. The $20/month Pro plan is worth it if you do more than 5 serious searches a day. But don’t use it for deep document analysis – it’s not built for that.

NotebookLM is the best for understanding complex documents and studying. If you’re a lawyer, a researcher, a student, or anyone who deals with long texts, NotebookLM is a godsend. And it’s free. Use it to digest PDFs, legal briefs, textbooks, and your own notes. But don’t ask it about the news – it’s clueless.

Google Gemini is the best for multimodal tasks and creative brainstorming. If you need to analyze a video, turn a photo into a plan, or generate creative ideas, Gemini is your tool. The 1 million token context window is amazing for code analysis. But don’t trust it for facts unless you enable internet search, and don’t expect great long-form writing.

My personal setup: I use Perplexity Pro for daily research and fact-checking. I use NotebookLM for deep dives into documents and study materials. I use Gemini Advanced only for multimodal tasks (video analysis, image-to-text, code review). I don’t use any of them for writing – they all produce mediocre long-form content.

If you forced me to pick only one, I’d choose Perplexity. It’s the most reliable for the most common task (getting accurate information quickly). But I’d be lying if I said I didn’t miss NotebookLM’s audio summaries and Gemini’s video analysis. The truth is, these tools are complementary, not competitive. Use them together, and you’ll be unstoppable. Use only one, and you’ll always feel like something’s missing.

Final Score (out of 10):

• Perplexity: 8.5/10 (loses points for weak document analysis and aggressive free tier limits)
• NotebookLM: 9/10 (loses points for no internet access and clunky interface)
• Google Gemini: 7/10 (loses points for accuracy issues, confusing UI, and mediocre writing)

The best tool is the one that matches your workflow. For me, that’s NotebookLM for deep work and Perplexity for quick answers. Gemini sits on the sidelines until I need to do something fancy with a video or a photo. Your mileage will vary, but at least now you know what each one actually does.