Learning AI

English

Tiếng Việt

English

Tiếng Việt

Appearance

Perplexity — The answer engine with citations that shows its sources ​

🔍

Hands-on — 30 seconds

You work in marketing, and your boss pings you at 9pm: "We have a meeting tomorrow morning. Send me a comparison of 3 ad platforms plus 2026 market-share figures, with sources so I can trust it." Open Perplexity and type: "Compare Meta Ads, Google Ads, and TikTok Ads in 2026 on cost and reach, build a table, and cite a source for each row." → 30 seconds later you get a coherent answer where every number carries a citation number you can click through to the original source. You open a few links to spot-check, then drop it into your slides. A 2-hour Google dig shrinks to 15 minutes. Real-world payoff: an "answer machine" that searches the web in real time and synthesizes the result with verifiable sources — the free tier already gets you going, and the $20/month Pro plan unlocks deep research. No more juggling 20 Google tabs.

"Perplexity isn't Google (which hands you 10 links) and it isn't vanilla ChatGPT (which answers from memory).It searches the web → reads many sources → synthesizes an answer, attaching a citation to each sentence. But it still hallucinates — knowing which parts to trust and which to verify is the difference that matters."

After this chapter you'll be able to

  • Tell apart Perplexity from Google, ChatGPT, and Gemini — and know when to reach for which.
  • Use search with citations: type a natural-language question, click the source numbers to verify, and use Focus to narrow scope.
  • Run Deep Research to get a long sourced report, and use Labs to generate a dashboard/file/app from a single prompt.
  • Install & use the Comet browser (agentic browsing) — and understand its real limits.
  • Call the Sonar API with curl if you're a developer, and understand token-based pricing.
  • Spot the risks: hallucinated hard facts, data used for training by default, and the copyright lawsuits still pending.

Note on the "shelf life" of this information

This reflects understanding as of mid-2026. AI tools change very fast (model version names, plan prices, feature limits) — many figures below come from cross-checked third-party sources because the official pricing page blocks automated access. In a few spots the sources contradict each other, and that's flagged explicitly. Just head straight to perplexity.ai and docs.perplexity.ai to check the latest before quoting anything as fact.

01 · What this tool is & when to use it ​

Perplexity (vendor: Perplexity AI) is an "answer engine" — a different kind of tool from both a traditional search engine and a plain chatbot. Instead of handing you a list of 10 blue links (like Google) or answering "from memory" with the risk of making things up (like vanilla ChatGPT), Perplexity searches the web in real time, reads many sources, and then synthesizes a coherent answer with numbered citations right inside the text. That's the core identity: every claim is tied to a source you can click and inspect.

By 2026, Perplexity has grown beyond the idea of a "search tool" into an AI workspace that includes: search with citations, Deep Research, Labs (generate files/apps), Spaces (collaborative workspaces), the Comet AI browser, and the Sonar API for developers. Main URL: https://www.perplexity.ai.

Three things that are easy to mix up (read carefully)

  • Perplexity (web app) = what you use at perplexity.ai for cited Q&A. Most of this chapter is about this.
  • Comet = a separate browser built by Perplexity (you install it on your machine), with an AI assistant that acts on your behalf — different from the web app.
  • Sonar API = a service for developers to call from code, billed per token — completely different from using the app. Non-coders can skip it.

What Perplexity does well (per research):

FeatureWhat it doesPlan needed
Search with citationsReal-time web search, answers with numbered sources. Runs the vendor's Sonar model by defaultFree
Pro SearchDeeper, multi-step queries with a choice of premium models. Free is capped (~5 per day)Free (limited) → Pro
Model switcherSwap between GPT / Claude / Gemini (per sources through mid-2026, versions like GPT-5.4, Claude Opus 4.6, Gemini 3.1 Pro — model version names change constantly, treat as reference)Pro and up
Deep ResearchSearches dozens of times, browses 100+ pages, cross-checks, then writes a structured report with citations. Takes ~2–5 minutes per queryFree (very limited) → Pro/Max
Labs (under the + menu, labeled "Create files and apps")Tackles 10+ minutes of work: generates reports, spreadsheets, dashboards, slides, mini web apps, and images — running Python and drawing chartsPro/Max
SpacesA collaborative space to store research by topic; the Enterprise version adds admin controls, audit logs, and SSOFree (basic) → Enterprise
Comet browserAn AI-native browser: summarize pages, ask questions across tabs, agentic browsing (add items to a cart, find coupons, draft an email from page content). Available on iOS/Android/Mac/WindowsThe browser itself is free; some agentic capabilities require Pro. A Comet Plus ~$5/month add-on unlocks premium publisher content (included free with Pro/Max)
Focus modesRestrict the search scope: Academic, Reddit, etc.Free
Image / video generationImage generation; video (sources around mid-2026 mention Sora 2 Pro on the Max plan). Image generation has been region-blocked at times — see Section 04Pro/Max
Sonar APIA search-grounded API for developers (see Section 03)Pay-per-use

Why "citations" are the headline feature

Perplexity's biggest difference from a regular chatbot: it attaches a source to each sentence by default. For anything you need to verify (news, technical or academic figures), you don't have to "take the AI's word for it" — you click the citation number, open the original page, and confirm it yourself. That's why researchers, journalists, and marketers love it. But don't confuse "has citations" with "always correct": it can still attach a source that doesn't actually support the claim, or fabricate a fact with no source at all — see Section 04.

Versus other tools — "when to pick which" ​

No tool "wins on every front." Perplexity is a specialist for sourced research — in raw "AI search" market share it's much smaller than ChatGPT/Gemini (ChatGPT leads by a wide margin, Gemini is second), but it's highly rated for tracking down verifiable sources. Note: 2026 market-share numbers swing wildly depending on measurement method (for example, ChatGPT ranges from ~57% to ~77%, Gemini from ~9% to ~25% depending on the source) — don't splice numbers from sources that use different methods; treat them as qualitative only. The table below is brief and neutral as of mid-2026:

ToolStrong atPick it when
PerplexityAnswer engine built for research; per-sentence citations by default; has Comet & the Sonar APIYou need a sourced, verifiable answer — tracking news/technical/academic facts, synthesizing research
ChatGPT Search (OpenAI)All-rounder, multi-step workflows, writing/coding/brainstorming; search is a supplementYou want a true "all-in-one" assistant for varied work beyond research
Google AI ModeAn AI tab right inside Google Search, with Google's infrastructure browsing huge numbers of pagesYou already live in Google Search and want AI right there
Gemini (Google)Creative, multimodal, integrated with Google WorkspaceYou prioritize images/video/multimedia, or you're already in the Google ecosystem
Claude (Anthropic)Reasoning, long-form writing, document analysis, codeYou're analyzing long text, writing, or coding — and can accept less search-grounding

When NOT to use Perplexity (the real limits)

Perplexity is great at research, but not every job suits it:

  • When you need absolute precision on hard facts (GPS coordinates, opening hours, phone numbers, official contact info): the community reports it fabricates very confidently in this area (see Section 06) → always verify, or use the primary source.
  • Heavy creative / multimedia tasks (artistic images, complex video, long brainstorms): Gemini/ChatGPT fit better.
  • Coding / long, in-depth technical writing: Claude/ChatGPT are usually better.
  • High-stakes agentic automation / money transactions via Comet: it still misclicks and gets stuck on payment pages — don't hand it risky work unsupervised.
  • Highly sensitive data on the consumer tier: by default your data is collected and used for training (unless you opt out) — see Section 04.

Quick summary

If your job is looking things up with trustworthy sources (news, figures, academic work, product comparisons) → Perplexity is the strongest choice. Need a versatile assistant that does a bit of everything → ChatGPT. Leaning toward creative/multimodal work, or already on Google → Gemini. Serious long-form writing or coding → Claude. Many people run them in parallel: Perplexity to find & verify, another chatbot to write & process.

02 · Sign-up / install & access ​

Pricing & access ​

Perplexity is available globally with no special setup — open the site, sign in, and start typing. There's a generous free tier, a paid web app (multiple plans), a downloadable Comet browser, and the Sonar API for developers. Below is a practical walkthrough of each, with verification steps so you know you've done it right.

Don't use a VPN to sign up or log in

This is a common trap: Perplexity blocks VPNs to fight spam, and it requires region-correct phone verification for Pro accounts. Using a VPN, or buying Pro through a foreign coupon, is a leading cause of accounts getting locked. Connecting directly on your normal network is the safest path.

Sign up / install ​

Web (nothing to install):

text
1. Open https://www.perplexity.ai
2. Sign up with email, or sign in quickly via Google / Apple.
3. Go straight to the question box → type and you're using it right away (free tier).

Comet browser (a real installer you download to your machine):

text
# Download at https://www.perplexity.ai/comet
# (needs >=2GB free disk space, 5-6GB recommended - per the official support page)

# macOS:
1. Open the .dmg file
2. Drag the "Comet" icon into the Applications folder
3. On first launch, grant permissions in System Settings -> Privacy & Security

# Windows:
1. Run the .exe file
2. Follow the install wizard
3. Open it from the Start menu

# Mobile:
- iOS: download "Comet" from the App Store
- Android: download from Google Play (package: ai.perplexity.comet)

There's no end-user CLI

Perplexity does not publish an official command-line tool (CLI) for end users — you interact through the app/browser. A real command line only exists on the Sonar API side (called via curl/SDK), see Section 03.

Plans & pricing (as of mid-2026) ​

Pricing-accuracy caveat

Perplexity's /pro page blocks automated access (403 error), so the pricing table below is compiled from multiple cross-checked third-party sources from 2026. There are contradictions in a few spots — flagged explicitly. Re-check the official page before buying or quoting anything as fact.

PlanMonthlyYearlyKey inclusions
Free$0$0Unlimited basic search (Sonar model), citations, ~5 Pro Searches/day; no premium models / full Labs / full Deep Research
Pro$20$200 (~$16.67/month, saves ~17%)Unlimited Pro Search, model switcher (GPT/Claude/Gemini), Deep Research (~20/day), file upload, image generation, $5 API credit/month, basic Comet/Computer access
Max$200$2,000Everything in Pro + Perplexity Computer (~10,000 credits/month; orchestrates 19 models to run complex tasks), unlimited Labs/Deep Research, Model Council (ask several models at once, then synthesize the answers), priority frontier-model access at peak hours, premium video
Education / Student Pro$10 (50% off)(usually monthly only)Same as Pro, for verified students/teachers
Enterprise Pro$40/user/month$400/user/yearPro for teams, shared Spaces, SSO/SCIM, audit logs, SOC 2 Type II; Deep Research ~500/day
Enterprise Max$325/user/month$3,250/user/yearAll of Enterprise Pro + unlimited Labs/Research, org-wide analytics

The UNCERTAIN bits (third-party sources — tagged with ~, treat as reference)

Per-plan Deep Research limits; frontier model version names; the "19 models" figure for Perplexity Computer (many sources say exactly 19, but the model versions change constantly). The Max price ($200/month, $2,000/year), Enterprise Pro ($40/seat/month, $400/seat/year), Enterprise Max ($325/seat/month, $3,250/seat/year), and Education ($10/month) match across multiple sources as of mid-2026 — but still verify at the official source before printing.

Payment & card access ​

  • Perplexity accepts international Visa / Mastercard / Amex / Discover credit and debit cards (processed via Stripe).
  • It also supports PayPal (Perplexity has integrated PayPal in some flows) — a fallback if a card is declined.

Note for Vietnam / SEA readers

Domestic-only cards from some countries (Vietnam included) are often blocked for recurring international SaaS charges. To reduce the chance of a decline, use a virtual card that supports international payments, or fall back to PayPal. Also, beware of sellers advertising "cheap Perplexity Pro" (a few dollars a year) — these are usually shared accounts or foreign-region coupons and carry a real risk of getting locked due to region-correct phone verification. Not recommended for serious work — losing the account mid-project costs more than it saves.

Tips for choosing a plan

  • Just trying it / light use → Free is enough for daily lookups, no card needed. Once you hit the ~5 Pro Searches/day cap, drop back to Quick Search.
  • Regular research use → Pro ($20): unlocks Deep Research, the model switcher, Labs, and full Comet. The best value for most people.
  • Only step up to Max ($200) when you genuinely need Perplexity Computer / unlimited Labs / premium video — don't overpay for what you won't use.

03 · The hands-on workflow — step by step (with real prompts/commands) ​

This is the "from question to done" process. Each part has a way to self-check (verify) so you know you did it right.

A. Research with citations (web app) ​

text
1. Go to perplexity.ai -> type a natural-language question.
2. Pick a Focus to narrow scope if needed: Academic, Reddit (community opinions).
3. Turn on Pro Search (Pro plan) for multi-step queries.
4. Read the answer -> click the citation numbers to open the original source and verify.
5. Use "related questions" to dig deeper; save it to a Space by topic for reuse.

A sample prompt for serious research:

text
Compare 3 options X, Y, Z on [cost, performance, regional support] as of 2026.
Use only sources <= 12 months old. Build a table + cite a source for each row.
If the data conflicts, present both and cite the sources.

→ Verify: the answer has clickable citation numbers; when you open them, the sources are real, on-topic, and not a 404 or an off-topic link.

B. Deep Research → a long sourced report ​

text
1. Choose the Deep Research mode (or via the + menu).
2. Give a clear brief: scope + the output format you want.
3. Wait ~2-5 minutes (the system browses 100+ pages and cross-checks).
4. Read the report; if you need a dashboard/slides -> push it on to Labs.

→ Verify: you get a dense, sourced synthesis — and you ALWAYS re-verify the key figures/sources before using them (Deep Research can still cite shallow or off-target sources, see Section 06).

C. Labs — generate an app / dashboard / file from a single prompt ​

text
1. + menu -> "Create files and apps" (Labs).
2. Write a fully detailed prompt -> Labs researches + runs Python +
   builds charts -> produces a mini web app / dashboard (~10-30 minutes).
3. Open the "Tasks" tab to see which steps it took.

A sample Labs prompt (dashboard):

text
Build a dashboard tracking [AI startups that raised funding]: funding data, founders,
investors, charts + a map, with filters by date/topic.
Use reputable sources and note the date of the data.

Labs is powerful but you must check the output

Per DataCamp's testing, Labs sometimes gets the current year wrong (thinks it's still an earlier year → stale data), breaks chart/map layouts, or produces broken exports. Don't use Labs output as an official document without reviewing every number.

D. Comet — agentic browsing (a browser that acts for you) ​

text
1. Install Comet (Section 02) -> sign in.
2. Give the assistant a high-level command, for example:
   - "summarize this page"
   - "find a video summarizing topic X"
   - "add items to the cart + find a coupon"
   - "read the Gmail tab, list important unanswered emails"
3. Comet acts across the tabs on its own.
   (The agent model, per sources through mid-2026: Pro defaults to Sonnet 4.6;
    Max defaults to Opus 4.6, with Sonnet selectable - versions change constantly, treat as reference.)

→ Verify: Comet does exactly what you asked. If it gets stuck on a complex form or a payment page → step in manually at that point; don't let the agent click "pay" on its own.

E. Sonar API — for developers (real commands) ​

Get your API key at console.perplexity.ai (Settings → API). A common snag: you must load prepaid credit FIRST before the Generate key button becomes active (this is separate from the "$5 API credit/month" that comes with Pro — that's consumption credit and doesn't unlock the console on its own). Reference docs: docs.perplexity.ai. Once you have a key → call the chat-completions–style endpoint:

bash
curl https://api.perplexity.ai/chat/completions \
  -H "Authorization: Bearer $PPLX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "sonar-pro",
    "messages": [
      {"role": "user", "content": "Summarize the latest news on [topic], with sources."}
    ]
  }'

The model names sonar, sonar-pro, sonar-reasoning-pro, sonar-deep-research are per the docs ~mid-2026. The endpoint syntax is confirmed against the docs; check the docs for specific fields, as they may change.

Sonar API pricing (~mid-2026, per the docs):

ModelInput /1M tokensOutput /1M tokensPer 1,000 requests (Low→High)
Sonar$1$1$5 → $12
Sonar Pro$3$15$6 → $14
Sonar Reasoning Pro$2$8$6 → $14
Sonar Deep Research$2$8+ citation $2/1M, search $5/1K, reasoning $3/1M

API cost note

In 2026 Perplexity dropped citation-token fees for Sonar/Sonar Pro (they remain only on Deep Research) → cheaper per response than in 2025. The Pro plan includes $5 API credit/month; heavy use requires buying more prepaid credit. The Sonar API is Zero Data Retention — it doesn't store the data you send through the API (see Section 04).

Small tricks to improve answer quality (web app)

  • Demand explicit sources: add "cite a source for each point, prioritizing sources ≤ 12 months old" → fewer stale/fabricated facts.
  • Allow "I don't know": add "if you can't find a trustworthy source, say so instead of guessing" → fewer fabricated hard facts.
  • Use the right Focus: academic questions → Academic; need real user experiences → Reddit.
  • Save a Space per topic: group research sessions on the same subject so you don't have to retype the context.

04 · Tips & common mistakes ​

🟢 High-value tips ​

6 tips for using Perplexity like a pro

  1. Always click the citation numbers to check the source — "has a citation" isn't "correct"; it can attach a source that doesn't actually support the claim.
  2. Give a constrained brief (source date cutoff, table format, "cite a source for each row") → results stay on-target and are far easier to verify than from a vague prompt.
  3. Let the AI say "not found" → markedly fewer fabricated hard facts (GPS, hours, phone numbers).
  4. For hard jobs → split Deep Research (find & synthesize) then push to Labs (build the output) instead of forcing one prompt to carry it all.
  5. Switch models when an answer isn't satisfying (Pro) — the community reports that picking a different model makes a real difference.
  6. Start on Free, step up to Pro ($20) when you hit the ~5 Pro Searches/day limit or need Deep Research/Labs.

🔴 Mistakes & pitfalls (read carefully — this section saves you) ​

🚨 Hallucination — still happens, even "with citations"

Perplexity still fabricates, especially on hard lookup facts: Hacker News community testing found it fabricated GPS coordinates that were almost 100% wrong, restaurant/attraction opening hours that were "mostly made up," and agency phone numbers/departments that were "almost 100% wrong." The main complaint: it answers confidently but wrong instead of saying "not found."

Warning signs:

  • A claim with no attached source, or a source that doesn't actually say that (click through and it's off-point).
  • Facts like coordinates/hours/contact info — the area it gets wrong most.
  • Re-ask a different way → you get a completely different answer.

→ ALWAYS verify hard facts against the primary source before using them for real work.

Privacy & data — read this if you use it for work

Per Perplexity's Privacy Policy effective 2026-02-05:

(a) Where does your data go?

  • On the consumer tiers (Free / Pro / Max), by default your prompts and queries are collected, stored, and used to improve/train models — unless you opt out yourself.
  • On Enterprise, customer data is NOT used to train/fine-tune (including chats, file uploads, metadata).
  • Sonar API: Zero Data Retention — data sent through the API is not stored and not trained on.

(b) How to turn off training (consumer tiers):

  • Go to Settings → turn off the AI data retention / training option if you're entering sensitive data.

(c) NEVER paste the following into the consumer tiers:

  • National ID numbers, bank card numbers, passwords, OTPs.
  • Contracts/NDAs, confidential documents, your company's proprietary source code.
  • Customers' personal data (names, phone numbers, addresses, records) — in most jurisdictions this can violate data-protection law (e.g. the EU's GDPR, and equivalent privacy regulations elsewhere).

(d) If you must process sensitive data → use Enterprise / the Sonar API (Zero Data Retention), not the consumer tier.

⚖️ Legal risks still in the air (worth knowing to judge durability)

This is the hot section, reported honestly — most of it is allegations in lawsuits, not court findings:

  • Tracking lawsuit (early Apr 2026): a class action under CIPA (the California Invasion of Privacy Act, filed around Mar 31–Apr 1, 2026) alleges Perplexity embedded "undetectable" tracking software that shares conversations with Meta/Google even in incognito mode. An important detail for assessing your own risk: per the complaint, the proposed class covers only free-tier accounts (roughly Dec 2022–Feb 2026) and excludes paying Pro/Max users; the class has not yet been certified by a court. (Still an allegation — worth watching how it unfolds.)
  • A wave of copyright/scraping lawsuits: as of ~May 31, 2026, around 9 organizations are suing Perplexity (CNN — a May 28, 2026 complaint alleging it copied 17,000+ works; The New York Times; News Corp/Dow Jones; the New York Post; the Chicago Tribune; Britannica; Merriam-Webster; Reddit; the Yomiuri Shimbun). Earlier (Aug 2024), Cloudflare documentation plus a Wired/Robb Knight investigation alleged Perplexity used a hidden crawler, ignored robots.txt, and spoofed its user-agent. The vendor's response (CCO Jesse Dwyer): "You can't copyright facts."
  • The practical upshot: if you rely on major news sources, these lawsuits could affect content/source relationships down the line → keep an eye on it.
FAQ & common errors (click to open)

"Image Generation Not Supported in This Region" — what's going on? Image generation rolls out by region/compliance (in early 2026 several regions, including parts of Southeast Asia, were temporarily blocked). It's not a problem with your account — wait for it to open in your region.

Out of quota / a vague error during search? Free is only ~5 Pro Searches/day; hitting the cap shows a generic error. Fix: switch back to Quick Search, or upgrade to Pro.

"Something went wrong" on every query? Possible causes: (1) a service outage, (2) a stale cache, (3) a blocking extension. Fix: check the service status; hard refresh (Ctrl+Shift+R) + clear cache; test in incognito with extensions off.

Won't load / broken UI? The #1 cause is an ad-blocker / privacy extension / VPN. Fix: disable extensions; don't use a VPN; try Chrome/Edge (Safari often breaks a few features).

Locked out / asked for phone verification? Usually from buying Pro via a foreign-region coupon; it requires region-correct phone verification. Fix: use a properly registered account for your region, and avoid foreign coupons.

Comet stuck / misclicking while acting on its own? Caused by complex forms, unfamiliar logins, dynamic elements, payment pages. Fix: step in manually at that point; break the task into smaller pieces; don't hand sensitive payment work to the agent.

App slow / not loading — do I need a VPN?No VPN needed (and you shouldn't use one). If it's slow, try a different network or disable extensions.

05 · Exercises / mini-projects ​

Actually do 2–3 of the exercises below to turn "I read it" into "I can do it." Each has clear completion criteria.

Exercise 1 — Sourced research & the verification reflex (basic) ​

Goal: practice the single most important skill when using Perplexity — clicking the source to verify.

  1. Go to perplexity.ai and ask a question that needs fresh figures, for example:
text
Top 3 most popular programming languages in 2026 per a reputable survey,
with the % and a source for each number. Use only sources <= 12 months old.
  1. Click each citation number, open the original page, and compare: does the number in the answer match what the source says? Is any source off-topic or a 404?

Done when: you can confirm each number matches its source (or catch a spot where the source was attached incorrectly). This is a reflex to keep forever.

Exercise 2 — Deep Research into a report (important) ​

Goal: experience deep research and learn not to "trust blindly."

  1. Choose Deep Research and give a clearly scoped brief:
text
Survey the home water-purifier market in 2026: the main brands,
price ranges, and consumer trends. Write a report with sections and cite each part.
If sources conflict, present both.
  1. Wait ~2–5 minutes. Read the report, then pick the 3 most important claims and verify them yourself by opening the sources.

Done when: you have a structured report + you've verified at least 3 key points, and you've noticed where the report "sounds authoritative but is shallow underneath."

Exercise 3 — Build a dashboard with Labs (needs Pro/Max) ​

Requires a paid plan

Labs needs Pro/Max. If you're on Free, do Exercises 1 and 2 first; save this one for when you upgrade.

Goal: build a real output from a single prompt — and practice the habit of checking for errors.

  1. + → Create files and apps (Labs), with this prompt:
text
Build a dashboard comparing 5 best-selling mid-range phones in 2026:
price, key specs, camera score, battery. Include a table + a bar chart.
Use reputable sources and note the date of the data.
  1. When it's done, open the Tasks tab to see what it did, then review: is the data year actually 2026? Are the charts breaking the layout? Do the numbers match the sources?

Done when: you have a working dashboard and you can name at least one thing to fix/verify before you'd dare hand it to someone else.

06 · Case studies & real use-cases (from the community) ​

This section gathers real examples from a tutorial (DataCamp), community discussion (Hacker News), and official announcements through ~mid-2026. The point: to show you how Perplexity actually behaves in the real world — both when it shines and when it becomes a trap.

Read carefully about source reliability

Some of the content below is personal experience / third-party testing, not independently verified:

  • Remarks like "beautiful," "accurate," "almost 100% fabricated" are one tester's judgment in a specific context → read as reference.
  • Numbers like "4.6/5 for Vietnamese," "89%/94% ease of use" come from self-published regional reviews, not vendor figures → qualitative reference only.
  • Whatever is official (the Perplexity blog) is labeled as such.

⚽ CS1 — An FC Barcelona dashboard with Labs (beautiful but a few details wrong) ​

  • Context: The DataCamp author tested Perplexity Labs by asking it to build a football-fan dashboard from a single prompt.
  • What they did: One prompt describing a dashboard with the last 5 results, the fixture list, the squad with photos + contract status, an injury list, and transfer news.
  • Result: The output was beautiful, the squad info was accurate, the club colors were right; but a few details were wrong about contract extensions/injuries (a research limitation, not a system bug).
  • Lesson: Labs produces an impressive visual product from a single prompt, but the factual details must be reviewed — don't treat the output as the final truth.
  • Source: DataCamp — Perplexity Labs tutorial.

CS2 — Analyzing Shopify's 10-K with Labs ​

  • Context: Testing Labs on a financial-analysis task (reading an annual 10-K report).
  • What they did: Labs read the 10-K data and generated a detailed report with its own charts, drawing on more sources than a typical AI research tool.
  • Result: The visuals made the numbers easier to present; the report had more depth thanks to multi-source synthesis.
  • Lesson: For number-heavy analysis, Labs excels at turning raw figures into charts + a report — the tool's sweet spot.
  • Source: Perplexity Labs documentation/tutorial.

CS3 — 5 experimental trackers/dashboards (including when it breaks) ​

  • Context: DataCamp built 5 dashboards with Labs: a climate heat map, a tech-conference calendar, an AI-startup funding tracker, a global-conflict map, and more.
  • What they did: Each mostly needed just 1 prompt, built in ~10–30 minutes.
  • Result / real failures: Many came out beautifully; but in one case Labs got the current year wrong (thought it was still an earlier year → stale data), in another it over-engineered and broke the pipeline down to 2 rows of data, and exports/charts sometimes broke.
  • Lesson: Labs is powerful but not "set & forget" — you have to check the output, especially the data year and the integrity of charts/exports.
  • Source: DataCamp.

CS4 — Comet as a Gmail assistant (handy but easy to get stuck) ​

  • Context: A Comet browser review (USAII / toolstack) tried using it as an assistant controlling tabs directly.
  • What they did: Asked Comet to summarize long emails and find important unanswered emails by acting on the Gmail tab; also tried a shopping use-case (add to cart + hunt for coupons + compare prices).
  • Result: Summarizing and filtering email worked; but the review warned that pages with complex forms / unfamiliar logins / dynamic elements easily cause Comet to misclick and get stuck on payment pages, sometimes slower than doing it by hand.
  • Lesson: Agentic browsing is useful for reading/summarizing, but for transactions/payments, supervise closely or do it manually.
  • Source: Comet review (USAII / toolstack).

🤝 CS5 — Hacker News community experiences (very mixed) ​

  • Context: The thread "Ask HN: What's Your Take on Perplexity AI?" and a thread discussing Deep Research.
  • Positive: One user, using it 2+ years for web search, said it "hallucinates less than before" and that "choosing a model makes a difference."
  • Negative (very specific): One user's testing found Perplexity fabricated GPS coordinates that were almost 100% wrong, restaurant/attraction opening hours that were "mostly made up," and government phone numbers/departments that were "almost 100% wrong" — the main complaint being that it answers confidently instead of saying "not found." Some criticized Deep Research as "authoritative-sounding, well-structured, but possibly shallow in content."
  • Lesson: Strong for synthesis & web search, weak for hard lookup facts. This is exactly why Section 04 hammers on verification.
  • Source: Hacker News — "Ask HN: What's Your Take on Perplexity AI?" + the Deep Research thread.

CS6 — Perplexity funding AI & journalism research (positioning its user base) ​

  • Context: (Reference, not a technical use-case.) Perplexity announced a $250k gift to Northwestern/Medill to research AI and journalism.
  • Meaning: It shows the vendor positions journalists / researchers as its core user base — consistent with its "sourced research" strength.
  • Lesson: If your work is research/journalism, this is a tool designed with you in mind — but still weigh it against the copyright legal risks in Section 04.
  • Source (official): Perplexity blog.

🧷 Takeaways from the community

  • Trust Perplexity most for: multi-source synthesis, tracking news/technical/academic facts, comparisons — things you can click through to verify.
  • Don't trust blindly for: hard facts (GPS, hours, contact info), and Labs output (data year, charts) — always review.
  • Golden rule: "has a citation" ≠ "correct." Clicking through to read the source is the only real verification.

07 · Summary & official sources ​

5 things to take with you

  1. Perplexity = an answer engine with citations — real-time web search, synthesis, a source attached to each sentence. Strongest for research that needs verification.
  2. "Has a citation" isn't "correct" — always click the source numbers to verify, especially hard facts (GPS, hours, contact info) that it often fabricates.
  3. Available globally, no VPN needed (and you shouldn't use a VPN — it risks a lockout). Free is enough for lookups; Pro $20 unlocks full Deep Research/Labs/Comet.
  4. The backbone workflow: sourced search → Deep Research (synthesize) → Labs (build the output) → Comet (act). Check the output at every layer.
  5. Privacy: the consumer tier trains on your data by default (unless you opt out); for sensitive data → Enterprise/API. Keep an eye on the pending copyright/tracking lawsuits.

These are the first-party pages to check for the latest information — always trust these over third-party round-ups:

Reliability notes (research through ~mid-2026)
  • Solid (multiple sources agree): the answer-engine + citations nature; Pro $20/month, $200/year; Max $200/month, $2,000/year; Enterprise Pro $40/seat/month ($400/year), Enterprise Max $325/seat/month ($3,250/year); Education $10/month; Free ~5 Pro Searches/day; the existence of Deep Research/Labs/Spaces/Comet/Sonar; Comet (the browser) free globally + Comet Plus ~$5/month; works globally with no VPN; Sonar API pricing per the docs; the copyright lawsuits and the Cloudflare/tracking allegations.
  • Reference (tagged with ~, changes constantly): per-plan Deep Research limits; frontier model version names (per sources through mid-2026: GPT-5.4 / Claude Opus 4.6 / Gemini 3.1 Pro); the "19 models" figure for Computer.
  • Self-published third-party numbers (not the vendor's): 4.6/5 for Vietnamese, 89%/94% ease of use — qualitative reference only.
  • The /pro page blocks automated fetches (403) → pricing comes from cross-checked third-party sources; re-verify at the official source before printing.

Figures (pricing, models, features) may have changed — always re-check at perplexity.ai and docs.perplexity.ai.