Perplexity will not find you a restaurant. Its own CEO says Google does navigational searches better than anyone, and a reviewer found Google still faster for a specific site, a quick fact or a simple how-to. Keep Google as your default.
For fifteen jobs that end in a sourced deliverable, the answer flips. Those are the jobs in this list: a comparison table, one contract, many sources, a live purchase, a whole folder. Perplexity wins because it shows where each line came from.
The honest cost: a chat answer lands in 10 to 60 seconds, a long PDF in 1 to 2 minutes, and Research or agent runs take several minutes. Your review and edits are always the bigger number, and never zero.
Each entry gives you the exact prompt, the Perplexity mode to use, the three-part time ledger, and the one check that keeps an invented source out of the answer. All fifteen run on the free plan at least once.
At a Glance
1. Comparing Five Laptops and Their Owner Complaints
Seventy-five minutes is what five laptops cost by hand. Search back and forth, five maker pages for the specs that matter, and the owner complaints at the bottom of review sections nobody scrolls to.
Pro Search builds the grid in about 45 seconds. The 20 review minutes go into opening the flagged cells, not into drawing the table, and every cell carries a link and a date so stale retailer specs and star ratings get caught.
Prompt to copy
You are a product analyst who does not trust a spec sheet. I am choosing among [PRODUCT A, PRODUCT B, PRODUCT C, PRODUCT D, PRODUCT E] for [USE CASE], budget [BUDGET], buying in [COUNTRY]. 1) Build one table, one row per product, columns for [SPEC 1], [SPEC 2], [SPEC 3], price and warranty. 2) Fill every cell from the maker's own page or a named review, and write 'unverified' rather than guessing. 3) Add the two most common owner complaints per product, each with a linked source. 4) End with the one product you would rule out first and the reason. Rules: no star ratings without a source, no price without a date, flag any spec only the manufacturer states. Output: the table, then five lines of verdict.
💡 Tip: Open the flagged cells first, not the whole table. Five links decide whether the verdict holds.
| Job | Best tool | By hand (minutes) | Perplexity's own time | Review and edits (minutes) |
|---|---|---|---|---|
| Five-product spec comparison | Pro Search | 75 | ~45 seconds | 20 |
| 40-page contract obligations | File upload | 120 | ~2 minutes | 25 |
| New-market briefing | Research | 240 | ~8 minutes | 35 |
| Twelve-claim fact-check | Pro Search | 90 | ~2 minutes (batched) | 18 |
| Steelman the other side | Pro Search | 45 | ~40 seconds | 10 |
| Literature scan with citations | Research | 300 | ~12 minutes | 40 |
| Lost statistic, found | Pro Search | 40 | ~60 seconds | 8 |
| Contract clause translator | File upload | 60 | ~90 seconds | 12 |
| Two-vendor deal comparison | File upload | 150 | ~2 minutes | 30 |
| Pre-meeting company dossier | Research | 120 | ~7 minutes | 20 |
| Earnings call autopsy | File upload | 90 | ~2 minutes | 15 |
| Six-outlet coverage comparison | Pro Search | 60 | ~50 seconds | 12 |
| Competitive landscape table | Research | 180 | ~9 minutes | 30 |
| Multi-tab price sweep | Comet Assistant | 35 | ~4 minutes | 10 |
| Downloads folder to cited index | Personal Computer | 100 | ~15 minutes | 20 |
2. The 40-Page Contract Obligations Prompt
Prompt to copy
You are a contract reviewer's assistant, not a lawyer. Attached is [CONTRACT PDF]. 1) Identify the parties, term, renewal and notice window, payment terms, termination rights, indemnity, liability cap and governing law. 2) For each, quote the section number and a ten-word excerpt, then explain it in one plain sentence. 3) List what the document does not state that a contract this size normally would. 4) Flag the three clauses most one-sided against [MY SIDE]. Rules: quote exactly, never round a number, write 'not stated' instead of inferring, add no legal advice. Output: a table of clause / section / plain meaning / risk, then the three flags.
That is the same rule as the product table, moved onto a document: quote what is written, mark what is absent. A shortlist carries a spec; forty pages carry a liability cap and a notice window that decides whether you can leave.
Upload a redacted PDF and the answer lands in about 2 minutes. Redact names. Never paste passwords, client files, or medical or financial records into any chatbot; personal plans may train on chats unless that is switched off. It is not legal advice.
The answer is a table of clause, section, plain meaning and risk, plus the three one-sided clauses named. Check every quote against the page. Renewal and notice are known before the first call, and absences come back as absences.
💡 Tip: Search the PDF for each quoted ten-word excerpt. If the quote will not match, the cell is wrong.
⚠️ COMMON MISTAKE
Do not paste the whole client file
Redact names, account numbers and anything medical or financial before uploading. On personal plans your chats may train the company's models unless you turn that off in settings. The prompt is not legal advice, and no clause leaves the page unchecked.
3. The New-Market Briefing Prompt
That is the question a market brief answers. Google serves SEO posts citing each other, with no category boundary, no dates, and no split between reported and estimated. By hand it runs 240 minutes.
Step 1: Open Perplexity, switch to Research mode, and paste the prompt below. You should see it work through many pages over about 8 minutes, then return a brief rather than a list of links.
Prompt to copy
You are an analyst writing an internal briefing for [COMPANY]. Topic: [CATEGORY] in [GEOGRAPHY], read in [MONTH YEAR]. 1) Define the category boundary and say what you excluded. 2) Size it from named sources, each with a date and a link, and mark each figure reported or estimated. 3) Name the five players and their positioning in one line each. 4) List three things everyone repeats that only one source actually supports. 5) List what changed in the last twelve months. Rules: a footnote link on every claim, no undated numbers, no figure older than eighteen months without a flag. Output: a 600-word brief, then the numbered source list with dates.
Step 2: Read the numbered source list first and open anything undated. The free plan includes 1 Research query a month, so a second brief means Pro, as of November 2026. Budget 35 review minutes for the brief itself.
What survives is the part a room cannot knock down: 600 words on numbered, dated sources, an explicit statement of what was excluded, and the three claims everyone repeats on a single source.
4. The Twelve-Claim Fact-Check Prompt
Three of the twelve lines came back contradicted, and they were not the three anyone would have guessed. The expensive part was not finding them.
It was the eighteen minutes spent opening each primary link to confirm the verdict held, because a supported label is only a lead until you read the filing yourself.
Prompt to copy
You are a fact-checker. Here is my draft: [PASTE DRAFT]. 1) Extract every checkable claim into a table with the exact wording. 2) For each, find the primary source – the filing, the study, the original statement – and mark it supported, contradicted or unsupported, with the link. 3) Flag every number, date, name and quotation separately, since those fail loudest. Rules: primary sources only, never cite an article that is itself citing the source, ban the phrase 'widely reported', write 'no primary source found' when that is the truth. Output: the table, then the three claims most likely to embarrass me and why.
The work by hand is an afternoon: one search per claim, then a judgment call on whether a write-up counts as the source. Perplexity answers all twelve in about two minutes.
But three Pro Searches a day will not cover a dozen claims in one sitting, so split the list across days or pay for Pro, and re-open every link the table calls supported.
💡 Tip: Run it on your own last draft first. It is the cheapest way to find out how the tool behaves.
5. The Steelman the Other Side Prompt
A counterargument that reads well and cites nothing is worse than no argument, because it makes you feel prepared. So the rule is fixed before you paste: a point without a link is deleted, not softened.
Expect the same trap at the end, since the model tends to hedge a genuinely weak side rather than say outright that it is weak.
The setup is the one fact-checking already taught: same sourcing standard, different target. Google hands back what matches the framing already used, and the strongest case against sits in sources nobody opens after a decision feels settled.
Prompt to copy
You are an advocate for the other side of my decision. My position: [POSITION], held because [REASONS]. 1) Give the three strongest arguments against it, each with a linked source and the best data behind it. 2) For each, name who makes that argument professionally. 3) State what evidence would change my mind. 4) Name who benefits if I am wrong. Rules: no strawmen, no both-sides mush, no argument without a source, and if the opposition really is weak, say so plainly. Output: three numbered arguments, each followed by one line on how I would answer it.
Pro Search returns in about 40 seconds what takes 45 minutes to assemble by hand. Ten minutes of review follows, and it goes on opening the links, not on rewriting the prose. What survives is three sourced objections with the professional camp behind each.
Then you walk in with a one-line answer prepared for every argument, and a written note of the evidence that would actually change your mind. That is the difference between defending a decision and having tested it before anyone else did.
The Words the Prompts Use
6. The Literature Scan With Citation Slots
The job is finding what has already been published on a question in a field whose vocabulary you do not have. This is the one place to slow down before trusting the machine.
A database is a searchable index of published papers, run by a publisher, a university or a government body, and it only answers a search you can already phrase.
Google Scholar works the same way. It needs the search terms, and those terms are the thing you are asking for. That is 300 minutes of blind reading by hand, about 12 minutes in Research mode and 40 minutes of review.
Chatbots invent citations, so every DOI or publisher page gets opened before it goes into your notes. A DOI is a permanent ID number a paper carries, and an entry that will not open gets deleted, not kept.
Prompt to copy
You are a research librarian. Question: [QUESTION]. Field: [FIELD]. Window: [YEARS]. 1) List the twelve most relevant papers with authors, year, venue, DOI or publisher page, and a one-line finding each. 2) Sort them into supporting, contradicting and method-critique. 3) Add two review articles that cover the field. 4) List eight search terms and two databases I should use next. Rules: only papers with a resolvable identifier, never invent an author or a title, write 'could not verify' instead of filling a gap, no summaries from the abstract alone without saying so. Output: the annotated list by group, then the search-term list.
You get twelve papers with resolvable identifiers sorted into three groups, two review articles and eight search terms for the databases. The 40 review minutes are twelve clicks, and each one that opens is a paper that exists.
7. The Lost Statistic, Found and Sourced Prompt
A number has been in the decks for two years. Nobody knows where it started, and every write-up paraphrases it a little differently. Searching for it means guessing the words someone else chose, which is exactly what you cannot do.
By hand that is forty minutes of dead ends. Pro Search gets a candidate in about sixty seconds, then eight minutes of your review. The prompt is built for a half-memory, not a clean quote.
Prompt to copy
You are a source hunter. I half-remember this: [HALF-REMEMBERED STATISTIC OR QUOTE], possibly from [YEAR OR PUBLICATION]. 1) Write five plausible phrasings of it. 2) For each, find the earliest named source and give the publication, date and link. 3) If only a secondary source exists, say so and name the primary source it points to. 4) Give the exact wording and date of the best candidate. Rules: no source, no answer; do not smooth my memory into a cleaner quote; stop after three dead ends and tell me it may not exist. Output: best source, exact quote, and a confidence line saying what would confirm it.
The catch: a chat assistant will produce a confident quote that never existed, in seconds, in exactly the tone of a real one. Nothing goes back on a slide until the original document is open in front of you.
What comes back is a best candidate with exact wording, date and a line saying what would confirm it. It may also come back saying the statistic may not exist. That retires a slide quietly, which beats a correction in public.
💡 Tip: Paste the five phrasings into Google too. If none of them appear, your memory of the number is probably off.
8. The Contract Clause Translator Prompt
Read alone, the answer hides in eight money-and-rights clauses scattered through boilerplate. The rule that found a lost statistic applies here: quote the section, and mark what the document does not say.
One realistic terms document takes sixty minutes by hand. Upload the PDF and Perplexity returns the analysis in about ninety seconds, then twelve minutes of checking every quote against the section number.
Prompt to copy
You are a plain-language translator for a document I must accept or refuse. Attached: [POLICY OR TERMS PDF]. 1) Pick the eight clauses that can cost me money, take a right away, or bind me to something later. 2) For each, quote the section number and the operative sentence, then explain it in one sentence a friend would understand. 3) Note every deadline, notice period and auto-renewal. 4) Say what happens if I do nothing. Rules: quote rather than paraphrase, flag anything that varies by state, never fill a gap with what is typical, add no legal advice. Output: a numbered list with section references and a one-line bottom line at the end.
Two limits stay fixed. Terms differ by state and change without notice, and this is not legal advice. Anything the document never states comes back as "not stated" instead of a reasonable guess.
The two times: 60 minutes by hand against 90 seconds plus 12 minutes of review, with eight clauses quoted by section and a bottom line that covers doing nothing.
💡 Tip: Skip the redaction here only if you are uploading a public policy page. For your own signed contract, redact first.
9. The Two-Vendor Deal Comparison Prompt
The tedious part is two PDFs that number their sections differently. One reads clause 7.2, the other page 14, and the exit clause hides in one of them during a by-hand pass that runs 150 minutes anyway. Two uploads to Perplexity take about 2 minutes.
Prompt to copy
You are comparing two vendor contracts for [MY COMPANY], committing for [TERM]. Attached: [VENDOR A CONTRACT] and [VENDOR B CONTRACT]. 1) Build a clause-by-clause table on term and renewal, price escalation, exit and notice, data ownership and portability, service credits, liability cap and termination for cause. 2) Note where either contract is silent and what that silence costs us. 3) Mark which is better for a one-year commitment and which for a three-year, and say when that changes the answer. Rules: cite page or section for every cell, no winner on price alone, no assumption that an absent clause is standard. Output: the table, then a five-line negotiation note.
The real cost: 150 minutes by hand against 32 minutes with the tool.
A price is not the answer alone, though; every cell cites page or section, and a headline rate you see does not settle a three-year commitment, so 30 minutes go into checking each silent clause and what it costs you.
💡 Tip: Ask for the three-year view explicitly. The deal that wins at one year is often not the one that wins at three.
When Google Is Still the Better Tab
Stay in Google
- Finding a nearby restaurant or a sports score
- Going to one specific website by name
- Checking a single fact you already know how to search
- Reading a product page with current pricing
- A simple how-to you will follow once
Open Perplexity
- The answer needs sources you can show someone
- You are still shaping the question
- The work ends in a table, brief or index
- You have a file, a folder or a transcript to read
- One fact has to survive being challenged
10. The Pre-Meeting Company Dossier Prompt
Private companies disclose almost nothing. Google returns their own press releases and a funding round from three years back with no period attached. Add a rule before you start: label anything unconfirmed as rumored, and write not disclosed rather than estimating a private number.
Prompt to copy
You are an analyst briefing an executive before a meeting. Company: [NAME]. Meeting: [DATE AND PURPOSE]. 1) Say what they sell and to whom, in three lines. 2) Give the latest revenue and profit figures with the period and the source. 3) List three strategic moves from the last twelve months. 4) List two risks named by someone other than the company. 5) Give three questions I should ask and one thing not to say. Rules: filings and named press only, dates on every figure, label anything unconfirmed as rumored, write 'not disclosed' for private numbers rather than estimating. Output: one page, bullets, dates on every number.
Run it in Research mode: 120 minutes by hand, about 7 minutes for the brief, 20 minutes of review.
PitchBook and Statista, the premium sources that help here, come with Pro. Quick check: one page of bullets with a date on every number, plus the line to leave unsaid.
✅ The Fact-Check Loop, Start to Finish
Extract the claims
Paste the draft and let the tool pull every checkable line into a table with its exact wording.
Demand primary sources
Each claim gets the filing, study or original statement, never an article citing the source.
Open every link yourself
A 'supported' verdict is a lead. Click it and read the line that carries the claim.
Flag numbers, dates and names
These fail loudest in public, so check them separately from the prose claims.
Fix or cut the contradicted lines
A claim with no primary source gets rewritten as a question or removed.
11. The Earnings Call Autopsy Prompt
Ninety minutes is the honest by-hand figure for a 60-page transcript. The second pass with a search box finds the changed guidance line. Uploaded to Perplexity, it reads the file in about 2 minutes, leaving 15 minutes of review.
Prompt to copy
You are analyzing an earnings call transcript for [COMPANY], quarter [QUARTER]. Attached: [TRANSCRIPT FILE]. 1) List five commitments or decisions, each with the speaker and a timestamp. 2) Compare guidance with the prior quarter and mark what changed. 3) Pull five verbatim quotes that matter, with timestamps. 4) Note the questions asked twice and how the answer changed. 5) Name what management was not asked or did not address. Rules: quotes verbatim only, never paraphrase inside quotation marks, 'not addressed' is a valid finding, no interpretation of tone without a quote. Output: five short sections in that order.
The two minutes buy the first draft; the risk is a timestamp a little off or a quote with one word swapped. Search each quoted line in the transcript before it goes anywhere. A paraphrased sentence inside quotation marks is the failure this job hides.
What survives review is the useful part: five commitments with speaker and timestamp, guidance against the prior quarter, and the question asked twice. The line management was never asked is the finding.
12. How Six Outlets Framed the Same Story
A comparison table lands in about 50 seconds in Pro Search: six outlets, their headlines and dates, the framing words each used, and the facts all six share.
By hand, reading six pieces properly runs about an hour, and 12 minutes of that is the review you still do.
Prompt to copy
You are comparing coverage, not opinion. Event: [EVENT], window: [DATES]. 1) Take six outlets across the spectrum – [OUTLET 1] through [OUTLET 6] – and give the date and headline of each piece. 2) Show how each frames the cause and who is responsible, quoting the framing words. 3) List the facts all six agree on. 4) List the facts only one reports and whether it names a source. 5) Name what none of them covers. Rules: attribute every point to an outlet, use no evaluative adjective of your own, and flag a claim with no named source. Output: a comparison table, then a three-line synthesis of what a reader would believe from each.
Framing is not falsehood, so keep your own adjectives out and check that each framing point names the outlet that used it.
The lone-fact row is the one to open: exactly one outlet reports it, and the prompt says whether that outlet names a source. What none of them covers is the third result.
13. The Competitive Landscape Table Prompt
Prompt to copy
You are mapping a market for [MY COMPANY]. Market: [MARKET], read in [MONTH YEAR]. 1) Pick eight players, including at least two that are not the obvious ones. 2) For each: what they sell, who pays, pricing model, latest disclosed revenue or funding and its date, and one real differentiator. 3) Show where the offers overlap and where the gaps are. 4) Name who is most exposed to [RISK] and why. Rules: a link in every cell, 'not disclosed' for private numbers, a date on every price, no market-size estimate without a named source. Output: the table first, then a 150-word read of what the table shows.
That prompt maps eight players, two of them off the obvious list. Research mode returns the table in about 9 minutes, against roughly 180 minutes of opening pricing pages yourself. Review adds 30 minutes, spent opening cells and checking dates.
The catch: private firms publish revenue and funding on their own schedule, so some cells read "not disclosed." Every price carries the day it was read.
The total: about 39 minutes against three hours, and the 150-word read names who is most exposed to your stated risk.
💡 Tip: Ask for two non-obvious players by name. It is the difference between a landscape and a list of the usual suspects.
14. The Comet Multi-Tab Price Sweep
Comet Assistant is a browser agent that reads the tabs you already have open and reports back, so the setup is one afternoon of training rather than a new habit every purchase.
Comet ships for iPhone, Android, Windows and Mac, with no native iPad app, and on iOS it defaults to Google.
Open one tab per retailer, then paste the prompt into the Assistant panel. Six tabs, one item, and the rule is that nothing is read that is not on the page.
Prompt to copy
You are a careful price checker working in my open tabs, one per retailer. Tabs open: [SIX RETAILER OR PROVIDER PAGES], item [EXACT ITEM AND SIZE]. 1) For each tab, capture sticker price, availability, shipping cost, delivery date and return window. 2) Put it in one table with the page address and the time you read it. 3) Flag any page where the price or the stock changed while you were checking. 4) Tell me the cheapest landed cost, not the cheapest sticker. Rules: read only what is on the page, never invent a discount code, stop and ask me if a login, a payment or a CAPTCHA appears, and do not buy anything. Output: the table, one recommendation, and the one thing to re-check tomorrow.
The run takes about 4 minutes, against 35 minutes of calculator and shipping pages by hand; review is 10.
Agent runs count against a monthly allowance that is larger on paid plans, and pages can move mid-check, so the agent stops rather than logs in, invents a coupon or buys.
💡 PRO TIP
Set a spending limit before the agent runs
Personal Computer and Comet Assistant run multi-step tasks on your machine or in your browser. Switch off any saved payment method you do not need, and keep secret keys out of anything the agent can read.
15. The Downloads Folder to Cited Index Prompt
150 PDFs sit in a downloads folder with meaningless filenames, three duplicates and one superseded version. By hand, every file gets opened just to find out what it is, which is about 100 minutes.
You paste the prompt below into Perplexity's Personal Computer feature, which runs multi-step tasks on your own machine. Set it up once and it pays back every time a folder gets this messy.
Prompt to copy
You are indexing a working folder on my computer. Folder: [FOLDER PATH], all [NUMBER] PDFs inside. 1) Open each file and record title, publisher, date, a two-line summary, three keywords, and any deadline or dollar figure it contains. 2) Group the files by theme. 3) Flag duplicates and superseded versions, naming which file replaces which. 4) List what a folder like this should contain but does not. Rules: never summarize from the filename, open every file, write 'unreadable scan' instead of guessing, change nothing and delete nothing. Output: one index table sorted by theme, then a five-line note on what is missing.
This is where to hold back. Personal Computer needs a paid plan (Pro, Max or Enterprise), and scanned files can come back unreadable. If most of the folder is scans, run OCR first or skip this job entirely.
Quick check: the prompt says open every file, not trust the filename. The run takes about 15 minutes and review 20. What you get is one index table by theme, duplicates named against what replaces them, and a note on what the folder lacks.
Frequently Asked Questions
Do I need to pay for Perplexity to use these fifteen prompts?
Most run on the free plan at least once. The fact-check prompt in entry 4 splits across days because the free plan gives 3 Pro Searches a day. Entries that need Research mode, a file upload over the free allowance, or Personal Computer need a paid plan. Pro is $20 a month as of September 2026, according to CloudZero.
How often can I run Research mode and file uploads?
Perplexity's help center states the free plan includes 1 Research query a month as of October 2026. Paid plans lift it to 50 a month, and the highest plan to 500. File uploads on paid plans extend to 100 sessions a week, and 1,000 on the highest plan.
Is it safe to upload contracts and transcripts?
Redact names, account numbers and anything medical or financial first. On personal plans, chats may be used to improve the company's models unless you switch that off in settings. The contract prompts are not legal advice, and every quote still gets checked against the page.
Do I need to know how to code to use Comet or Personal Computer?
No. Comet Assistant reads the browser tabs you already have open, and Personal Computer indexes a folder on your machine. Both are set up through normal sign-in screens. Comet ships for iPhone, Android, Windows and Mac with no native iPad app.
Why does Perplexity still lose to Google for some things?
Perplexity's own CEO said Google does a much better job at navigational searches like finding a nearby restaurant or checking a sports score. A reviewer found Google faster for a specific site, a quick fact, a product page or a simple how-to. Keep Google as your default and use Perplexity for these fifteen jobs.
What happens if I hit a limit mid-task?
The task stops returning full answers until the limit resets or you upgrade. Perplexity's help center says the free plan fits light users and occasional answers. Pro fits frequent research and file analysis. Check Perplexity's pricing page for current limits before you commit.
Keep Google as Default, Give Perplexity a Second Tab for These 15
Google stays the default for navigational and one-fact queries: a nearby restaurant, a specific site, a sports score, a product page. Nothing in this list changes that. Perplexity earns a permanent second tab for the fifteen jobs that end in a sourced, checked deliverable.
The first move is free. Run the fact-check prompt in entry 4 on a real draft today, or the contract-obligations prompt in entry 2 on a redacted PDF. Both run on the free plan, and both show you the tool's behavior before you pay anything.
Only move to Pro at $20 a month once a job needs Research mode more than once a month or file uploads past the free allowance. Then keep the review minutes on the calendar, because those are the real cost of every entry here.