Free preview · no sign-up required

Perplexity: Intelligent Research and Instant Knowledge: full syllabus and an entire chapter

Search, verify, cite, and build your own knowledge base — to deliver research that holds up when someone clicks the link

Researching with AI is easy. The hard part is standing behind the research afterward — when someone opens the link you cited and the page doesn't say what you claimed. This course is about that second part, which is where the real work has moved.

The centerpiece is Perplexity, from the inside: Projects, search modes, Deep Research, Model Council, Comet, and the API, each with the real plan limit and the date that limit was read. But the course doesn't end there. Every question has a tool that answers it better, and you'll compare ChatGPT, Gemini, Claude, Copilot, Kagi, and Exa by task — not by brand.

The discomfort comes early, on purpose. The Tow Center study at Columbia tested eight AI search engines on 1,600 queries, and the best one got attribution wrong 37% of the time; in the same test, the paid version performed worse than the free one. The EBU study with the BBC evaluated over 3,000 responses across 18 countries and found significant problems in 45%. This isn't an opinion about AI: it's published measurement, with open methodology and a date. Module 1 is entirely about what to do with that information.

30 chapters · 6 modules · 18+ hours

What you learn, chapter by chapter

Module 1: Researching is asking, verifying, and standing behind it

What changes when the answer arrives ready-made with numbered citations. The three ways a citation can fail, what independent studies measured, what already happens to those who don't verify, and the six-step protocol the rest of the course uses.

  1. 1. The answer arrives ready: what it takes from you and what it gives back

  2. 2. A citation existing doesn't mean it supports the sentence

  3. 3. What independent measurement showed: 37% at best

  4. 4. 1,922 court decisions later: what happens to those who don't check

  5. 5. The six-step protocol, and the notebook that records the verdict

Module 2: Perplexity from the inside — and the expiration date of every number

A module dated on purpose: feature names, plans, limits, and models changed three times in twelve months. Projects, search modes, what the company stopped publishing, the model list, and the Model Council, with the reading date next to every number.

  1. 6. Collections became Spaces, which became Projects: the name that ages

  2. 7. The search modes: Search, Pro Search, Deep Research, and Learn Mode

  3. 8. When the provider deletes the number from the site, the number becomes a variable

  4. 9. Choosing the model inside Perplexity — and the documentation that contradicts itself

  5. 10. Model Council: paying to see where models disagree

Module 3: There's no best tool: there's the right one for the question

ChatGPT, Gemini, Claude, Copilot, Kagi, and Exa compared by task and by published evidence — who cites better, who goes deeper, who sees your documents, who respects your privacy, and who publishes how long it takes.

  1. 11. ChatGPT with search: deeper, less precise in attribution

  2. 12. Google: the largest scale and the worst attribution score

  3. 13. Claude with search: clean citations, and an evidence gap

  4. 14. Copilot: your own documents — and the feature that just died

  5. 15. Kagi and Exa: paying to be the customer, and the only one that publishes the clock

Module 4: Deep research: the report that looks ready

How to ask for deep research, how to read the report with calibrated skepticism, how much it costs and how long it takes in each service, and what you're signing off on when you deliver — because the volume of text and the density of citations create authority that accuracy doesn't sustain.

  1. 16. What deep research does that you wouldn't do in three days

  2. 17. Writing the request: scope, date range, and source criteria

  3. 18. Reading thirty pages with calibrated skepticism

  4. 19. How much it costs and how long it takes, service by service

  5. 20. From report to delivery: what changes when your name goes on it

Module 5: Your own knowledge base: when the answer has to come from your material

Gemini Notebook and Perplexity Projects side by side: what fits in the free tier, what a closed knowledge base solves, what it doesn't solve, and how to build and maintain the base for a real project without turning it into a dumping ground.

  1. 21. Gemini Notebook, formerly NotebookLM: the base that doesn't invent outside your material

  2. 22. Perplexity Projects: your sources plus the live web

  3. 23. What your own knowledge base can't fix

  4. 24. Building the base for a real project: selection, name, and date

  5. 25. Keeping the base alive: what goes out, what gets rechecked, who gets in

Module 6: Where verifying is an obligation, not diligence

Legal, healthcare, journalism, and academia already have written rules and enforced sanctions. What each requires, what changes with the European AI Act, and how to build a system that survives the next feature rename.

  1. 26. Legal: the rule already exists, and so does the penalty

  2. 27. Healthcare: inform the patient, record in the chart, you decide

  3. 28. Publishing: what the European AI Act started requiring in August

  4. 29. Academic: AI is a search tool, never the source

  5. 30. Building a system that survives the next rename

Chapter 1 in full

This is the complete chapter, just like the one inside the course — text, images, and videos.

Overview

You type a question. In seconds, the screen shows a paragraph answering it, with a numbered citation at the end. It feels like the research work is done. It isn't — it just moved somewhere else.

The old search engine gave you a list of links, and you judged which ones were worth opening. AI gives you the conclusion already judged. The judging work didn't disappear: it was done earlier, by someone else, and you just need to check whether their judgment holds up. Those who don't move their own work to after the answer deliver someone else's judgment as if it were their own. This chapter installs that: the ready answer is the beginning, not the end.

Key Concepts

Before, the work was in the choosing. You read three titles, opened two links, compared. The decision was yours. Now Perplexity makes that choice for you and hands you the final sentence. What's left for you is verification: is the answer right? Does its source say what it claims?

Perplexity responds with paragraphs and numbered citations. Each number is a link to the original source — but the link doesn't guarantee the source supports the sentence.
Perplexity responds with paragraphs and numbered citations. Each number is a link to the original source — but the link doesn't guarantee the source supports the sentence.

The numbered citation creates an illusion of security. It's there, with a link, looking like proof. But the citation only points to an address. It doesn't guarantee that the content at that address supports the sentence you read. Perplexity itself writes on the Pro Search page: "it's crucial to validate information by referencing the sources linked in your answer" (07/21/2026). Even the provider tells you to validate.

The intent of the answer is to give you a conclusion, not a map. You need to open the source to know if the conclusion is yours.
The intent of the answer is to give you a conclusion, not a map. You need to open the source to know if the conclusion is yours.

The mental model is this: AI is an assistant that read faster than you, but that may have misunderstood. It's not a reliable witness; it's a witness that talks a lot. Your job isn't to repeat what it said. It's to judge whether it could have said that. That's new work, which happens after the answer, and no tool does it for you.

Execution Flow

  1. Open Perplexity on the free plan (Standard). In the search field, type a concrete question about your industry. Use "packaging factory Joinville" if you want to test now. Press Enter.
  2. Read the entire answer before clicking any citation. Mentally mark the sentences you'd use in a report. Those are the ones you'll verify.
  3. Click the first numbered citation in the sentence you marked. A tab opens with the original source. Note the URL, the site title, and the publication date.
  4. Compare the source text with the answer's sentence. Ask: does the source say exactly this? Or does it say something similar, but different? Often the source is about another topic, or another time.
  5. Record the verification result. If the source supports the sentence, note the source date next to the number. If it doesn't, cross out the sentence from the answer. That record is your working material.
See the full flow: question, answer, clicked citation, opened source, and sentence-by-sentence comparison.

Applied Scenarios

Renata prepares a report for the commercial board of a packaging factory in Joinville. She asks Perplexity: "What's the price trend for corrugated cardboard next quarter?" The answer brings a paragraph with citation 3. Before, Renata copied the paragraph and pasted it into the presentation, with the citation number. Now, she clicks citation 3. The source is a 2019 blog about something else. The answer's sentence isn't there. Renata crosses out the paragraph and looks for another source. The presentation doesn't repeat the mistake — and the director doesn't open a broken link in the meeting.

Renata's flow: the answer brings a citation, but the source doesn't support the sentence. She crosses it out before presenting.
Renata's flow: the answer brings a citation, but the source doesn't support the sentence. She crosses it out before presenting.

Second scenario: Renata needs to answer legal about a food packaging labeling regulation. Perplexity responds with a law and citation 5. Before, Renata passed along the law number without opening the source. Now, she opens citation 5: it's a summary from a news site, not the original law. The law number is wrong. Renata looks up the regulation on the official government site and corrects it. Legal gets the right information — and the factory avoids a fine for incorrect labeling.

Common Mistakes

  • Not clicking the citations. You read the answer, trust the number, and don't open any links. The citation becomes decoration.
  • Opening the source and not comparing sentence by sentence. You see the link works and assume everything is fine. The link can work and say something else.
  • Confusing the answer's date with the source's date. The answer was generated today, but the source could be from 2015. For cardboard prices, that changes everything.
  • Copying the answer without marking what is a claim and what is an opinion. AI mixes both in the same paragraph.
  • Thinking the free plan isn't good enough for serious work. It gives practically unlimited basic searches, 3 Pro Searches per day, and 1 Research per month (official pricing page, updated 22/07/2026). The limit isn't an excuse not to verify.
Common mistake: the citation points to a link that opens, but the content doesn't support the sentence. The link working isn't the same as the source confirming it.
Common mistake: the citation points to a link that opens, but the content doesn't support the sentence. The link working isn't the same as the source confirming it.

Pro Tip: Treat each citation as an invitation to check, not as proof. Before using any number in a report, open the source and look for the number in it. If you can't find it, the answer might even be right, but you don't know that. The citation is the start of verification, not the end.

The tip in practice: how to find the number in the source and what to do when it's not there.

Practical Exercise

Open Perplexity on the free plan. Ask a question about your company's industry — for example, "what's the average profit margin for a packaging factory in Brazil?" Wait for the answer. Pick a sentence that has a numbered citation. Click the citation, open the source, and verify: does the sentence from the answer appear in the source? Write a paragraph of no more than five lines saying whether the source supports the answer and why. You're done when you can answer "supports" or "doesn't support" with a justification based on the source's text, not your opinion. Save that paragraph — you'll use it in chapter 2.

Implementation Checklist

  • I can explain why the ready-made answer doesn't eliminate the research work, it just moves it later.
  • I can open a Perplexity citation and identify the URL, title, and date of the source.
  • I can compare the sentence in the answer with the source text and say whether there's an exact match.
  • I know to record the source's date next to every number I'm going to use.
  • I can distinguish between "the source exists" and "the source supports the sentence."

Chapter Summary

  • The ready-made answer with a numbered citation looks like the end of research, but it's the beginning of verification.
  • The judging work has moved: now you judge after, not before.
  • The citation points to the source, but doesn't guarantee the source supports the sentence.
  • Perplexity's free plan has limits — 3 Pro Searches per day and 1 Research per month — that don't exempt you from checking.
  • In the next chapter, you'll see the first way a citation fails: it exists, but doesn't support the sentence.

---

Previews of similar courses

View all courses →
FAYAI