Perplexity searches the web and places citations beside its answer. That does not make the answer a source; it makes the answer a map. You still need to open each link and decide whether its evidence supports your claim.

You will turn a content idea into answerable questions, run a broad-to-focused search, audit cited pages, and build a research brief. It will separate evidence from interpretation and leave honest blanks where Pakistani data is unavailable.

// concept

Turn a Topic Into an Answerable Question

“What does my audience care about?” is too broad. An answerable question names the audience, period, behaviour, and evidence.

Use this pattern:

// prompt — copy me4 lines

Among [specific audience] in [place/channel], what [questions/problems/decisions]
about [narrow topic] appear in [named evidence types] during [time range]?
Return each finding with its source and label any evidence gap.
Do not estimate popularity unless a source measured it.

Research in three passes. Here are three copyable example queries for a creator planning content for Pakistani home-based food sellers:

// prompt — copy me16 lines

# 1 — Broad discovery
What questions do Pakistani home-based food sellers appear to ask about taking
orders through WhatsApp? Look for themes in public, attributable sources from
2024 onward. Return source links and do not infer how common a theme is.

# 2 — Focused research
For a home-based food seller in Karachi taking orders through WhatsApp, what
order details can reduce avoidable misunderstandings with customers? Prefer
official platform guidance and original Pakistani sources. Separate documented
guidance from your inference, and identify missing Pakistan-specific evidence.

# 3 — Verification
Verify only this claim: “A WhatsApp order form should capture delivery area,
requested date, item quantity, and allergy information.” For each element, show
the original source passage or mark it as a practical recommendation rather
than a sourced fact. Do not use another AI summary as evidence.

The first query discovers themes, the second narrows the decision, and the third tests one publishable statement.

// concept

Search in Perplexity, Then Leave Perplexity

Start a Perplexity search and paste the broad query. The interface may offer Search, Pro Search, or Research modes. Names, limits, and plan access change, so check the official plan page. Standard Search is enough: run smaller queries and inspect each citation. If Research is available, use it for discovery, not as permission to skip the audit.

After the answer appears:

Open the citation attached to the sentence you may use.
Find the claim on the page; topical relevance alone is not support.
Record publisher, date, URL, evidence location, and what it establishes.
Prefer official documentation, government data, or research with visible methods.
Follow secondary coverage to the original report or dataset and log that.
Cross-check important claims independently. Two articles repeating one release are one evidence chain.

Reject undated pages, absent evidence, mismatched populations, and summary-only chains. Never paste a private brief, unredacted DM, phone number, CNIC, order sheet, or paid report into a public search tool. Use authorised, redacted inputs and review Perplexity’s current data settings.

// concept

Keep a Citation Ledger, Not a Link Dump

Add accepted sources to the source log created in Lesson 1.2. One row should support one precise statement.

ID	Candidate statement	Source and date	Evidence location	Status	Notes
S1	What the source directly shows	Original URL · date	Heading, table, or page	Evidence	Safe to paraphrase narrowly
S2	Your interpretation of S1	Link back to S1	Your analysis	Inference	Must be labelled, not cited as fact
S3	What you still need to know	None	—	Unknown	Ask audience or remove claim

Use three labels consistently:

Evidence: the source directly supports the statement for the named audience and period.
Inference: your reasoned interpretation of evidence, written as “this may suggest,” not as a measured fact.
Unknown: no suitable source was found, the data is too old, or the source does not cover Pakistan.

The brief needs the audience decision, three queries, a five-to-eight-row ledger, cautious inferences, unresolved questions, and one supportable content angle. Keep citations attached when the brief becomes a script or carousel.

// worked_example

Worked Example

// prompt — copy me5 lines

For Pakistani home-based bakers taking customer orders through WhatsApp, what
information should an order-confirmation checklist contain? Prioritise official
WhatsApp guidance and original Pakistani sources. For every recommendation,
label it Evidence, Inference, or Unknown. Include links and do not claim that a
practice is common unless a source measured adoption.

The first draft suggests product, quantity, delivery details, payment status, and dietary notes. It also claims that “most order disputes come from missing delivery details.” Its citation opens a marketing article that links to a second roundup; the second gives no dataset and links back to the first. This is circular, not evidence.

She deletes “most order disputes” from the brief. She then asks a verification follow-up:

// prompt — copy me4 lines

Find an original source that measured causes of WhatsApp order disputes among
Pakistani home-based food sellers. Exclude blogs that cite each other. If no
source with a visible method and relevant sample is available, answer “not
established” and suggest a first-party question I can ask my own customers.

In this sample scenario, no suitable original study appears, so the claim becomes Unknown. The creator labels the checklist as practical advice and writes: “Confirm delivery area and time in writing before accepting the order.” It is not presented as a national finding.

The brief retains verified product guidance, practical recommendations, and one first-party question: “Which detail did you have to clarify after placing your last order?” An optional poll can collect answers without names or order details. The content idea survives with a narrower claim.

// failure_cases

Failure Cases to Diagnose

7 cases to diagnose

The query asks for “top concerns” without a measured dataset.
Replace “top” with “recurring themes found in these named sources,” or request frequency evidence explicitly.
A citation opens a page that discusses the topic but not the claim.
Search within the page for the key phrase; if it is absent, reject the citation.
Five links repeat the same announcement.
Trace them to their common origin and count the origin once.
A global survey is presented as Pakistani audience evidence.
Label it global context and seek a Pakistan-specific source or first-party observation.
Perplexity’s wording enters the script as a quotation.
Quote only text you read in the original source and confirm its context and reuse conditions.
A polished gap is silently filled.
Put it under Unknown and design a poll, interview, or comment-mining task rather than asking AI to guess.
Private audience material is searched.
Stop, redact identifiers, obtain permission where needed, and use only the minimum authorised text.

// pakistan_angle

Pakistan Angle

Pakistan-specific audience evidence is often thinner than US or UK evidence, especially for informal sellers, regional-language audiences, and WhatsApp commerce. Search in English and in the language your audience uses: “home baker order maslay,” “WhatsApp order confirm kaise karein,” and Urdu-script variants can reveal different public vocabulary. Treat Karachi, Lahore, smaller Punjab cities, Khyber Pakhtunkhwa, and rural audiences as separate hypotheses. If a source covers only urban banked consumers, do not extend it to all Pakistanis.

Prefer original Pakistani sources when the claim depends on local policy or behaviour: the Pakistan Bureau of Statistics, State Bank of Pakistan, a regulator, or a report whose Pakistan sample and method you can inspect. When no source answers the question, say so and gather a small first-party signal from your own consenting audience. Save source pages or notes for offline review during load-shedding, but do not download or scrape private groups. Redact phone numbers, addresses, CNICs, payment screenshots, and identifiable DM details before any AI-assisted analysis.

// hands_on

Hands-On Exercise

6 steps

Build a cited audience-research brief for one content piece.

Name one narrow Pakistani audience, one channel, and one decision your content will help them make.
Write broad, focused, and verification queries using the pattern above.
Run them in Perplexity. Check the current plan page if a suggested mode is unavailable; continue with Standard Search if needed.
Open every citation attached to a claim you may publish. Reject summaries, circular chains, missing passages, and mismatched populations.
Create a ledger with at least five rows labelled Evidence, Inference, or Unknown.
Draft a one-page brief containing the audience, research questions, supported findings, unresolved questions, source links, and one evidence-safe content angle.

// completion_rubric

Completion Rubric

6 checks — tick as you verify

0/6

// sources

Sources

5 official sources — check every claim yourself

// check_yourself

Check yourself

4 questions · answers and options are taken word-for-word from this course

0/4

1 / 4 · diagnose
Your work shows this failure mode: “The query asks for “top concerns” without a measured dataset.” What does the lesson tell you to do about it?