Perplexity AI for Teachers: A Sourced Project Research Flow, and Teaching Pupils to Check Citations

A Secondary 4 Citizenship and Social Development project comes in with one line under references: "Source: AI". That is not laziness. The tool the pupil used simply does not show its sources. Every sentence Perplexity produces carries a clickable citation number, which makes it the best available tool for teaching verification — not because it is cleverer, but because it puts the provenance in front of the pupil so there is something to teach. Before recommending it to a class, read one clause: users must be at least 13, so primary pupils do not qualify, and there is no teacher monitoring of any kind.

What this tool actually is

Perplexity is an answer engine. You ask a question, it searches the web, synthesises a reply, and cites a source behind each statement. The difference from a chatbot is the starting point: a chatbot generates from model memory and then decides whether to search, while Perplexity searches first and synthesises second.

It comes in three parts: the web and mobile apps, the Comet browser (Chromium-based, with the AI assistant in a sidebar), and the Sonar API for developers.

Who is it really built for? Knowledge workers who need sourced answers quickly — analysts, journalists, researchers. Education is not the design target: no classes, no assignments, no teacher monitoring, no student safety gating. There is an Education Pro plan, but it is a discounted subscription for individually verified students and educators, not a school product.

By school phase, senior secondary fits best — Citizenship and Social Development, History, Geography and English project work, and any assignment that requires a reference list. Junior secondary works under supervision. Primary does not meet the age terms.

Key features

As officially described:

• Cited answers — every response carries source links you can open and check against the original.
• Pro Search — deeper multi-source searching with a choice of which model answers.
• Model selection — Pro users switch models at the bottom left of the prompt box, including the default Best that picks automatically, OpenAI's GPT models and Anthropic's Claude models among others.
• Deep Research — runs multiple rounds of searching on a topic and produces a research report.
• Learn Mode — a study-oriented mode Perplexity lists among the Education Pro inclusions.
• The Comet browser — free, with Ask and Summarize buttons for the current page, support for most Chrome extensions, page translation, built-in ad blocking, and natural-language browser commands for tab management.
• Model Council — a Max feature that asks one frontier model each from Anthropic, OpenAI and Google simultaneously, then has a synthesiser model resolve a single answer that shows where the models disagree.
• Enterprise administration — organisation-level management, collaboration, security and privacy controls on the Enterprise tiers.

Full tutorial

The scenario: you teach Secondary 4 Citizenship and Social Development and the unit is healthcare and population ageing in Hong Kong. Pupils have three weeks to produce a 1,500-word project citing at least six reliable sources. You want two outcomes at once: that they finish, and that they learn to judge whether a source is trustworthy.

Step 1 — run it yourself before deciding whether pupils should. Register a free account with your school email. The official pricing page describes the free plan as good for limited daily usage with cited answers. Run your own topic and look at what it cites — the Census and Statistics Department and the Hospital Authority, or a content farm. That answer determines what you will be teaching pupils to look for.

Step 2 — create a dedicated space. In the left panel create a Space (some versions call these Projects) named "S4 CSD healthcare and ageing", and put the task, the set readings and the marking rubric inside it so subsequent questions run in that context.

Step 3 — model one good question. Project the block below and explain each line as you go.

Topic: pressure on Hong Kong's public healthcare system from population ageing, 2015 to 2025

Answer the four points below. Every point must cite its source, and prefer Hong Kong government departments, statutory bodies or university research:
1. The change in the proportion of the Hong Kong population aged 65 and over, with specific years and figures.
2. The change in waiting times for new specialist outpatient cases in public hospitals, with specific specialties and median figures.
3. Three relevant government policies or schemes introduced in this decade, one sentence each.
4. Two main points of contention identified by academics or policy commentators, with one source for each opposing position.

Constraints:
- If an official source cannot be found for a figure, write "no official source found" rather than substituting a figure from secondary reporting.
- State the statistical year for every figure, and do not place figures from different years side by side as a comparison.
- Label each source as a government document, a news report or a commentary piece.

Step 4 — hand out the citation-checking sheet, because this is the actual lesson. The commonest pupil error is treating an AI summary as fact. Require one row per source for all six:

Citation check sheet, one row per source, submitted with the project

Source number:
Title and website:
Source type: government document / statutory body / academic research / news report / commentary / other
Publication date:
Did I open the original: yes / no
Where in the original did I find the AI's claim: (copy one sentence of the original)
Do the figures in the original match what the AI quoted: match / do not match (explain)
Does this source hold a position: yes / no (explain)
If I quote this sentence in my project, how would I rewrite it:

Rule: any source where "did I open the original" is answered no may not be cited in the project.

Step 5 — make pupils find one error themselves. This works better than any explanation: require every pupil to locate at least one place where the answer departs from the original — an out-of-date figure, a news report presented as official statistics, a district figure described as territory-wide. Whoever finds one presents it to the class. That is information literacy done live.

Step 6 — separate gathering from writing. State it explicitly: Perplexity is for stage one, locating sources and evidence, and the analysis and argument must be written by the pupil. Put it in the rubric and explain why in class — not because AI is forbidden, but because the project assesses their judgement.

Step 7 — if you use Comet, brief them first. Comet is free and its Summarize and Ask buttons are genuinely useful on long government reports. But it is a full browser that sees a pupil's browsing history and signed-in sessions, so anything installed on school machines should go past the IT coordinator first.

Pros and cons

Pros

• The citation interface is the best in its class. Source numbers behind each statement open instantly, and for teaching verification that matters far more than the answer itself.
• Search first, synthesise second, which makes it markedly more accurate than memory-based tools on current affairs, statistics and policy.
• The free tier is genuinely usable, with cited answers not behind a paywall.
• The Comet browser is free, supports Chrome extensions and page translation, and helps pupils working through long English reports.
• Comparatively transparent about models: Perplexity publishes the third-party models it uses and their terms, covering OpenAI, Anthropic, Meta, xAI and Google, and notes that its own Sonar model is fine-tuned from Meta's Llama 3.1 70B.
• Perplexity explicitly mentions that schools, universities and not-for-profits may qualify for Enterprise discounts, which gives a school something to negotiate.

Cons

• A minimum age of 13, so primary is out, and minors above 13 need a parent or guardian to accept the terms, which means a consent process for the school to run.
• No teacher monitoring, no classes, no assignments, no student safety gating. It is a search tool, not a teaching platform.
• We could not find a student data protection agreement aimed at schools. Before pupils log in with their own accounts, get written confirmation of data use, retention and whether inputs improve the models.
• Education Pro's price and eligibility could not be verified officially. Third parties report roughly US$10 a month and eligibility possibly limited to post-secondary institutions, which may exclude school teachers.
• No verified Traditional Chinese (Hong Kong) interface and no Cantonese features.
• A citation is not a guarantee of accuracy. It will faithfully cite an unreliable source, or place figures from two different years side by side. That is precisely why the checking sheet cannot be skipped.
• Max at US$200 a month is irrelevant to schools, and Enterprise per-seat pricing gets expensive at 100 staff.

Pricing and fee breakdown

The number after multiplying by 100. Enterprise Pro at US$400 per seat per year is US$40,000 a year for 100 staff, roughly HK$312,000 at 7.8 to the US dollar — and that covers no student seats at all. Adding pupils adds per-head cost again. Perplexity does say schools and not-for-profits may qualify for discounts, but the size of the discount is unpublished, so the budget cannot be fixed before negotiation.

The fragmentation risk of individual subscriptions. Without a whole-school purchase, what usually happens is that a few keen teachers pay US$20 a month themselves while pupils use the free tier or a parent-funded account. That creates three problems: pupils sitting the same assignment with unequal tools, no single data policy for the school, and teacher out-of-pocket spending that cannot be claimed under a grant.

What happens when the free tier runs out? Nothing breaks, but the deeper research features and top models are unavailable and daily limits apply. The practical teaching consequence is that running Deep Research with 30 pupils simultaneously is not viable, so demonstration-led teaching is the safer design.

Compared with the alternatives

Honestly: if your school already has Google Workspace for Education, Gemini Notebook is the better junior secondary choice precisely because it cites only teacher-supplied material. But for teaching pupils to judge sources on the open web, Perplexity's interface is still the best teaching aid.

Side by side with Edor.ai

Where Perplexity genuinely wins: the citation experience is better than ours and better than most education platforms. Clickable sources per statement, a clean source list, traceability back to the paragraph — that is its refined core. Our knowledge base follows the same principle in a different direction: we cite the school's own documents rather than the whole internet. For teaching information literacy, Perplexity is the better instrument; for school-based teaching, a school knowledge base is the right tool.

Why a school still needs a school-based platform

The gaps Perplexity leaves are exactly the ones a school cannot ignore.

No gating on the student side. The age-13 clause, no content filtering, no teacher monitoring — a secondary pupil using it unsupervised leaves the school unable to know what was asked or to intervene when something inappropriate comes back. See student AI safety, gating, PII and monitoring.

It does not know your teaching materials. Beyond project work, what pupils most often need is "what is the key point of this lesson", and that answer should come from the notes their teacher released, not from whatever the web offers. That is the problem a school knowledge base with RAG exists to solve.

Per-seat pricing works against schools. Enterprise Pro for 100 seats is US$40,000 a year with no student access, and the discount is unpublished. A per-school annual fee covering staff and pupils together, with usage reports attached, is the clearest structure for an EDB funding submission.

Data agreements. Schools have to handle parental consent, retention and export, and we could not find a student data agreement aimed at primary or secondary schools, so this is a question the school must put in writing. The evaluation questions in 12 questions to ask before buying a school AI platform cover the ground.

Edor.ai is built for Hong Kong schools, teachers and students, it is the only thing we do, and we serve them with our best effort: nine modules covering lesson planning, AI tutoring, marking, differentiation, oral practice, student data analysis, administrative workload, the PIE planning cycle and a school knowledge base, with data belonging to the school, free one-click full export at any time, and an on-premises option. None of this argues against using Perplexity. It belongs in information literacy lessons, and the more it is used there the better.

Takeaway

Perplexity is the tool we would recommend to a Hong Kong teacher for teaching verification: clear citations, a usable free tier, and a free browser. Putting it into a course means handling three things — the age-13 clause and parental consent, the absence of teacher monitoring, and getting the student data policy confirmed by the vendor. The most important point is that a citation is not proof, which is why the checking sheet is worth more than the tool. To bring the same "answers must show their sources" discipline to your own materials, read the school knowledge base and RAG; to compare tools, start with the 2026 overview of AI tools for teachers.