How ExamExplained is built: the AI-first methodology (2026)

What this page is for

ExamExplained is a study library for senior secondary courses that is openly AI-built. Guides, quizzes and explainers are drafted by Claude Opus, Anthropic's AI model. We are also upfront about what we don't do yet: no page has been reviewed by a named teacher or examiner before going live. This page explains how that works, what AI is good at, what AI is bad at, and how we handle the gap.

If you are a student, teacher or parent deciding whether to trust the site, this page sets out what we do and what we don't.

Why we did this

A traditional study resource is bottlenecked by tutor time. A single tutor can write one good maths guide, then needs another month to write a Biology guide, then the syllabus updates and the first one needs rewriting. The result is the same pattern across the industry. A handful of strong guides in the most popular subjects, and thin, outdated, or missing coverage everywhere else.

We took a different approach. We pointed the most capable AI available at the entire problem.

Specifically, Claude Opus works from the published syllabuses and study designs of the official exam authorities and, where they are used for a page, public past papers, marking guidelines and admissions-centre scaling reports. Claude drafts the subject hubs, topic guides, practice question banks and system explainers you see on the site from that material.

The structural advantage is breadth and consistency. Claude does not get tired in week 12 and does not have a favourite subject. That is why this resource can cover what it covers.

What Claude is actually good at

In our experience building this site over the past year, Claude is reliably strong at the following.

• Summarising long official documents. Reading a 70-page official syllabus and producing a 1,500-word guide that captures the structure, the assessment, and the common exam patterns. This is what AI does best.
• Cross-curriculum comparison. Explaining how one curriculum's maths course maps to another's, or how the same subject is assessed differently across exam boards. A human tutor specialised in one system struggles with this.
• Maintaining consistency. Across the whole library, the tone, structure, and conventions stay aligned in a way that is hard to achieve with many different human authors.
• Synthesising historiography. For Modern History and English, summarising the major schools of interpretation (intentionalist vs structuralist, orthodox vs revisionist) is something Claude does well from the published literature.
• Producing many practice questions in the style of past papers. Pattern-matching against published exam papers to generate similar but not identical practice items.

What Claude is not good at

This is where you, the reader, need to verify against the official source before it matters for your mark. The site does not have an individual tutor checking every page before it ships, so the responsibility for the next list lives with you.

• Specific current statistics. Scaling figures, mean scaled marks, cut-off ranks all change year-to-year. Claude will sometimes confidently produce a figure that was true two years ago. Treat any specific figure on the site as a starting point and check the most recent published report from the official authority or admissions centre before relying on it.
• Recently changed syllabus content. Exam authorities update their syllabuses, study designs and prescribed texts on rolling cycles. If a change is very recent, Claude may not know about it. Where we know of a change, we flag it in the relevant subject hub and update pages as we go.
• Hyper-local context. Whether your school weights the Module B essay more or less, or which textbook your teacher actually uses, or the unspoken culture at your selective school, Claude cannot know. For that, you want a tutor or a senior student.
• Marking your specific draft. Generic feedback on essay structure, yes. Specific feedback on your sentence-by-sentence work, no. That is a human tutoring job.
• Subtle pedagogy choices. Whether to introduce a topic with intuition first or with formal definitions, whether to drill harder or to lecture longer for a specific student, these are tutor judgements that benefit from knowing the student.

When you read a page on the site, that page is good at the first list and is not the right tool for the second list.

The pipeline

Every page ships through the following pipeline before going live.

1. Source ingestion. We give Claude the relevant primary sources, syllabus documents, past papers, marking guides, and any scaling data.
2. Draft generation. Claude Opus produces a draft in our standard guide format (frontmatter with FAQs, body with sub-headings, named examples, worked traps, exam-pattern section).
3. Validation. Automated validators check the frontmatter schema, banned punctuation, minimum body length, duplicate pages, placeholder syllabus references, callout structure, the quick-answer block on every dot point, and a few other structural rules. Drafts that fail get fixed and re-run.
4. Publishing. The page is committed to git and deployed. The URL is added to the sitemap.
5. Post-publish fixes. If we find errors after publishing, or syllabus updates land, the page is corrected, the last-updated date is bumped, and the new version replaces the old at the same URL.

What this pipeline does not yet include is a teacher or examiner reading each draft against the syllabus and for factual accuracy before it goes live. That would be the gold standard, and we don't claim to do it. We are setting up a reviewer program: when a named reviewer checks a page, the page will say who, with their credentials and the date, and link to their profile on the reviewers page. Until a page says that, it has not been reviewed by a person.

How we keep pages fresh

Every page has a last_updated date in its frontmatter. That date drives four things on the site. (1) The "Last updated" line in the "About this page" box at the foot of every dot point and guide. (2) A public What's new page listing the 50 most recently updated study pages, grouped by week, plus a /whats-new.json feed for machine consumers. (3) The article:modified_time tag and the structured-data dateModified in the page head, so Google and other crawlers can see the date without parsing the body. (4) A pnpm audit:freshness script that walks every content file, reports how many pages fall into each age bucket, and lists the oldest files due for a refresh. Stale pages are refreshed by re-running them through Claude Opus against the current syllabus and marking guidelines.

The dot-point depth layer

Underneath the subject hubs and the topic guides sits a deeper layer: one page per official syllabus dot point. Exam authorities structure their syllabuses and study designs as a hierarchy of modules, inquiry questions, and individual dot points, the smallest atom of "you must know this". The site now has a dedicated answer page for each dot point we have shipped, with a focused explanation, optional worked past exam questions, and cross-links to related dot points in the same module. Coverage differs by subject: some subjects have a page for every dot point, others are still being filled in, and each subject hub shows what is there. Every dot point page is drafted by Claude Opus from the official syllabus wording and, where used, public marking guidelines and past papers, then validated against our schema and published. Browse them from any subject hub via its syllabus link.

What we capture, what we don't

Reading the site requires no account, no email, no cookies. The pages are pre-built static HTML and there is no chatbot or generative endpoint at runtime. A narrow set of things does leave the browser, each for a specific reason, and the honesty rule on the site applies to this list too.

• Anonymous crash envelopes to Sentry when something throws an error. We use this to find and fix bugs faster. Session replay is not enabled.
• Cookieless page-views and click events to PostHog (EU Cloud). We use this to see which topics need more work and which navigation paths confuse readers. Persistence is set to memory only, so PostHog does not write to your browser's localStorage or set cookies.
• Aggregate page-view counts to Vercel Analytics and Cloudflare Web Analytics, and to Plausible when the deployment configures it. These are cookieless; we see counts and referrers, not individual users.
• Form submissions when you choose to write to us from /get-help. The message, your name and email (when supplied), and the source URL are written to a private Supabase table and trigger a notification email so a person can reply. The intake table is not publicly readable.

Everything else stays on your device. Your study state - subjects, planner, bookmarks, quiz attempts, flashcards seen - lives in your browser's localStorage and is not sent to our servers. No quiz answers, no reading history, no progress data leaves the browser unless you submit a form.

For the line-by-line breakdown, including your data rights and the privacy law that applies, see the privacy notice.

How to use the site, given all of this

If you are a student, use ExamExplained the way you would use a very strong textbook plus a generic study companion. Use it to understand the structure of your subject, to memorise named examples, to drill practice questions, to plan study, to understand the scaling and ATAR system. Do not use it to verify a specific statistic for an assessment without checking the source. Do not paste a page into your essay; copying a study guide is plagiarism, and it is not what your teacher is marking you on anyway.

When you want the things AI cannot give you, feedback on your specific draft, a tutor who knows your school, accountability week-to-week, exam-day strategy for your situation, that is when you book a session with a tutor.

Versions and dates

Content is generated with the current version of Claude Opus; older content was generated by earlier versions of the same model family and is refreshed over time. The last-updated date on each page records when it was last regenerated or fixed.

If you find an error, we want to know. Use the question form at /get-help with the page address and what looks wrong, or open an issue on the public GitHub repository.