How to Build an AI Study Guide From Your Course Files

Still choosing a tool? Our AI study assistant guide has a one-chapter test for that. This article assumes you have picked one and want the output to hold up on exam day.

Why "upload everything and summarise" fails

A typical module folder holds twelve lecture decks, two textbook chapters, a spreadsheet of lab values and a recording or two. Upload all of it with one prompt and you get something long and fluent. What you do not get is proof that it covers what the exam can ask. Three things go wrong again and again:

• Silent omissions. A model condensing many files has to drop something. It tends to drop short slides, tables and exceptions, which are exactly what exam questions love.
• Blended conflicts. Lecture 8 defines a term one way, the textbook another. The draft merges them into one smooth sentence that matches neither.
• File order as structure. The guide follows lecture 1, 2, 3 instead of the questions you will be asked, so it helps with reading but not with answering.

None of this means AI is useless here. It means generation is the first pass, and the checking is on you.

Step 1: build a source inventory

Create a small table with one row per source. Five columns are enough: name, type, lecture number or date, expected topics, and exam relevance (high, medium, low). Add the syllabus and any "what's on the exam" email from the lecturer as separate rows, because they define scope.

Then flag two things by hand. First, conflicts: if two sources disagree, write it down so the guide keeps both versions instead of averaging them. Second, emphasis: for recordings, note timestamps where the lecturer said "this comes up every year". Ten minutes here saves an hour of correcting later.

What each file type needs before you upload it

Step 2: organise around what the exam asks you to do

Build the topic map from the syllabus, then decide the "shape" of each topic from the exam format. A law student organises by rule, exception and application. A medical student often uses definition, mechanism, clinical picture, tests and management. An engineering student may use assumptions, model, derivation and failure conditions.

Give every topic a stable label and reuse it everywhere: in the guide, the flashcards, the quiz and your oral practice. Stable labels are what later let you see that "Bayes theorem" has a summary and six cards but no applied question, while "sampling bias" sits in the syllabus and in no generated material at all.

Step 3: generate a draft with source anchors

Ask for the topic map first and the full guide second. A long first output tends to become the structure by default, even when it is the wrong one. These are the prompts I use; adapt the bracketed parts:

The "not in sources" instruction matters more than it looks. It turns a confident guess into a visible gap you can fix.

Step 4: run a coverage matrix

Use the syllabus topics as rows. For each row, run four checks:

• Present: does the guide address the topic at all?
• Supported: can you point to at least one source location for it?
• Usable: does it support the task the exam sets (a calculation, a comparison, a case)?
• Practised: is there at least one retrieval question or application task for it?

Worked example: a renal physiology module

Example Here is what the matrix looked like for a six-topic excerpt of a renal physiology module after the first draft:

Two points from this example. "Present" alone would have scored five out of six and hidden the real problem with clearance, where a two-sentence description is useless for a calculation question. And the missing topic was in a recording, the file type most often under-used when many sources are merged.

Step 5: repair the failed rows only

When a row fails, select just the sources for that topic and regenerate that one section. Do not rebuild the whole guide unless the topic map itself was wrong. Selective repair is faster, and it avoids a common trap: a full regeneration fixes one gap and quietly introduces two new ones elsewhere.

Copy this coverage matrix

Paste the table into a spreadsheet or your notes app, one row per syllabus topic. Fill it after each draft, not once at the end: the "Action" column is your repair queue for the next session.

Each failed check has a different fix. Regenerating the whole guide is almost never one of them:

Step 6: turn each topic into active practice

Coverage is a content check. Learning needs retrieval. In the review by Dunlosky and colleagues (2013), practice testing and distributed practice were the only two techniques rated high utility, while rereading, highlighting and summarisation were rated low. Roediger and Karpicke (2006) found that testing beat restudying on delayed tests, even though students who restudied felt more confident.

So for each important topic write three prompts: a short recall question, an application question, and "explain the exception". Answer before looking, check against the anchor, and write one corrective sentence. Then space those attempts over the days before the exam; our spaced repetition schedule guide shows how to fit them into a calendar.

What a finished study guide should contain

For each topic, aim for one screen, not five pages:

1. One-line answer: what you would say first if asked about it.
2. Core explanation: the mechanism, rule or derivation, with source anchors.
3. Must-know list: 3-5 facts, numbers or steps that are examinable on their own.
4. Common trap: the confusion or exception the exam likes to test.
5. Practice: the three retrieval prompts from step 6.

A 45-minute session for one module

1. Minutes 0-8: update the source inventory and syllabus topics.
2. Minutes 8-15: build or check the topic map (prompt 1).
3. Minutes 15-25: generate the draft sections with anchors (prompt 2).
4. Minutes 25-32: run the coverage matrix (prompt 3 plus your own check).
5. Minutes 32-40: repair the two highest-risk rows.
6. Minutes 40-45: answer three retrieval questions without notes.

The deliverable is not "a summary". It is a list of covered topics, the evidence for each, and a short queue of weak spots to practise next.

What source anchors look like in practice

Anchors are easier to demand once you have seen them. In the screenshot below every summary point ends with the slide range, page or audio timestamp it came from, and the left column shows a per-topic coverage count. Those are the two things steps 3 and 4 ask you to check by hand in any tool.

You can open the same sample notebook and click through the summary, flashcards, quiz and oral tabs. It is a fixed example, so nothing you type is sent to an AI model.

Doing this in Memoniq

Memoniq is built around this loop. Mixed files (PDF, slides, Word, spreadsheets, audio, video, YouTube lectures, images) sit in one notebook; the summary, flashcards, quizzes, mind map, lessons and oral practice all reuse that source set; and a coverage check lists missing topics per source so you can generate only the ones you select. Sources per notebook are capped by plan (5 on Free, 15 on Student) because merged output gets less coherent as the pile grows.

If a weak topic has to be explained aloud, continue with how to prepare for an oral exam. If you are still comparing source-grounded tools, see the NotebookLM alternative comparison. Flashcards from the same files? PDF to flashcards covers that step.

When not to trust the workflow

Do not treat a generated guide as the final authority for clinical, legal, safety-critical or fast-changing facts; check them against current primary sources or official guidance. Poor scans, ambiguous diagrams, spreadsheets without units and noisy recordings need extra review. And a coverage check only evaluates the files you supplied: it cannot find the lecture you forgot to upload.

Sources and further reading

• Dunlosky et al. (2013), Improving Students' Learning With Effective Learning Techniques
• Roediger & Karpicke (2006), Test-enhanced learning
• Karpicke & Blunt (2011), Retrieval practice produces more learning than elaborative studying with concept mapping
• Cepeda et al. (2006), Distributed practice in verbal recall tasks