Send a question paper as a PDF to a WhatsApp group and your students get the questions. They do not get an exam. Four things go missing, and every one of them matters on the day: a clock the student cannot pause, a score the student did not calculate for themselves, any signal back to you about who struggled where, and the exam-day interface itself. An online mock test is simply the same paper with those four restored. Everything else about the PDF - the speed, the reach, the fact that it needs no account - is a real advantage, and worth being honest about before the pitch.
This is written for the person who already runs a batch this way and is wondering whether the move is worth it. If you are weighing a bot instead, a Telegram quiz bot against a full exam simulator is the closer comparison. If you have already decided and just want the mechanics, how a question paper becomes a computer-based test walks the conversion.
What the WhatsApp PDF Genuinely Gets Right
Start by granting the PDF its wins, because they are not small. There is no setup: you export from Word and you are done. There is no learning curve on either side - nobody has to be taught what a PDF is. It reaches a student on a patchy 2G connection with a cracked screen, because it downloads once and then needs nothing. It works on the bus with the data off. A parent can print it. A student can keep it forever and solve it again in March. No account, no login, no password reset, no support question late at night. Any replacement that does not acknowledge all of that is selling, not advising. The question is not whether the PDF is bad. It is what the PDF cannot do, and whether those things are the ones that decide a result.
Loss One: A Clock the Student Cannot Pause
On a PDF the clock is an honour system, and the honour system loses. The student answers a call, finishes the paper across the evening, stops to check a formula, and still reports it as done in time. Nobody is cheating; the clock simply does not exist. On MockSetu the countdown is real machinery. It runs in a background worker so a tab left in the background keeps accurate time instead of drifting. The deadline is written into the attempt row in the database the moment the clock starts, so a refresh resumes the same deadline rather than minting a fresh full-length one - closing the refresh-for-a-fresh-clock loophole any student could previously hit by accident. A warning fires once when five minutes are left. At zero it submits itself with whatever is on the screen - the whole paper, or just that section on a paper that runs sectional clocks.
Be honest about the edges too. Walk away from the device for longer than five minutes and the open sitting is sealed and filed with the answers already saved, counted as a normal ranked attempt - the student does not get that time back. And the clock length still reaches the server from the browser, so a determined candidate with developer tools can still interfere with it. What the database clock kills is the accidental exploit, which is the one that actually distorts your batch's scores.
Loss Two: A Score the Student Cannot Fudge
With a PDF the answer key lands in the same group afterwards and the student marks their own paper. Half-marks appear. A question that was "basically right" gets counted. Negative marking is applied with a generosity nobody admits to. The number that reaches you is not a score, it is a self-report. Online, marks are computed against your stored key at the moment the paper is submitted, using the scheme you set - marks for a right answer, marks taken off for a wrong one, marks for a blank, resolved from the question first, then its section, then the exam default. That chain is what lets one paper carry two rules at once. SSC MTS is the standard case: 90 questions for 270 marks across two sessions of 45 minutes each, where Session I is qualifying only and carries no negative marking while Session II counts towards merit and deducts one mark for a wrong answer. A PDF cannot enforce that distinction. It can only describe it and hope.
Loss Three: Any Signal Back About Who Struggled Where
This is the loss that quietly costs you teaching time, and the easiest one to not notice. A PDF returns nothing. Students message you a number, some do not message at all, and the next class gets planned around the loudest few replies. The analytics page on a published paper returns the paper's own behaviour instead: a Most Skipped list, a Most Reviewed list counting the questions candidates flagged to come back to, and a Common Misconceptions list that names, for each badly answered question, the wrong option the batch chose most often. There is a score distribution, average accuracy per section, and a Top Students board of the first three. Per-question figures - accuracy and average time - are averaged over everyone who attempted that question, so a question nobody reached does not drag the rest down.
Know the shape of what you get. These are counts and averages across the batch, not a dossier: there is no CSV or Excel export of results and no per-student report card, and the only names anywhere on that page are the usernames on the top-three board. Used properly that is enough. A wrong option the batch converged on is a misconception with an address, and it is a slice of Monday's lesson you did not know you needed.
Loss Four: The Exam-Day Interface Itself
Make this one concrete, because it is the one that gets dismissed as cosmetic and then costs marks. A student who has only ever attempted on paper meets the question palette for the first time in the examination hall. The palette is the numbered grid beside the question, and it is colour-coded: green for attempted, red for marked for review, purple for seen but not answered, plain for untouched. Beside it sits a Mark for Review button that flips to read "Marked" once pressed. None of that is intuitive the first time. A candidate meeting it cold spends real minutes learning that the grid is clickable, that marking a question for review does not submit anything, and that purple does not mean wrong. Those minutes come out of the paper.
The rehearsal starts before question one. The instructions run over two screens: how an exam works here, then this paper at a glance - a table of serial number, section name and number of questions, with maximum marks and sectional timing alongside them where the paper carries those. The second screen ends on a declaration the candidate has to tick before Start will do anything. It arrives un-ticked every single time, on purpose, because a pre-ticked box is a box nobody reads. A student who has walked those screens several times is not reading them for the first time in a hall where the clock has already begun.
A PDF asks whether your student knows the answer. An exam asks whether they can find it in time, on a screen, with nobody sending the key round afterwards.
Moving the Paper You Already Have
You do not retype anything. The universal route for a paper that already exists as a PDF is to have an AI read it into structured JSON and then upload that file. Server-side extraction straight from a PDF does exist, but it is off by default and switched on for a creator only on request, so plan around the prompt-plus-upload route - it works for everybody today.
- Run the published extraction prompt from the JSON upload guide in whatever AI you already use, and upload the JSON it hands back.
- Check the figures. The picture itself is not in the JSON, but the extraction records each figure's page and bounding box, the upload crops it out of your source PDF, and you get thumbnails plus a Re-snip button to drag a new box around anything it framed badly.
- Add the numeric, TITA and match-the-column questions by hand. The importer deliberately leaves those for you rather than guessing them wrong.
- Set the marking scheme from the exam default downwards, overriding only the sections and questions that genuinely differ.
- Write the scheme into the general instructions in plain words, section by section, in every language you publish.
- Preview the paper on your own account and sit a few questions. A creator preview records nothing - no attempt, no score, no entry in your analytics - so it cannot pollute the batch's numbers.
- Publish, then copy the link. Sharing refuses while the exam is still a draft, which is the one guard rail against a half-built paper reaching a group chat.
What an Online Mock Still Cannot Do
Being the honest option means listing this before you are asked. There is no proctoring of any kind - no webcam, no lockdown browser, no tab-switch detection - so an online mock is not a secure examination, it is a rehearsal. Published papers are public: anyone with the link may attempt them, and there is no payment, paywall or private delivery to one batch only. There is no question shuffling and no cap on attempts. There is no Excel or Word import, no certificates, no white-labelled domain and no mobile app - it is a browser page. Nothing notifies your students by email, SMS or WhatsApp that a test has gone up; the platform sends transactional account mail for signup and password reset and nothing else, so announcing the test is still your job. And the AI only extracts from a PDF you supply. It will not write a paper from a syllabus.
Keep WhatsApp. Change What You Send.
The honest conclusion is not that WhatsApp is the problem. The group is still where your students already are, and nothing here replaces it. What changes is the payload: a link instead of a file, and a results conversation instead of a key. Sharing an online test with students covers how to post it so people actually sit it rather than scroll past. Everything else - what a creator account can build, including sections with their own clocks, bilingual papers and a listing in the public library - is free and needs no card. Run your next paper both ways if you want the proof. Send the PDF to one group and the link to another, then compare what each route can tell you afterwards about which question the paper actually broke on.