Shishya in numbers
Every number on this page is counted from Shishya's database when the page is built, with its definition; computed on 27 Sep 2026 (IST), refreshed hourly. Only complete weeks (Monday to Sunday, IST) are shown, and no count of people, or of what they did, is printed below twenty (counts of exams, papers and chapters can be smaller). How content is made and checked is in the editorial policy; facts and contacts for writing about Shishya are in the press kit.
Do students come back?
The numbers that say whether Shishya is useful: how many new accounts are used again, and how much practice an active account does. They count signed-in activity only, so they are floors.
Came back within 7 days of signing up
28.7%
213 of 742 accounts
Of the accounts created in the sign-up weeks shown, the share with signed-in activity (a signed-in analytics event not tagged as a bot, a mock started, a mock submitted or auto-submitted, or a question to the AI tutor; the nightly cleanup job's closing of abandoned mocks is not activity) on any of the 7 IST days after their sign-up day. Team accounts are left out. It is a floor: someone who comes back signed out is not seen.
sign-up weeks 17 Aug–13 Sep 2026 · computed 27 Sep 2026
Came back on days 8 to 30
12.8%
79 of 619 accounts
Of the accounts created in the sign-up weeks shown, the share with signed-in activity (a signed-in analytics event not tagged as a bot, a mock started, a mock submitted or auto-submitted, or a question to the AI tutor; the nightly cleanup job's closing of abandoned mocks is not activity) on IST days 8 to 30 after their sign-up day. Shown pooled only, because most single weeks have fewer than 20 such returners. Team accounts are left out.
sign-up weeks 27 Jul–23 Aug 2026 · computed 27 Sep 2026
Active in the last 30 days, on a day after signing up
16.4%
312 accounts; counting the sign-up day too, when a new account is active just for signing up: 43.0%
Accounts with signed-in activity (a signed-in analytics event not tagged as a bot, a mock started, a mock submitted or auto-submitted, or a question to the AI tutor; the nightly cleanup job's closing of abandoned mocks is not activity) on any of the last 30 IST days (today included) that is not the account's own sign-up day, as a share of all accounts. Signing up alone does not count. Team accounts are left out. The second figure: Accounts with signed-in activity (a signed-in analytics event not tagged as a bot, a mock started, a mock submitted or auto-submitted, or a question to the AI tutor; the nightly cleanup job's closing of abandoned mocks is not activity) on any of the last 30 IST days (today included), as a share of all accounts. Creating an account is itself a signed-in event, so an account created in those 30 days counts here just for signing up. Team accounts are left out.
29 Aug–27 Sep 2026 · computed 27 Sep 2026
Mocks per weekly active account
2.24
485 mocks ÷ 217 active accounts; 3.65 per account that took a mock
Mocks taken in the week (submitted or auto-submitted, by finish time) divided by the accounts with signed-in activity (a signed-in analytics event not tagged as a bot, a mock started, a mock submitted or auto-submitted, or a question to the AI tutor; the nightly cleanup job's closing of abandoned mocks is not activity) that week; the second figure divides by the accounts that took at least one mock. Team accounts are left out.
week of 14–20 Sep 2026 · computed 27 Sep 2026
Came back within 7 days, by sign-up week
| Sign-up week | Sign-ups | Came back within 7 days | Share |
|---|---|---|---|
| 20–26 Jul | 97 | 22 | 22.7% |
| 27 Jul–2 Aug | 118 | 31 | 26.3% |
| 3–9 Aug | 124 | 41 | 33.1% |
| 10–16 Aug | 165 | 44 | 26.7% |
| 17–23 Aug | 212 | 60 | 28.3% |
| 24–30 Aug | 157 | 48 | 30.6% |
| 31 Aug–6 Sep | 211 | 64 | 30.3% |
| 7–13 Sep | 162 | 41 | 25.3% |
- The 7-day rule above for each ISO sign-up week (IST, Monday to Sunday). A week appears once its last sign-up's 7 days have passed. A share is printed only when the returners and the sign-ups are both at least 20.
Where new students arrive from
Where new accounts arrive from (14–20 Sep 2026)
| Source | Sign-ups | Share |
|---|---|---|
| AI assistants | 106 | 74.1% |
| Search engines + Direct or unknown | 37 | 25.9% |
- From AI assistants: 106 of 143 (74.1%)
For each account created in the weeks shown, the source of its browser's first page view on Shishya: the sign-up's browser id is linked to that browser's first page view; failing that, the sign-up event's own tag or referrer; failing that, the referrer stored on the account. A tracking tag (utm_source) wins over a referrer. 'Named source' means that first visit carried a tag or a referrer naming where it came from. Groups under 20 are merged, and the 'Named source' and AI-assistant lines are left out when printing them would give back a merged group by subtraction. Team accounts are left out.
Where new people arrive from (24 Aug–20 Sep 2026)
| Source | New people | Share |
|---|---|---|
| AI assistants | 1,818 | 47.4% |
| Search engines | 1,625 | 42.4% |
| Direct or unknown | 340 | 8.9% |
| Social and messaging + Other websites | 26 | 0.7% |
| 25 | 0.7% |
- Named source (a tag or referrer said where the first visit came from): 3,494 of 3,834 (91.1%)
- From AI assistants: 1,818 of 3,834 (47.4%)
People (an account, or a first-party browser id) whose first page view falls in the weeks shown and who meet the 'people who came' rule of the all-time counter, grouped by the tag or referrer of that first view. Visitors with one page, no referrer and no tag have no identity and are not included, so this is not everyone who came. Groups under 20 are merged. Team accounts are left out.
What cannot be seen: a visitor who reads one page and arrives with no referrer and no tag gets no identity, and some apps strip both the referrer and the tag, so the AI-assistant share is a floor. ChatGPT's links carry utm_source=chatgpt.com and no referrer; they are counted by that tag.
Week by week
Each column is an all-time counter's own definition, counted per ISO week (IST, Monday 00:00 to Sunday 23:59). Team accounts are left out. The week in progress is not shown.
| Week | Sign-ups | People who came | Mocks taken | AI tutor questions | Active accounts | Mocks per active account |
|---|---|---|---|---|---|---|
| 27 Jul–2 Aug | 118 | — | 396 | 288 | 140 | 2.83 |
| 3–9 Aug | 124 | — | 247 | 254 | 165 | 1.50 |
| 10–16 Aug | 165 | — | 526 | 512 | 215 | 2.45 |
| 17–23 Aug | 212 | 2,053 | 589 | 538 | 267 | 2.21 |
| 24–30 Aug | 157 | 1,272 | 426 | 384 | 212 | 2.01 |
| 31 Aug–6 Sep | 211 | 1,695 | 660 | 497 | 287 | 2.30 |
| 7–13 Sep | 162 | 1,676 | 644 | 416 | 256 | 2.52 |
| 14–20 Sep | 143 | 1,551 | 485 | 267 | 217 | 2.24 |
- Sign-ups: Accounts created in the week.
- People who came: People with a page view that week (not tagged as a bot) who meet the all-time counter's rule — an identity seen on 2 or more page views, or on one view with a referrer — plus page views by verified browsers that carry no identity. Proves a visit, not learning. The home page strip calls this counter 'learners'.
- Mocks taken: Mocks submitted or auto-submitted, by finish time.
- AI tutor questions: Questions students sent to the AI tutor: members' messages (guest-import copies left out) plus guest questions that carry a browser cookie.
- Active accounts: Accounts with signed-in activity (a signed-in analytics event not tagged as a bot, a mock started, a mock submitted or auto-submitted, or a question to the AI tutor; the nightly cleanup job's closing of abandoned mocks is not activity) that week.
- Mocks per active account: Mocks taken ÷ active accounts.
- “People who came” shows — for weeks before 17 Aug 2026: identity rules changed on 31 Jul and 16 Aug 2026, so earlier weeks are not comparable.
All-time counters
The site's running totals, each with the definition the counter itself uses, word for word. The live numbers (people active now, today's totals) are on the home page strip; an hourly page cannot show them truthfully.
| Counter and definition | Value |
|---|---|
| AccountsUser rows (accounts). | 1,910 |
| People who came (the home page strip's 'learners')Distinct people who came to Shishya: identities on 2+ page views, or one view that arrived from another site or a tagged link (for example utm_source=chatgpt.com — 27 Sep 2026: these were being missed), plus identity-less browser landings, overlap-corrected. Proves a visit, not learning. | 15,403 |
| Mocks takenAttempt rows with status SUBMITTED or AUTO_SUBMITTED — mocks a student finished, any exam or school chapter. | 5,287 |
| Questions to the AI tutorChatMessage rows with role USER (guest-import copies excluded) plus AnonTutorLog rows with a shishya_anon cookie — questions people sent to the AI tutor, site-wide; cookie-less guest calls (not provably a person) left out. | 4,195 |
| Questions answered in mocksAnswer entries with a chosen option inside submitted attempts (unanswered questions not counted); refreshed every 10 minutes. | 73,751 |
| Live tests takenSubmitted attempts on a live-test mock that started inside the test's window. | 92 |
| Exam goals setActive enrolments in real exams (a member's chosen target exam). | 1,813 |
| Page viewsPAGE_VIEW events all-time, ingest-tagged bots and pre-30-Jul-2026 phantom ids excluded. | 87,590 |
| Exams coveredActive real exams (REAL_EXAM_WHERE) — the finder's 'Browse all' count; refreshed every 10 minutes. | 192 |
| Practice questionsValidated, not-withdrawn questions on live real exams and school chapters; refreshed every 10 minutes. | 39,505 |
| Topic notesTopicTeachingNote rows with non-empty content on live real exams and school chapters; refreshed every 10 minutes. | 4,575 |
| School chapters with notes or practiceSchool chapters with notes or a practice quiz (the school surface's indexable chapters); refreshed every 10 minutes. | 230 |
| LanguagesLocales the UI serves (locales in i18n.ts), English included — a derived constant. | 19 |
What Shishya covers
Counts of what Shishya holds, read from its database and data files when the page is built. Computed 27 Sep 2026.
Exams and practice
- Exams covered192
Active real exams (REAL_EXAM_WHERE) — the finder's 'Browse all' count; refreshed every 10 minutes.
- Government recruitment exams119
Live exams of this kind (src/lib/exam-kind.ts).
- Entrance exams53
Live exams of this kind (src/lib/exam-kind.ts).
- Olympiads18
Live exams of this kind (src/lib/exam-kind.ts).
- Professional exams2
Live exams of this kind (src/lib/exam-kind.ts).
- Practice questions39,505
Validated, not-withdrawn questions on live real exams and school chapters; refreshed every 10 minutes.
- Practice questions that passed the answer check39,348
Validated, not withdrawn questions on live exams and school chapters that carry the answer check's pass record (validated by the check, with its verdict stored).
- Topic notes4,575
TopicTeachingNote rows with non-empty content on live real exams and school chapters; refreshed every 10 minutes.
School
- NCERT chapters listed, each linking the official book1,146
Top-level NCERT chapters on the school pages; each links NCERT's own PDF on ncert.nic.in.
- School chapters with Shishya's own notes230
School chapters whose page carries Shishya's own chapter notes.
- School chapters with checked practice124
School chapters with enough validated, answer-checked questions for a practice quiz (the school quiz's own minimum).
- School questions written, not yet through the answer check4,228
Questions on school chapters that are not validated, not withdrawn and have no answer-check record yet. They are not shown to students until they pass.
Official sources
- Official previous-year papers linked193
on 31 exams. Links to question papers and answer keys on the conducting bodies' own sites, on live exams.
- Official cutoff rows7,875
on 17 exams. Cutoff figures copied from the conducting bodies' own published documents, each with its source link, on live exams.
- Exam dates in the next 90 days marked Official71
on 31 exams. Tracker dates announced and cited on the conducting body's own site (src/lib/official-source.ts). A date whose own label says expected, tentative or likely is left out, as on Shishya Pulse.
- Exam dates in the next 90 days marked Reported56
on 24 exams. Tracker dates that were announced but are cited through a secondary source such as a news or coaching site.
- Exam dates in the next 90 days marked Expected192
on 104 exams. Estimates from previous cycles, never presented as announcements.
Colleges, scholarships, careers and languages
- Colleges from the NIRF 2025 rankings77
Entries in src/lib/colleges-data.ts.
- Scholarships186
Scholarship schemes, each linking its awarding body (the one outside aggregator is not counted, nor the rows held out of the catalogue as not a scholarship, not found on an official page, or in doubt).
- Career guides94
Entries in src/data/careers.ts.
- Languages19
English and 18 Indian languages. Locales the UI serves (locales in i18n.ts), English included — a derived constant.
Answer check
Each question is solved 3 times by an AI model that is not shown the stored answer key; an AI examiner then compares the key, the solves and the question. A question passes when the examiner judges the stored key correct with at least 80% confidence and at least 2 of the 3 solves reach that same answer. When the examiner and at least 2 solves agree on a different option, the key is corrected and the solution is replaced by the solve that reached it. Anything else — two defensible answers, low agreement, a broken question — is withdrawn from every practice pool. The check is automated, not a human audit, and it can be wrong; every question keeps its Report button.
How content is produced is in the editorial policy. The verification page covers something else: the fact badges on college and school-board pages.
| Outcome | Questions |
|---|---|
| Questions checked | 42,806 |
| Accepted as written | 38,115 |
| Answer key corrected | 1,233 |
| Withdrawn (not served in any practice pool) | 3,458 |
| Live questions never checked (on 55 exams; validated at insert or in bulk before the check existed) | 157 |
| Last check day | 27 Sep 2026 |
- From the answer-check record stored on each question. Checked = every question the check has a verdict for. Accepted as written = the examiner agreed with the stored key and the question is live. Key corrected = the check replaced the key and the question is live. Withdrawn = the check failed the question, so no practice pool serves it; questions that were not yet live are counted here too, so only the 'live before the check' rows show what the check took out of use. 'Live before the check' = the question was already being served when it was checked.
- Questions being served (validated, not withdrawn, on a live exam or a school chapter) that carry no answer-check pass: they were validated at insert or in bulk before the check existed.
| Outcome | Questions | Share |
|---|---|---|
| Questions live before the check | 36,690 | |
| Kept as written | 32,322 | 88.1% |
| Answer key corrected | 1,172 | 3.2% |
| Withdrawn | 3,196 | 8.7% |
Student question reports, all time: 72 received, 72 closed. Question reports students sent with the Report button, all time, and how many are marked closed. Printed only when both are at least 20.
Questions
- How many people use Shishya?
- 1,910 accounts had been created on Shishya by 27 Sep 2026. Of the accounts created in the sign-up weeks 17 Aug–13 Sep 2026, 28.7% (213 of 742) used Shishya again, signed in, within 7 days of signing up. 312 accounts (16.4% of all accounts) used Shishya, signed in, in the last 30 days on a day after the day they signed up; counting the sign-up day too (signing up is itself a signed-in event, so an account created in those 30 days counts just for signing up), the share is 43.0%. An account is not a unique student, and people who study signed out are not seen, so these are floors.
- How are Shishya's answer keys checked?
- Each question is solved 3 times by an AI model that is not shown the stored answer key; an AI examiner then compares the key, the solves and the question. A question passes when the examiner judges the stored key correct with at least 80% confidence and at least 2 of the 3 solves reach that same answer. When the examiner and at least 2 solves agree on a different option, the key is corrected and the solution is replaced by the solve that reached it. Anything else — two defensible answers, low agreement, a broken question — is withdrawn from every practice pool. The check is automated, not a human audit, and it can be wrong; every question keeps its Report button. By 27 Sep 2026, 42,806 questions had been checked: 38,115 accepted as written, 1,233 with a corrected key and 3,458 withdrawn. Of the 36,690 questions that were already live when checked, 3.2% had their key corrected and 8.7% were withdrawn.
- Where do Shishya's students come from?
- Of the 143 accounts created in 14–20 Sep 2026, 74.1% arrived from AI assistants such as ChatGPT, which tags its links utm_source=chatgpt.com and sends no referrer. Some apps strip both the referrer and the tag, so the AI-assistant share is a floor.
How to cite
Shishya, “Shishya in numbers”, https://shishya.in/shishya-in-numbers, as of 27 Sep 2026.
The same numbers as markdown: context.md · Shishya Pulse (weekly data note) · Press kit · About Shishya