AI math solver: photograph or type a problem to get step-by-step solutions, AI video explanations and practice problems; also has physics/chemistry/accounting versions.
Two Cornell CS undergrads registered the exact-match domain "math gpt" after the ChatGPT boom (2023-08) and built an AI solver that turns an uploaded screenshot into solution steps. The cold start relied on co-founder Nour's own on-camera TikToks ("if Chegg and ChatGPT had a baby", reaching 1.5M+ people with 60K+ student shares) and hands-on recommendations in math homework-help Discord servers; it then scaled on "math gpt" brand-keyword search, multi-subject sub-tools and AI animated explanation videos, claiming 2M+ students.One-line growth story, compiled by us from public sources; speculation is marked separately below.
The main engine is holding search positions: the domain is literally math gpt, so students looking for "ChatGPT for math" land on it with one search, and the top keywords are mathgpt, math gpt, math solver and math ai.
The founder's TikTok was a take-off: a hit video with the co-founder on camera demoing the product reached 1.5M+ people and was shared by 60,000+ students; students coming back every day to solve problems during homework season is the lubricant.
By the estimate, over 40% of visits are students coming back to solve problems while doing homework; about 0.9 per 100 visits click into the checkout page.
Model: Lenny's racecar growth framework. Engine = holding search positions (the exact-match domain "math gpt", plus a calculator page, multi-subject sub-tools and a blog covering search); turbo (take-off) = the founder's hit TikTok on camera; lubricant = students coming back again and again to solve problems during homework and exam seasons, with classmates and Discord groups recommending it to each other.
Real data: 2M visits, month-over-month +36.87%, the largest channel direct at 51.71%, the top-three channel ranking (direct, organic search, referrals), organic search at 94.37% of search and paid search 5.63%, top search terms (mathgpt, math gpt, math solver, math ai, mathsgpt), top referrals (waimao518.com 6.79%, msn.com 3.35%, google.com 3.34%), social sources (Discord, YouTube, Instagram), countries (US 35.53%, Australia 9.86%, UK 5.82%, Canada 3.77%, India 3.14%), bounce rate 43.39%, 3.27 pages per visit and average time on site 3 minutes 51 seconds come from the Similarweb public page for last month; the public page does not give the share of the second and third channels. 17,192 visits (about 17,200) to the Stripe checkout page come from quietmoney's checkout page data, with 9,217 the month before last and 12,741 the month before that. The "direct 64.5%, search 35.5%" in the profile is on Semrush's desktop basis and is not used as a share of the whole site.
Supporting evidence: Two Cornell computer science undergraduates registered math-gpt.org after the 2023 ChatGPT wave and built a tool that turns an uploaded screenshot into solution steps; the first users came from math Discord help servers, and the take-off came from co-founder Nour's hit TikTok (reaching 1.5M+ people, shared by 60,000+ students); when Cornell reported on it in 2024 there were 385,000 registered accounts and 3.5M+ problems submitted, and that year a Show HN said it had served 2M+ users; in 2025 it added versions for physics, chemistry, accounting and other subjects.
Inference: The two channel circles are real; how each path splits is an inference: direct URL typing 1,034,200 = students returning 700,000 + seeing demo videos 180,000 + classmate and Discord recommendations 154,200; other channels 965,800 = searching math gpt and similar terms 550,000 + searching the name after a classmate's recommendation 165,800 + students returning 150,000 + videos 100,000. Semrush desktop data shows that apart from direct it is basically all search, so most of the second circle is assigned to search. "math gpt" is both a brand name and a generic phrase, and how many people searching it came for the generic need is an inference. The ratios of 10% clicking in after searching, 5% clicking in after seeing a video and 40% opening after being recommended are guesses.
Checkout: checkout.stripe.com
“'was created by two Cornell Engineering students'; app 'available on the iOS App Store and Google Play Store'”Quote from the site
New-user source: SEO (Non-brand search). Direct 64.5%, search 35.5%, brand keywords (mathgpt, math gpt) 85.8%; non-brand search about 14%, from math solver and math ai. Referrals include waimao518.com 6.79%, msn.com 3.35%, google.com 3.34%; also iOS/Android apps, returning users come back directly.
Semrush source breakdown (desktop)
| Attributable new users, last month | Share |
|---|---|
| SEO | 14% |
| Other | 86% |
Direct + brand search 86%. Direct and brand search count as viral only when the site shows sharing mechanics; the largest share wins, under 5% means Other.
| Country | Share of visits |
|---|---|
| United States | 35.53% |
| Australia | 9.86% |
| United Kingdom | 5.82% |
| Canada | 3.77% |
| India | 3.14% |
Countries (Semrush): United States 34.7%, Australia 9.2%, Bangladesh 6.3%, Canada 6.1%, France 4.5%
English
Discord, YouTube, Instagram; Messaging: Discord
| Source | Share |
|---|---|
| ChatGPT | 0.17 |
| math-gpt.ai | 0.1 |
| Gemini | 0.08 |
AI referrals (Semrush): ChatGPT 4,188 (0.17%), math-gpt.ai 2,452 (0.10%), Gemini 1,943 (0.08%), mathgpt.ai 1,593 (0.07%), Claude 755 (0.03%)
2.6K keywords · brand ~86% · organic 94.37% / paid 5.63%
| Similarweb top 5 keywords |
|---|
| mathgpt |
| math gpt |
| math solver |
| math ai |
| mathsgpt |
Non-brand queries: math solver, math ai
10 referring sites
| Site | Share of referrals |
|---|---|
| waimao518.com | 6.79% |
| msn.com | 3.35% |
| google.com | 3.34% |
| Category | Share of referrals |
|---|---|
| Education | 18.66% |
| Graphics Multimedia and Web Design | 7.19% |
| Search Engines | 3.99% |
| Programming and Developer Software | 3.35% |
| Computers Electronics and Technology - Other | 3.3% |
About 17,192 last month (Similarweb estimate), #193 of 606; the month before was 9,217. Opening checkout is not paying, so this is not revenue.
New-user source: SEO (Non-brand search). New users mostly come from need-based searches Direct 64.5%, search 35.5%, brand keywords (mathgpt, math gpt) 85.8%; non-brand search about 14%, from math solver and math ai. Referrals include waimao518.com 6.79%, msn.com 3.35%, google.com 3.34%; also iOS/Android apps, returning users come back directly.
Not found
Dossier confidence: medium. Parts 1 and 2 come from reading the site and Similarweb data; part 3 from public reporting, founder posts, Wayback snapshots, WHOIS and similar, with speculation marked.
The number: checkout visits is Similarweb's estimate of visits from the site to a checkout payment page; opening checkout is not paying, so this is not revenue. Small sites land on fixed buckets.
Data through last month. Page updated 2026-10-31.