Four failures preceded the writing product
Park started with a clothing business in high school, then left college to pursue a technology startup. That company failed, as did later dating and social applications. The record matters because Jenni AI was not a first attempt or an overnight response to a model release.
Jenni began from Park's literature background and his co-founder's engineering ability. An early GPT-2 version helped agency writers work slightly faster and operated as an internal tool. After GPT-3, the team moved from agency work toward a broader software product, but its ideal customer remained unclear.
Revenue stayed near $2K per month for about eighteen months
The Indie Hackers case says Jenni generated about $2,000 per month and remained flat for almost eighteen months. The founders moved to Malaysia to reduce expenses and were nearly out of funds. External investment extended the runway while they searched for a stronger market.
That plateau cannot be removed from the later growth story. The eventual ARR figure followed years of failed projects, a long period without growth, financing, and continued customer research. It was not produced by changing one prompt or adding a larger model.
The turning point was a narrower customer and fewer features
Park cold-called and interviewed users. The team observed that students used the product heavily, so it focused on students, academics, and researchers. It also removed features that did not strengthen the central writing and citation workflow.
Customers were paying to complete academic writing and research work. AI shortened parts of generation, but product value still depended on customer selection, acceptable output, citation experience, reliable delivery, and pricing. A clearer customer changed the roadmap more than another broad feature set.
The reported $6.4M ARR came with non-repeatable distribution
The editorial case reports that Jenni AI later reached about $6.4 million ARR and approximately $533,000 in monthly revenue. ARR is annual recurring revenue, not net profit, audited cash, or Park's personal income. The case also mentions a monthly EBITDA figure, but it is still a source-reported number rather than an independent audit.
Growth relied on early investment, a viral social post, short-form video, influencer partnerships, SEO, and paid advertising. Some video campaigns accumulated hundreds of millions of views. A new founder cannot compress those financing and distribution conditions into a claim that changing models will reproduce the outcome.
Interview three users before rebuilding the product
Talk to one paying user, one churned user, and one person who refused to pay. Ask how each completes the task today, which step hurts, what outcome deserves payment, and what failure would cause cancellation. Avoid questions designed only to collect compliments.
Then choose one anonymized real sample and run a minimum test with a separate project key. Record usage, correction time, accuracy, and a stopping budget. If the interviews point to a different customer, rewrite the product promise before deciding which features or models to keep.
Source and evidence boundary
The main source is the Indie Hackers editorial case 'How a college dropout built an AI edtech with $6.5 million ARR in two years.' It reports the failed startups, the $2,000 monthly plateau, investor funding, customer repositioning, distribution, and Park's cancer diagnosis during the growth period.
The $6.4 million figure remains reported ARR rather than audited profit. Public material does not disclose a complete cost, refund, tax, and labor ledger. The case provides no evidence that Park or Jenni AI used APIToken, and the story is not an income promise for APIToken users.