The research behind next4
Before we wrote a line of the product's code, we spent two months trying to talk ourselves out of building it. This is what that looked like: a qualitative pass over 1,226 coded comments, a fielded survey of 84 parents and students, a six-segment competitive teardown, and a market frame, with the methods, the caveats, and the claims that got thrown out.
Why any of this happened first
The idea was a five-feature bundle: essay coaching, an AI-writing detector, an admit-chance predictor, interest-based discovery, and a swipe interface. It sounded good, which is the problem. Five features that each sound good is the most common way a product gets built for nobody.
So the research had one job: find out which of the five, if any, a real person would choose when choosing costs them something. Three of the five did not survive it.
The qualitative pass: 1,226 comments
Surveys tell you how people answer questions. They don't tell you what people say when nobody is asking. So the first pass read applicants talking to each other.
Method
Sizing
Four keyword passes over twelve months of public Reddit via a read-only scraper, yielding 279 unique threads, 225 of them core-relevant.
Depth
The 39 highest-signal threads were pulled to comment level, 1,226 comments in all, and coded on three axes at once: theme, sentiment, and segment, by four coders working in parallel.
Adversarial check
Five headline claims were then fact-checked against the corpus by someone trying to break them. Three held. Two were softened. One of the two was a conclusion we had already half-committed to.
Hygiene
No usernames or personal information stored. Short verbatim quotes with permalinks, nothing more.
The stated limit, which matters: Reddit is a vocal, high-achieving, selective-school-fixated minority. Every finding here is a hypothesis for the survey to test. None of it is a market share, and it was never treated as one.
The finding that reframed the competition
We went in believing the threat was the funded startups. The corpus said otherwise, by saying nothing about them at all.
| Who students actually named | Mentions | Sentiment |
|---|---|---|
| ChatGPT / general LLMs | 157 | Mixed to negative on writing; accepted for brainstorming |
| Human consultants (generic) | ~137 | Conditional; hourly praised, packages resented |
| Crimson | 33 | Negative-skewing: overpriced, cookie-cutter |
| ESAI | 4 | Negative, plus founder self-promotion |
| CollegeVine | 3 | Neutral; its chancing called "too optimistic" |
| Kollegio | 0 | One rave, flagged by peers as a fake promo |
| AdmitYogi | 0 | None |
Only ChatGPT and the human consultants cleared a 25-mention bar. The two companies at the center of our competitive brief could not be scored from organic discussion at all. That absence was the finding. It moved the funded one-stops from "the threat" to "a funding and marketing threat," and moved a general-purpose chatbot into the position of the real default competitor.
What people actually wanted, in order
- College list, matching, discovery, fit — the strongest theme by a distance.
- Tracking and deadline planning — spreadsheets, whiteboards, and dread.
- Financial aid, merit and real cost.
- Chancing — sought and distrusted in the same breath. "No chance calculator is accurate."
- Essay feedback — concentrated among students without counselor access.
Nobody asked for an all-in-one app by name. The pain of fragmentation was everywhere; the concept that solves it was nowhere. That means the integrated product has to be sold, not assumed, which is a harder go-to-market than we had been planning for.
The survey: 84 people, and one question that did the work
The qualitative pass produced hypotheses. The survey tested them against people who weren't self-selected into a forum.
n = 69 parents, 15 students, fielded August 2026. Scored against a rubric written before the data came in, with pre-registered thresholds and a 50% discount applied to stated intent. It's a convenience sample measuring stated intent, not revealed preference. It is directional evidence that sharpens a plan, not a forecast.
The forced choice
Respondents rated every feature, and then were made to keep exactly one. The two results disagree, and the disagreement is the most useful thing the survey produced.
| Feature | Kept as the only one | Rated value (1–5) |
|---|---|---|
| Browse colleges, build a shortlist | 28 | 3.96 |
| Quiz that learns interests, suggests fits | 24 | 3.52 |
| AI feedback on essay drafts | 6 | 3.96 |
| AI essay brainstorming | 5 | 3.94 |
| Deadline / requirement dashboard | 3 | 4.18 |
| Admission-chance ranges + steps | 3 | 4.19 |
The 52 of 69 who kept discovery are the reason next4's front door is a card sort and not an essay tool. Essay tooling, the thing the entire competitive set is racing to build, took 11 votes out of 69 combined.
The positioning gate
We asked parents directly about AI and essays, because the answer determined whether the product was viable at all:
- 61% (42): fine for brainstorming and feedback, as long as my child does all the actual writing.
- 30% (21): AI should stay away from college essays entirely.
- 5 respondents would allow drafting.
The feedback-never-writing posture passed. But the 30% hard-no is a real constituency, and it changed a product decision: preserving student authorship isn't enough if it's invisible. It has to be visibly disclosed in the interface, which is a trust-design requirement rather than a copy problem.
Who actually buys
61% of parents described the same motion: "my child finds it and asks; I decide whether to pay." Among students, 80% said a parent's money is involved. So acquisition is student-facing and conversion is parent-facing: two different audiences, two different surfaces, one funnel.
The competitive teardown
Run as a fan-out research harness rather than a reading list, so the claims could be checked rather than collected.
| Segment | Players | What it meant for us |
|---|---|---|
| Free AI one-stops | Kollegio | The direct substitute, and a $0 anchor |
| Paid essay / coaching | AdmitYogi, ESAI, Sups | Feedback-not-writing is table stakes, not a wedge |
| Free discovery & social | Appily, Niche, ZeeMee | Discovery and chancing are worth $0 standalone |
| School-side B2B | Scoir, Naviance | Chancing already reaches students free, via schools |
| Human consulting | Crimson, Prompt, Empowerly | The price umbrella we're productizing beneath |
| General LLMs | ChatGPT, Claude, Gemini | The substitution floor, and the real competitor |
The frame that changed the pitch
The scariest-looking competitors are free because colleges are their customer, not students. Kollegio's business pays out when a student engages and is delivered into an institution's CRM; its matching metrics are self-reported enrollment lift for colleges, not fit accuracy for students. CollegeVine has completed the same pivot, and student access there is free because institutions pay.
That reframes a price gap as a structural misalignment, which is durable in a way that being cheaper never is. It's also the origin of next4's sharpest line: a product paid for by families has only one incentive, which is fit.
Two honest counterweights we wrote down at the same time. A direct competitor already occupies next4's exact positioning and price band. And the central claim — that next4 matches more accurately — was asserted, not demonstrated. A head-to-head test against the incumbents on real student profiles remains the highest-value proof we have not yet run.
The market frame
Sized top-down, then cut down to something defensible.
| Figure | What it is | |
|---|---|---|
| Volume | ~1.5M | Distinct first-year applicants filing ~10.2M applications, 2024–25 |
| Per applicant | 6.8 | Schools applied to, up 46% in a decade: the complexity driver |
| TAM | ~$3.4B | Core admissions consulting, flat at ~1% growth |
| Adjacent | ~$40B | Test prep and online tutoring, growing ~11% |
| SAM | ~$90–180M | The "missing middle": 400k–800k households priced out, not uninterested |
The number that reframed the opportunity is the SAM. Paid counseling concentrates among wealthy families because of price, not because demand stops. That makes this market expansion rather than share capture, which is a different and better story, and a harder one, because it means the buyers have never paid for anything in this category before.
What the research actually changed
- The chance predictor was cut. Most-wanted and least-trusted in the same corpus, and statistically unstable in the literature. We built the screen, looked at the number, and deleted it.
- The AI-writing detector was cut. Not because detectors misfire often. The best 2026 tools are under 1% false-positive, and the corpus contained the study that says so. Because the harm they do is silent, unappealable, and lands on non-native speakers.
- Discovery became the front door, against our own instinct to lead with essays.
- Authorship disclosure became visible in the UI, because 30% of parents needed to see it, not be told it.
- The competitive story was rewritten around structural alignment instead of feature breadth.
- Acquisition and conversion were split across two audiences once the buyer motion was clear.
What it doesn't answer
Stated intent is not a credit card. The survey can tell you 87% of parents like the idea; it cannot tell you what fraction will pay, or stay. Reddit can tell you what a vocal minority argues about; it cannot tell you what a median applicant in Ohio does in October.
Three things remain genuinely open: real paid conversion, retention across a full application season, and whether the matching is actually more accurate than the incumbents. That is the one claim the whole product rests on, and the one the research never proved. Only the shipped app answers any of them, which is why it shipped.
The through-line: every one of these passes was designed so it could come back negative, and two of them did. That's the only reason the ones that came back positive are worth anything.