know.

CookbookStudy and examsNo. 14

A college searchtwelve schools, compared from official sources

Recipe No. 14Study and exams

For
A high-school senior and a parent narrowing twelve colleges down to the ones worth applying to
You bring
A list of twelve colleges, the major the senior is leaning towards, a rough idea of what the family can pay, and the AI assistant you already use
You get
A college shelf with a document per school, every deadline, requirement and cost linked to the college’s own page, your visit notes in your own words, and a comparison document with the finalists side by side
Time
An evening for your assistant to build the shelf, a weekend to check it, then half an hour after each visit or new deadline
Works in
know.sh on the web · your AI assistant, through MCP
Keep it
Private while you work; share one document by a read-only link when it is ready.

By October of senior year the list has twelve colleges on it and the facts are in forty browser tabs. One school’s early action deadline is 1 November and another’s is 15 November; one wants the CSS Profile as well as the FAFSA; one says it is test-optional and its supplement asks for scores anyway. The spreadsheet a parent started in August has three versions and no sources.

This recipe gives each college a research document with the same findings in the same order: deadlines, requirements, costs, aid, the programme, outcomes, your visit notes and the questions still open. Whichever assistant you use, Claude, ChatGPT or a local model, researches all twelve from official sources and builds the shelf, every number linked to the page it came from. You check it in the editor, run each net price calculator yourselves, add what you saw on the visits, star the finalists, and keep a comparison document you can both read.

The applications, the essays and the financial forms stay where they are submitted. know.sh holds what you learned about each college and where you learned it.

What you will use

Shelf
One shelf, College search — class of 2027, with what matters to you as its line: “Environmental science, under 8,000 students, within a day’s drive.”
Research document
One document per college, titled with its name, plus The finalists, side by side for the comparison tables.
Finding
The same eight findings for every college; confirmed facts filed as Evidence, anything unconfirmed as Question, deadlines marked Key.
Your AI assistant
Researches each college from its own pages, its Common Data Set and the College Scorecard, fills in the eight findings with source links, and builds the comparison tables.
The editor
Where you check each number against its source, write the visit notes, and correct what the assistant got wrong.
Margin notes
Where the two of you see a college differently: the note sits beside the line it is about.
Star and Focus
Star the finalists; put the college with the next deadline in Focus for its last week.
Public link
A password link to the comparison document for the other parent or the school counsellor, with your cost estimates left out.

Method

  1. 1

    Decide whose library it is and what matters

    A know.sh library has one owner. In this recipe it is the parent’s, and the two of you work on it together at one screen; the senior reads the comparison from a link. Before anything is filed, agree the three or four things that decide the list — the major, the size, the distance, what you can pay — and put them in the shelf’s line.

    Then connect your assistant, following the support page for Claude, ChatGPT or a local model, and give it the twelve colleges. Ask for a document per college with these findings in this order: Deadlines, Requirements, Cost of attendance, Aid and net price, The programme, Outcomes, Visit notes and Questions for admissions.

  2. 2

    Have it research each college from official sources

    Ask your assistant to research each college on the web and fill in its findings, one college at a time. Name the sources you trust: the college’s own admissions and financial aid pages, its most recent Common Data Set (section C for admission, G for costs, H for aid), and the US Department of Education’s College Scorecard for graduation rates, typical debt and earnings. Rankings, forums and admissions blogs are not sources for a number.

    Ask for the year beside every figure (“2026–27 cost of attendance”) and a link beside every line, and for “not found” rather than a guess. This needs an assistant with web search; Claude and ChatGPT have it, and many local set-ups do not.

  3. 3

    Check every number in the editor

    Now sit down with the college’s pages open and press Edit on each document. Check each deadline, fee, requirement and cost against its source. Numbers change every year and colleges update pages over the summer; if the Common Data Set and the admissions page disagree, the admissions page is usually the newer.

    Mark each deadline you mean to use as Key. Change a finding to Question where you could not confirm it, and put what you need to ask in Questions for admissions: whether a campus job counts against aid, whether the major is capped. Revisions show which lines are still the assistant’s.

  4. 4

    Run each net price calculator yourselves

    Every college has a net price calculator on its own website, and it gives a far better estimate for your family than the published cost. Run it for each college, on the college’s site, and record only the result and the date in Aid and net price: “Estimated net price, 2027–28: $24,300. Calculator run 4 October.” Note whether the college asks for the CSS Profile as well as the FAFSA, and its priority aid deadline.

    The calculator asks for income and savings; that goes into the college’s site, not into know.sh. So do Social Security numbers, the FSA ID and every portal password. Write where they are kept, never what they are.

  5. 5

    Write your own visit notes

    After each visit, virtual tour or information session, write Visit notes the same day, the senior first and in their own words: the class you sat in on, the dorm, whether you could see yourself there. The parent adds a paragraph underneath, marked as such. Highlight the line that matters most.

    Where you disagree, keep both. Select the line and leave a Note: “Mom liked the co-op programme; I thought the campus felt empty on a Friday.” These notes are the part of the shelf no assistant can write, and in March they are the part you will reread.

  6. 6

    Build the comparison and star the finalists

    Ask your assistant to read the twelve documents and add tables to The finalists, side by side: one finding for deadlines and requirements, one for costs and your net price estimates, one for the programme and outcomes. Each row should name the college document and finding number it came from. Check the rows against your corrected findings; fix any cell that is wrong.

    When you have your short list, Star those colleges. Contents, under Starred, is now the list you are applying to. Move the documents you ruled out to a second shelf rather than deleting them; a college you dropped in October sometimes comes back in December.

  7. 7

    Keep application weeks in Focus, and the essays yours

    As each deadline comes near, put that college in Focus for its last week, and ask your assistant which starred colleges have something due in the next fourteen days. It can list each supplement’s prompt and word limit from the college’s page.

    It does not draft, outline or polish the essays. Application essays must be the student’s own, and many applications ask you to attest that they are. Keep the drafts in whatever you write them in, not on this shelf.

  8. 8

    Share the comparison on a password link

    Press Share on The finalists, side by side, set a password, choose Never expire, and send the link and the password separately to the senior, the other parent or the school counsellor. Leave out the costs finding so your estimates stay private. Readers see the tables without your highlights or notes, and cannot edit; what they think reaches you by message, and you file it.

    Keep the senior’s name, grades and test scores off the shared document. After decisions arrive in spring, turn the link off.

Specimen

  1. You

    Using know.sh, research Kestrel State University and fill in the findings of the document “Kestrel State University” on my shelf “College search — class of 2027”. Use only Kestrel’s own admissions and financial aid pages, its most recent Common Data Set and the College Scorecard. Put the year beside every figure and a source link on every finding. Write “not found” rather than guess.

  2. ChatGPT

    I searched the web and read Kestrel’s admissions, deadlines, tuition and aid pages, its 2025–26 Common Data Set and its College Scorecard entry. Filing now.

  3. research_list_findingsKestrel State University: 8 findings, all empty
  4. finding_updateWrote “Deadlines” (Evidence): EA 1 November, RD 15 January, merit priority 1 December, 1
  5. finding_updateWrote “Requirements” (Evidence): test-optional, one recommendation, one supplement, 2
  6. finding_updateWrote “Cost of attendance” (Evidence): 2026–27, in state and out of state, 3
  7. finding_updateWrote “Outcomes” (Evidence): graduation rate, median debt and earnings from the Scorecard, 6

  8. ChatGPT

    Six findings filled; Visit notes and Questions for admissions are left for you. Things to check: the Common Data Set online is for 2025–26, so its costs are a year older than the tuition page; I used the tuition page and noted both. The Scorecard’s debt and earnings describe students who received federal aid, not every graduate. I could not find whether the environmental science major has a cap, so I filed that as a question. Run Kestrel’s net price calculator yourselves; I did not.

ChatGPT filling in one college’s document through the know.sh connector, with the year and source for each figure and an honest list of what to check. The college is invented.

College search — class of 2027The finalists, side by side

1of 3

Deadlines and requirements, the five finalists

Insight, key finding, 1 highlight, 1 note, 5 sources, 212 words

Built from finding 1 and 2 of each college’s document. Checked against each admissions page on 21 September.

CollegePlanDeadlineTestsRecsSupplement
Wrenfield CollegeEarly action1 NovemberOptional2250 words
Kestrel State UniversityEarly action1 NovemberOptional1300 words
Ashcombe UniversityEarly decision II1 JanuaryRequired2Two, 150 each
Linden Hill CollegeRegular15 JanuaryOptional1None
Tarrow UniversityRegular5 JanuaryOptional2200 words

Ashcombe early decision II is binding. Only if it is first choice in December, and only after comparing its net price estimate with the others.

One comparison table after the family’s check, with a proposed note from their assistant about a deadline that changed. The colleges and figures are invented.

  1. You

    Using know.sh, which of the starred colleges on my shelf “College search — class of 2027” ask for the CSS Profile, and what are their priority aid deadlines?

  2. A local model through Ollama

    Two of your five starred colleges ask for the CSS Profile as well as the FAFSA:

    • Wrenfield College: CSS Profile and FAFSA, priority deadline 15 November.
    • Ashcombe University: CSS Profile and FAFSA, priority deadline 15 January for early decision II.

    Kestrel State, Linden Hill and Tarrow ask for the FAFSA only. Linden Hill’s Aid and net price finding is still marked as a Question, so check its aid page before relying on that.

    From your library

    1. Wrenfield College, 4: Aid and net price
    2. Ashcombe University, 4: Aid and net price
    3. Linden Hill College, 4: Aid and net price
A small question to a local model: it reads the starred documents and answers from the findings the family has already checked.

Prompts to try

  1. Your assistant, connected to know.sh (Claude, ChatGPT or a local model)

    Using know.sh, make a shelf called “College search — class of 2027” with the line “Environmental science, under 8,000 students, within a day’s drive.” Add one document per college on this list, titled with its name, and give each these findings in this order: Deadlines; Requirements; Cost of attendance; Aid and net price; The programme; Outcomes; Visit notes; Questions for admissions. Leave them empty for now. Also add a document “The finalists, side by side”. Colleges: [paste your twelve].

  2. Your assistant, connected to know.sh (Claude, ChatGPT or a local model)

    Using know.sh, read the eight findings in each starred document on my shelf “College search — class of 2027”. Recheck every deadline and application fee against the college’s own admissions page today, and leave a proposed note, citing the page, on any finding that no longer matches. Do not change the findings yourself.

  3. Your assistant, connected to know.sh (Claude, ChatGPT or a local model)

    Using know.sh, read the documents on my shelf “College search — class of 2027” and list everything due in the next fourteen days, by college, with the finding it comes from. For each college with a supplement due, give the prompt and the word limit from the college’s page, and nothing else about the essay.

Variations

  • Plan the visits as a road trip with a group trip guide: one document for the route, each campus tour and overnight a finding.
  • Add a Scholarships document, one finding per outside scholarship with its deadline and a link to the sponsor’s own page. A scholarship that charges a fee to apply is a reason to stop.
  • Applying as a transfer student, the same shape works with section D of the Common Data Set in place of section C.
  • Once the senior is at college, their own course study guide starts on a shelf of its own, in their own account.

Where it falls short

  • Nothing here watches the colleges’ websites. A deadline that moves after you checked it stays wrong until you or your assistant check again.
  • The Common Data Set runs a year behind and not every college publishes one; Scorecard earnings and debt describe students who received federal aid and can be several years old.
  • One owner per library. The senior and the parent cannot both edit from their own accounts; one of you keeps it, and the other reads a link or sits alongside.
  • Web research needs an assistant with web search, and smaller local models call tools less reliably. Check what any assistant files; Revisions shows every change.

A note on essays, money and the numbers

Indexed under

College admissionsCommon Data SetCollege ScorecardNet price calculatorsApplication deadlinesFinancial aidApplication essays

See also