The Night Before: AI Document Maker Triage for Interviews
The Night Before: AI Document Maker Triage for Interviews
The interview is at 10 a.m. tomorrow. It is now 7 p.m. There is a job description open in one tab, a company website in another, and a vague plan to "read up on them" that has not survived contact with dinner.
This is the most common interview prep scenario, and it is also the most wasteful. Most candidates spend the night before reading. Reading feels productive and produces nothing. The candidates who walk in with a clear point of view are the ones who spent that same evening writing — turning scattered research into a small set of documents that force clarity and double as leave-behinds.
An AI document maker compresses that writing work from six hours to about ninety minutes. This is the full triage system: what to build, in what order, with the exact prompts, and what to do differently depending on how much time is left on the clock.
Why Writing Beats Reading as Interview Prep
Reading about a company creates recognition. Writing about a company creates recall. In an interview, recognition is useless — nobody hands out multiple-choice questions. Recall under pressure is what gets tested.
There is a second reason. Writing exposes gaps. A candidate can read a company's product page three times and feel prepared. The moment that same candidate tries to write one paragraph explaining how the company makes money and who it competes with, the fuzziness becomes obvious. Better to discover that at 8 p.m. than at 10:15 a.m.
The third reason is tactical: several of these documents can be sent or handed over. A one-page plan emailed after a final-round interview is a differentiator that costs nothing but an evening.
The Five-Document Triage Stack
Not all interview documents are equally valuable, and nobody has time to build all of them. The stack below is ordered by return on time invested. Work down the list and stop when the clock runs out.
- The Company Brief — one page on what the company does, how it makes money, and what is changing.
- The Story Bank — six to eight structured examples from a career history, ready to deploy.
- The Question Map — likely questions matched to specific stories and evidence.
- The 30-60-90 Plan — a one-page view of the first three months in the role.
- The Leave-Behind — a polished PDF summarizing a point of view on the role's biggest challenge.
Documents one through three are for the candidate. Documents four and five are for the interviewer. The split matters: internal documents can be messy and dense; external ones must be clean, short, and formatted well. That is precisely where an AI document maker earns its keep — the same raw material gets reshaped into two very different outputs without retyping anything.
Document 1: The Company Brief (20 Minutes)
The goal is not a research report. It is a single page that answers five questions well enough to speak to each for sixty seconds.
- What does this company sell, and to whom?
- How does it make money, and what drives that number up or down?
- Who are the obvious alternatives a customer would consider?
- What has visibly changed in the last year — new products, new markets, new leadership, new pricing?
- What does this specific role exist to fix?
The workflow: gather raw material first, then let the model structure it. Paste in the job description, the company's "about" and product pages, any recent announcements, and the interviewer's public bio if available. Then prompt:
"Using only the source material below, write a one-page interview brief on [Company] for a candidate interviewing for [Role]. Structure it as five sections: What They Sell, How They Make Money, Who They Compete With, What Changed Recently, Why This Role Exists. Keep each section to 80 words or fewer. Where the source material does not support an answer, write 'Gap:' followed by the specific question a candidate should ask. Do not speculate."
Two details in that prompt do the heavy lifting. "Using only the source material below" keeps the output grounded in real inputs rather than general knowledge about the industry. The "Gap:" instruction converts missing information into interview questions — which is exactly what should happen to a gap the night before an interview.
The output of this step is often the most valuable artifact in the stack, because the gaps become the thoughtful questions asked at the end of the conversation. Those questions are frequently what interviewers remember.
Document 2: The Story Bank (25 Minutes)
Most interview answers fail not because the candidate lacks experience but because the experience arrives unsorted. A story bank fixes that by pre-building a small set of examples that can be adapted to many questions.
Start with raw input, not polished prose. Dump bullet points about past projects into the chat — rough is fine, fragments are fine. Include numbers wherever they exist: team size, timeline, budget, percentage change, volume handled. Then prompt:
"Below are rough notes on my work history. Convert them into a story bank of 7 examples. For each, use this structure: Title (5 words), Situation (2 sentences), Action (3 bullets, each starting with a verb), Result (1 sentence with a number if one exists in my notes), and Reusable For (list 3 interview question types this story answers). Do not invent numbers or details not present in my notes. Flag any story that lacks a measurable result."
The "Reusable For" field is the part most candidates skip and the part that matters most. A single story about rescuing a late project can answer questions about conflict, prioritization, stakeholder management, and failure — but only if the candidate has consciously mapped those connections beforehand. Under pressure, nobody makes that leap in real time.
The instruction to flag stories without measurable results is a quality filter. A story with no number is not disqualified, but it should be used for questions about judgment and values rather than questions about impact.
Pressure-Test the Bank
Once the bank exists, spend five minutes stress-testing it in the same conversation:
"Act as a skeptical hiring manager. For each story above, write the single toughest follow-up question you would ask to test whether the result was really driven by this person. Keep each question to one sentence."
Having a second model review the output adds another layer. Running the same stories through a different model in the AI Doc Maker chat — where ChatGPT, Claude, and Gemini are all available in one place — surfaces different blind spots. One model tends to probe for specificity, another for causality. The overlap is where the real weakness sits.
Document 3: The Question Map (15 Minutes)
This document is a lookup table, not an essay. One column lists likely questions. The next column lists which story from the bank to use. The third column holds a one-line opening sentence so the answer starts cleanly instead of with "so, um, there was this one time."
"Using the job description and my story bank above, build a table of the 15 most likely interview questions for this role. Columns: Question | Story to Use | Opening Line (one sentence) | Risk (what could go wrong in my answer). Order by likelihood. Include at least 3 questions that probe my weakest area based on the gap between the job description and my experience."
That last clause is uncomfortable and necessary. Every candidate has a gap between the posting and their resume. Asking the model to name it directly produces the questions that would otherwise cause a stumble.
This is also a document that benefits from spreadsheet format rather than prose. Generating it as a table in an AI document maker means it can be scanned in two minutes on the morning of, rather than re-read in full.
Document 4: The 30-60-90 Plan (20 Minutes)
A 30-60-90 plan is the single highest-leverage document a candidate can bring to a final-round interview. It shifts the conversation from "can this person do the job" to "here is how this person would do the job." That is a different and far more favorable frame.
The common mistake is making it generic. A plan that says "meet the team, learn the systems, build relationships" signals nothing. A plan that references the company's actual products, named processes, and the specific problem in the job posting signals genuine preparation.
"Draft a 30-60-90 day plan for the [Role] position at [Company], using the company brief and job description above. Structure: three sections, each with Objectives (3 bullets), Key Actions (4 bullets), and Success Looks Like (2 bullets). Days 1-30 should focus on learning and quick wins. Days 31-60 on delivering a first measurable output. Days 61-90 on scaling or systematizing. Reference specific products, teams, or challenges named in the source material. Keep the whole document to one page. Use no buzzwords — no 'synergy,' 'leverage,' or 'align.'"
Two rules for the result. First, stay humble in days 1–30: a plan that proposes restructuring a department in week two reads as arrogance. Second, make the day-60 deliverable concrete and modest — something a reasonable person could actually ship with sixty days of context.
Export this one as a clean PDF. Formatting is part of the message. A plan that looks like a rushed text file undercuts its own content.
Document 5: The Leave-Behind (10 Minutes)
The leave-behind is a one-page point of view on the hardest problem implied by the job description. If the posting mentions "improving onboarding efficiency," the leave-behind is a short take on how onboarding efficiency actually gets improved — three observations, two risks, one recommended first step.
It is not a consulting deliverable. It is evidence of thinking. Keep it to 400 words.
"Write a one-page point-of-view document titled '[Problem]: A First Take.' Structure: three observations about why this problem is hard (each 2 sentences), two risks of the obvious solution, and one recommended first step. Write from the position of an informed outsider who has not seen internal data — acknowledge that limitation once, briefly, without undermining the argument. Target 400 words. Plain language, no jargon."
The instruction to acknowledge the outsider limitation is important. A candidate who confidently prescribes solutions without internal data looks naive. A candidate who frames a view as a hypothesis to be tested looks like a colleague.
Time-Boxed Versions of the Same System
Not every interview gets a full evening. The stack compresses.
Ninety Minutes
Build all five. Allocate 20 / 25 / 15 / 20 / 10 minutes as described above.
Forty-Five Minutes
Build documents one, two, and three. Skip the external-facing documents entirely. Internal clarity beats an unpolished leave-behind every time.
Twenty Minutes
Build a condensed version of the company brief and the question map only. Use a single combined prompt:
"From the job description and company material below, produce two things: (1) a 150-word summary of what this company does and why this role exists, and (2) the 8 most likely interview questions with a one-sentence note on what the interviewer is really testing with each. Nothing else."
Five Minutes in the Parking Lot
Open the question map. Read the opening lines only. Pick the three stories most likely to be used. That is the whole drill.
The Mistakes That Waste the Evening
Generating Answers Instead of Material
Asking a model to "write my answer to 'tell me about yourself'" produces a script that sounds like a script. Asking it to structure genuine material into retrievable form produces something that can be spoken naturally. The distinction is between outsourcing the thinking and organizing it.
Skipping the Raw Input
A prompt with no source material produces generic output, because generic input is all it was given. The ninety minutes described here are maybe 20% prompting and 80% having pasted in the right raw material: the actual posting, the actual product pages, the actual career notes.
Over-Polishing Internal Documents
The story bank and question map will never be seen by anyone else. Spending fifteen minutes formatting them is fifteen minutes stolen from the 30-60-90 plan, which someone will actually read.
Building Fresh Every Time
The story bank is reusable across every interview a candidate will ever do. Build it once properly, save it, and the per-interview cost drops to the company-specific documents. By the third interview in a search, the whole stack takes forty minutes.
After the Interview: One More Document
Within two hours of walking out, generate a debrief. Dump everything remembered — questions asked, answers given, what landed, what did not, names mentioned, concerns raised — into the chat and prompt:
"Turn these raw notes into a structured interview debrief with sections: Questions Asked, My Weakest Answer and Why, Signals About What They Care About, Open Concerns They Raised, and Three Things to Address in a Follow-Up Email. Then draft the follow-up email in under 150 words, addressing the single biggest concern directly."
This document serves two purposes. It produces a follow-up email that actually responds to the conversation instead of thanking everyone generically. And it feeds forward: the "weakest answer" section tells the next prep session exactly what to fix.
The Underlying Principle
Interview preparation is a document problem disguised as a knowledge problem. The information needed is mostly public and mostly available. What is missing is structure — the conversion of scattered inputs into a small number of retrievable, usable artifacts.
That conversion used to take an entire evening of writing, which is why most candidates skipped it and read instead. It no longer does. A well-prompted AI document maker handles the structuring, which frees the candidate to do the part that cannot be outsourced: supplying honest raw material and deciding what to actually say.
Ninety minutes tonight. Five documents. One of them might be the reason the offer arrives.
About
AI Doc Maker
AI Doc Maker is an AI productivity platform based in San Jose, California. Launched in 2023, our team brings years of experience in AI and machine learning.
