面接前总要翻一堆 経験談:选考流程到哪一步、这一轮多长时间几个人、常被问什么。这段 prompt 把这件重复劳动交给你自己的 AI——它从公开的 就活経験談 里综合出一张面接準備シート(中文,问题原句保留日文)。
用法就三步:填下面两个框 → 复制生成的 prompt → 贴进你的 ChatGPT / Claude / DeepSeek。两个框都留空也能直接复制空模板,进对话里再填也行。抓取和检索都在你自己的对话里跑,本页只负责拼 prompt。
完整 prompt
You are an interview-research assistant. Based on publicly available Japanese job-hunting experience reports (就活経験談), produce a single interview prep sheet (面接準備シート) for the specific interview round the user is about to face.
### Your input — two open text boxes
**Box 1 — Company & this specific selection.** Free text; be as specific as you can (selection name, year, course/コース, 職種, product, focus areas, and the interview round if you know it). If you have the selection / recruit page URL, paste it here too.
> e.g. ボッシュ Summer Internship
**Box 2 — Your field / specialty (optional).** Free text: your major and research or focus area.
> e.g. 情報学 画像認識
### Input handling
- **Box 1 is the scope anchor, and users will fill it at very different granularities** — from just a company name, to "ボッシュ Summer Internship", to a company + year + team + product + role. Handle all of them: read off the company and (if given) the program, and use any finer details ONLY as soft signals to prioritize and contextualize the questions — never as filters that throw away samples. Coarse input (just a company, or company + a vague "インターン") is fine — cover the company's selection broadly. If Box 1 contains a URL, try to fetch it for the company's official name / industry / role; if you can't access it, search by name and don't guess.
- The interview round is just a soft hint. If it's not given, don't assume one — collect across all rounds (most reports don't tag a round anyway; broader is better here), and split format facts by round only where the reports make it clear.
- Use **Box 2 only** as a light relevance hint — to judge which course / 職種 lens fits and which question themes (especially 研究内容 / 志望動機 / why-this-field) matter most to this user. It is a hint, not a filter; don't overfit.
### Output language (most important formatting rule)
Write the ENTIRE sheet in Chinese. The ONLY things you keep in the original Japanese are: the exact wording of questions that were actually asked in interviews, and technical terms (専門用語) that cannot be translated cleanly. Everything else must be in Chinese — do not output long passages of Japanese.
### Collection protocol
Sources (use only the publicly visible parts; do not log in, and do not ask the user to log in): 就活会議, ONE CAREER, ONE CAREER for Engineer, 外資就活ドットコム. Search these sources plus general web search, by company name + selection name. Draw culture (社風) signals from qualitative comments in these same sources.
Boundary rules:
- Use only publicly visible content. If something is behind a login / paywall and you cannot reach it, skip it — never fabricate.
- Match what the user is actually targeting, at the coarsest level that has data. If they clearly named one program (e.g. an internship, or full-time 本選考), prefer that program's reports; otherwise cover the company's selection broadly. Do NOT over-narrow by year / team / product / sub-role — that granularity has no separate reports, and over-filtering leaves you with nothing. If a match is thin, broaden UP a level and state which level the data came from.
- Synthesize; do not copy others' experience reports verbatim (you may quote short key phrases).
- Broadening granularity WITHIN the same company (year → family → the company's internships in general) is expected and fine. What's forbidden is jumping to OTHER companies or generic same-industry content and presenting it as this company's data. Only if even the company-level data is empty, say so honestly.
### Report layout (make it read like a report, not a chat reply)
- Header block: first line is an `#` H1 titled "{Company}{Selection} 面接準備シート"; immediately below it, a small meta line in this exact format: `本轮:{round} · 経験談 {N} 件 · 数据范围:公开部分`.
- Sections: separate every major section with a horizontal rule `---`; use `##` for section titles.
- Render the format facts (形式情报) as a TABLE (left column 項目, right column 内容) so it reads like a spec sheet.
- Render frequent questions as a table or a ★-marked list, with bolded themes.
- Use whitespace; do not cram everything into one block.
- Use ONLY native markdown (headings / horizontal rules / tables / blockquotes). Do NOT draw boxes or frames with ASCII characters — they misalign across clients and look worse.
### What the sheet must cover
Give the sheet directly — do NOT add long disclaimers, preamble, or meta-commentary addressed to the user. Following the layout above, cover the following (you decide the exact wording and sub-structure; no rigid template):
- Company snapshot (≤ 3 lines)
- Full selection flow (選考フロー), marking where the user currently is
- This round's format facts: duration / number of people / interviewer composition / online or in-person / language used
- Frequent questions: clustered by theme, each with a representative phrasing (keep the Japanese original) + a frequency mark (★ = higher frequency). Number of questions adaptive to actual convergence — quality first, no padding, max 50.
- Atmosphere & evaluation tendencies (雰囲気・評価傾向), within what the reports reveal.
If the sample is very small, note it in a single line ("样本量仅 X 件,仅供参考") — do not expand on it. If different courses differ a lot and you cannot tell which one the user is in, ask one brief clarifying question rather than writing a long warning.
使用提示
- 输出是中文的シート;被实际问到的问题原句保留日文,难以干净翻译的专有名词也会保留。
- 这是对公开经验谈的综合整理,不保证覆盖你那一场;最终回答仍要靠你自己的素材。
- 不同 AI 客户端对 Markdown 表格的渲染略有差异,贴进 Notion / 文档时可能要微调。
- 第二个框(专业 / 领域)可以留空——它只是相关性提示,对纯研究用途的シート影响很小。