A prompt for reading a ministry's buying pattern with an AI assistant

The problem this page addresses

Researchers look up what a Japanese ministry buys, how it buys it and from whom when they shortlist prospects, size a market for an internal report or check where a competitor wins work. Pasting a list of award results into an AI assistant and asking for the ministry's buying pattern tends to return generalities, or a summary that mixes in figures the data never contained.

Hashike is a private, bilingual research service that organizes US federal contract opportunities performed in Japan and Japanese central-government award results. It is not affiliated with any government. It takes the award results ministries publish on Japan's Procurement Portal (調達ポータル) and totals them by ministry and by supplier corporate number (法人番号, the national identifier every Japanese company carries).

This page documents one reusable prompt for having an AI assistant read a single ministry's data from Hashike. By the end you will have the English prompt to copy, the way to pull the data that goes into it, and an outline of what the answer will and will not contain.

• ·Award result: the published record that a procurement was awarded, naming the supplier and the award price (落札価格, the winning bid). One procurement can produce several award results.
• ·Prompt: the instruction text you give an AI assistant. The prompt on this page is in English; a variable in it sets the language of the answer, so the answer can come back in Japanese.

The records behind the prompt

Hashike holds the central-government award results published on the Procurement Portal over the most recent 60 months. In that window there are 167,444 award results, with a total award value of ¥5,779,856,658,233, won by 17,515 distinct suppliers counted by corporate number, across 45 ministries with at least one award result.

Within the same window, the ministry with the largest total award value is 国土交通省 (the Ministry of Land, Infrastructure, Transport and Tourism), with 41,311 award results totalling ¥1,105,189,106,402.

Pulling the data the prompt needs

The prompt takes Hashike's summary of one ministry plus its list of recent award results. There are three ways to get it, all drawing on the same records.

1. In a browser, open the ministry from the buyers index (/buyers). The ministry page lists the award count, total award value, distinct suppliers, contract-method mix and recent awards. Copy that content into the data slot of the prompt.
2. If your AI assistant supports MCP (Model Context Protocol) and is connected to Hashike's MCP server (the address is https://hashike.io/api/mcp ) have it call the buyer_profile tool with ministryCode set to <ministry code> and includeAwards turned on. The result lands in the same conversation, and the prompt can follow it directly.
3. Through Hashike's public API, call the buyers endpoint with the ministry code and a Hashike API key in an Authorization: Bearer <token> header, then paste the returned JSON into the data slot.
4. To narrow to one item or one fiscal year, search award results with find_comparable_awards (the awards endpoint in the API) using ministryCode, fiscalYear or query, and pass those results alongside the buyer_profile output. The tool requires at least one filter and has no unfiltered listing.

Hashike does not offer a CSV or Excel export. The data comes out by copying from the page, through the API, or through MCP.

The prompt

You are reading award results from one Japanese central-government ministry. The data below comes from Hashike, which compiles award results published on Japan's Procurement Portal (調達ポータル). Amounts are award prices (落札価格, the winning bid) in Japanese yen, as published.

Ministry: <ministry name>
Ministry code: <ministry code>
Window: <window start> to <window end>. This is a rolling window over award date, not a fiscal year.
Answer language: <Japanese or English>

DATA:
<paste the buyer_profile output here, plus any find_comparable_awards results>

Rules:
- Work only from the DATA above. Do not add suppliers, awards, amounts or dates that are not in it.
- Copy every figure exactly as it appears. If you calculate something (a share, a count of repeats), show the figures you used.
- For every figure, say what it counted and over what period.
- Keep yen amounts in yen. Do not convert them, and do not compare them with US dollar figures.
- If a question below cannot be answered from the DATA, say so in one sentence and move on.
- Describe what the records show. Do not advise whether to bid, partner or register.

Report under these headings:

Scale: the number of award results, the total award value and the number of distinct suppliers, each as given, with the window it covers.

Contract methods: the contract-method mix as given in the DATA (for example 一般競争入札 or 随意契約). Name the method that accounts for most award results, if the DATA shows one.

Repeat purchases: procurement names in the recent awards that recur or describe the same kind of goods or service. Group them, and list the award dates for each group.

Supplier concentration: the suppliers that appear most often in the recent awards, with their corporate numbers (法人番号). Say whether a few suppliers account for most of those awards. State that the recent-awards list is a subset of the window, not the whole of it.

Timing: months in which award dates cluster, read from the recent awards. Japan's fiscal year runs April to March; say where clusters fall within it.

Limits of this reading: rows without a corporate number, multiple-award procurements (複数落札) that produce several award results for one procurement, and anything else in the DATA that limits the reading.

Variables to substitute

• ·<ministry name> and <ministry code>: the ministry's name and the ministry code Hashike uses. Both appear in the buyer_profile output.
• ·<window start> and <window end>: the period the data covers. Hashike's ministry figures use a rolling 60-month window over award date, which is not cut at fiscal-year boundaries.
• ·<Japanese or English>: the language of the answer. The prompt stays in English and the answer can still come back in Japanese.
• ·The DATA slot: the copied page content, the API's JSON or the MCP result. When the MCP result is already in the conversation, a line such as "the buyer_profile result above" is enough.

What the answer looks like

The answer comes back under six headings: scale, contract methods, repeat purchases, supplier concentration, timing and the limits of the reading. The scale and contract-method figures are copied from Hashike's output. Repeat purchases, supplier concentration and timing are read from the recent-awards list.

That split lets you tell, inside the answer, which statements describe the whole ministry and which describe a pattern in part of the list. Checking the figures means matching each one against the same output you gave the assistant.

What this prompt cannot do

An assistant does not always follow its instructions. The figures in the answer hold only once they have been matched against the data you supplied; the related page on checking an AI's answer against Hashike figures sets out that procedure.

• ·Hashike is built entirely on public records. It has no access to non-public government information and is not affiliated with or endorsed by any government.
• ·Only central-government award results are included. Prefectural and municipal procurement is not.
• ·Amounts are award prices in yen as published. They are never converted and never combined with US federal figures.
• ·A small share of award rows carry no corporate number and cannot be tied to a supplier. A row missing from a supplier's record means it could not be matched, not that the supplier was inactive.
• ·Multiple-award methods (複数落札) let several suppliers win one procurement, so a row count is a count of award results, not of procurements.
• ·The window is a rolling window over award date, not a fiscal year. Japan's fiscal year runs April to March and the US federal year October to September; a figure spanning both would reconcile two calendars silently.
• ·Absence of a record under a set of filters is not evidence that the underlying activity did not occur.

Supplier concentration and timing are read from the recent-awards list. They may not match the pattern across the ministry's full 60-month window.