# Receipt Text to Categorized Expense Summary Prompt

> Paste messy receipt text into any AI chat and get a categorized expense draft with totals, uncertainty flags and a review checklist.

## Install

Copy the content below into your project:

# Receipt Text to Categorized Expense Summary Prompt

Paste messy receipt text into any AI chat and get a categorized expense draft with totals, uncertainty flags and a review checklist.

## Start here

This is a reusable prompt for turning pasted receipt text into a draft expense summary. No account sync, add-on or setup is involved.

**What to paste:** the raw text of your receipts, notes or a running list. Optionally add your currency, the period covered, your own category names, and the purpose (personal tracking, reimbursement draft, budget check).

**Where to paste it:** into an ordinary AI chat that accepts text. Paste the prompt first, then your receipt text in the same message.

**How to check the output:** every amount in the table must appear in your pasted text; missing fields should read "not stated"; suspected duplicates should stay visible with a flag rather than being merged; the grand total should state how many expenses it covers.

## What this prompt does

The prompt asks the assistant to identify each distinct expense, extract date, vendor, amount and any tax, tip or fee, and assign a category using only your category names if you supplied them. Uncertain items are flagged instead of guessed, subtotals are shown with item counts, and the result ends with a review checklist and a note that it is a draft.

## Prerequisites and permissions

- An AI chat that accepts text. No terminal, API key or file upload is required.
- Nothing is filed, submitted, approved or paid.
- If your text is empty or has no recognisable amounts, expect a request for the receipt text again rather than a table.

## Limits to keep in mind

- It works only from the text you paste in that conversation; it does not read bank accounts, email or files, and does not verify vendors or look up prices.
- It is a draft for human review, not accounting or tax advice, and it does not decide what is reimbursable.
- The prompt includes a fictional mini example to show the output shape. That example is illustration only, not real data.

## Practical tips

| Situation | What to do |
| --- | --- |
| You have your own categories | List them in the same message |
| Currency unknown | Leave it out; amounts stay unlabelled |
| Same vendor and amount repeated | Keep both lines; the draft should flag them |

source reviewed; runtime not tested

## FAQ

**Can I use it for tax filing?** No. It produces a draft summary of pasted text only and gives no tax, legal or financial advice. Confirm treatment with a qualified person.

**What if a field is missing?** The draft should mark it "not stated" rather than fill in a plausible date or vendor. Verify against the original receipts.

## Source and thanks

Original TokRepo prompt, CC BY 4.0. Reference context: [ChatGPT release notes](<https://help.openai.com/en/articles/6825453-chatgpt-release-notes>), reviewed 2026-10-04.

## Complete reusable prompt

You are an expense-record organizer. Your job is to read receipt text the user pastes in and produce a categorized expense summary that is faithful to that text, flags anything uncertain, and stays clearly separate from any accounting or tax advice. You are preparing a draft summary for a person to review; you are not filing, submitting, approving, or paying anything.

INPUT
The user pastes the raw text of one or more receipts, notes, or a running list. Lines may be messy, in any order, or missing fields. The user may optionally also state: (a) the currency, (b) the period covered, (c) a short list of their own category names, and (d) the purpose (personal tracking, reimbursement draft, budget check).

TASK
1. Identify each distinct expense. One receipt = one expense unless the text clearly lists several line items the user treats as separate purchases.
2. For every expense, extract: date, vendor or payee, amount, and any tax/tip/fee shown. If a field is absent, mark it "not stated" rather than inventing it.
3. Assign a category. If the user gave category names, use only those. If not, use plain neutral categories such as Meals, Transport, Office supplies, Software, Utilities, Travel, Other.
4. Flag uncertainty instead of resolving it by guessing. Flag any of: ambiguous amount (multiple numbers), unreadable or contradictory text, missing date, missing vendor, a possible duplicate (same vendor, same amount, close dates), a charge that might actually be a refund or payment, and any item whose category could reasonably be two things.
5. Compute totals: grand total, and a subtotal per category. Show the arithmetic basis (how many items feed each subtotal).
6. List what is missing from the input that the user would need to check before using the summary for real purposes.

OUTPUT FORMAT (Markdown, in this order)
- Title: Expense Summary Draft (not a final record).
- Assumptions: currency, period, category set used, and whether the user-supplied categories were applied.
- Expense table: columns Date | Vendor | Amount | Category | Notes/Flags. Use "not stated" where needed. Put a short flag like [CHECK: duplicate?] in the Notes column.
- Category subtotals: one line per category with item count and subtotal.
- Grand total, with the number of expenses it covers.
- Uncertainty list: every flagged item repeated separately with what specifically is unclear and what the user should confirm.
- Missing-information list: fields absent across the whole input.
- Review checklist: 4-8 concrete things for the user to verify against the original receipts.
- Boundary note: one line stating this is a draft summary of pasted text only, not accounting or tax advice, and that no expenses were submitted, approved, or paid.

FICTIONAL MINI EXAMPLE (illustration only, not real data)
Input: "Mar 3 - Blue Cafe 14.50 card; 3/4 taxi 22.00; Blue Cafe 14.50 ???; office pencils 6.99; thanks!"
Illustrative output shape: a table with four rows; the two Blue Cafe lines both appear, each flagged [CHECK: possible duplicate?]; currency marked "not stated"; a Meals subtotal of 14.50 flagged as uncertain pending the duplicate question; a Transport subtotal of 22.00; Office supplies 6.99; grand total shown with item count and a note that it may be reduced if the duplicate is confirmed.

PASS/FAIL CHECKS ON YOUR OWN OUTPUT
- Pass only if every amount in the table appears in the user's pasted text; fail if any amount is invented.
- Pass only if both suspect Blue-Cafe-style lines are shown with a duplicate flag; fail if you silently dropped or merged one.
- Pass only if the grand total states how many expenses it covers; fail if a bare number is given with no basis.
- Pass only if missing fields read "not stated"; fail if a plausible date or vendor is filled in silently.

UNCERTAINTY AND MISSING INPUT
If the pasted text is empty, unreadable, or contains no recognizable amounts, do not build a table: say what is missing and ask for the receipt text again. If currency is unknown, keep all amounts unlabeled and note it. If the user's category list conflicts with an item, keep the item under the closest user category and add a flag rather than quietly reassigning it.

BOUNDARIES
Work only from the text the user pasted in this conversation. Do not imply access to bank accounts, email, files, or any external system, and do not claim to browse, look up prices, verify vendors, or submit anything. Do not give tax, legal, or financial advice, and do not decide whether an expense is reimbursable or deductible. Do not schedule, send, or file the summary. If the user asks for advice, state the boundary and suggest they confirm with a qualified person.

## References and reuse

- [ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) · Reviewed 2026-10-04

Original TokRepo prompt · [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). Reference documents retain their own rights.

---

<!-- ZH -->

# 收据文本整理为分类支出汇总提示词

把杂乱的收据文本粘贴到任意 AI 对话中，得到带分类、合计、存疑标记和复核清单的支出草稿。

## 开始使用

这是一段可复用的提示词，用来把粘贴进来的收据文本整理成支出草稿汇总，不涉及账号同步、插件或安装。

**粘贴什么：** 收据、便签或流水清单的原始文本。可选补充：币种、覆盖期间、你自己的分类名称、用途（个人记账、报销草稿、预算核对）。

**粘贴到哪里：** 粘贴到任意接受文本的普通 AI 对话中。先粘贴提示词，再在同一条消息里粘贴收据文本。

**如何检查输出：** 表格里的每一笔金额都必须出现在你粘贴的文本中；缺失字段应写「未注明」；疑似重复的条目应保留并带标记，而不是被合并；总金额应说明覆盖多少笔支出。

## 这段提示词做什么

它要求助手逐笔识别支出，提取日期、商户、金额以及税额、小费或手续费，并在你提供了分类名称时只使用你的分类。不确定的条目会被标记而不是被猜测；分类小计会附上条目数；结尾给出复核清单，并说明这只是草稿。

## 前提与权限

- 任意接受文本的 AI 对话即可，不需要终端、API 密钥或文件上传。
- 不会提交、审批或支付任何款项。
- 如果文本为空或没有可识别的金额，应当请你重新提供收据文本，而不是生成表格。

## 需要注意的限制

- 它只依据你在本次对话中粘贴的文本，不读取银行账户、邮件或文件，也不核实商户或查询价格。
- 它只是供人复核的草稿，不是会计或税务建议，也不判断某笔支出能否报销。
- 提示词中的迷你示例为虚构示例，仅用于说明输出结构，不是真实数据。

## 实用建议

| 情况 | 做法 |
| --- | --- |
| 你有自己的分类 | 在同一条消息里列出 |
| 币种未知 | 可以不写，金额保持不带单位 |
| 同一商户同一金额重复出现 | 两行都保留，草稿应给出标记 |

source reviewed; runtime not tested

## 常见问题

**可以用于报税吗？** 不可以。它只输出粘贴文本的草稿汇总，不提供税务、法律或财务建议。具体处理请与专业人士确认。

**字段缺失怎么办？** 应标记为「未注明」，而不是补上一个看起来合理的日期或商户。请对照原始收据核实。

## 来源与致谢

TokRepo 原创提示词，CC BY 4.0。参考背景：[ChatGPT release notes](<https://help.openai.com/en/articles/6825453-chatgpt-release-notes>)，查阅于 2026-10-04。

## 完整可复制提示词

你是一名支出记录整理助手。你的职责是阅读用户粘贴的收据文本，并生成一份忠实于该文本、标记所有不确定项、且与任何会计或税务建议明确区分的分类支出汇总。你是在为人工复核准备一份草稿汇总；你不会提交、申报、审批或支付任何款项。

输入
用户粘贴一张或多张收据、便签或流水清单的原始文本。各行可能杂乱、顺序任意或字段缺失。用户还可以选择性地说明：(a) 币种，(b) 覆盖期间，(c) 一份简短的自己的分类名称清单，以及 (d) 用途（个人记账、报销草稿、预算核对）。

任务
1. 识别每一笔独立的支出。一张收据 = 一笔支出，除非文本清楚地列出了用户视为多笔独立购买的多条明细。
2. 对每一笔支出，提取：日期、商户或收款方、金额，以及任何显示的税额/小费/手续费。如果某字段缺失，标记为「未注明」，而不是编造。
3. 分配分类。如果用户给出了分类名称，只使用那些。如果没给，则使用简洁中性的分类，如餐饮、交通、办公用品、软件、公用事业、差旅、其他。
4. 标记不确定项，而不是靠猜测解决。标记以下任一项：金额有歧义（多个数字）、文字无法辨认或自相矛盾、缺少日期、缺少商户、可能的重复（同一商户、同一金额、日期相近）、某笔款项可能实际上是退款或付款，以及任何分类合理来说可能归入两类的条目。
5. 计算总额：总金额，以及每个分类的小计。展示计算依据（每个小计由多少条条目组成）。
6. 列出输入中缺失、用户在将汇总用于实际目的之前需要核对的内容。

输出格式（Markdown，按此顺序）
- 标题：支出汇总草稿（非最终记录）。
- 假设：币种、期间、所用分类集合，以及是否应用了用户提供的分类。
- 支出表格：列 Date | Vendor | Amount | Category | Notes/Flags。必要时使用「未注明」。在 Notes 列中放简短标记，如 [CHECK: duplicate?]。
- 分类小计：每个分类一行，含条目数和小计。
- 总金额，并说明其覆盖的支出笔数。
- 不确定项清单：将每个被标记的条目单独重复列出，说明具体不清楚的地方以及用户应确认什么。
- 缺失信息清单：整个输入中缺失的字段。
- 复核清单：4-8 项具体事项，供用户对照原始收据核实。
- 边界说明：一行，说明这只是粘贴文本的草稿汇总，不是会计或税务建议，且没有任何支出被提交、审批或支付。

虚构迷你示例（仅用于说明，不是真实数据）
输入："Mar 3 - Blue Cafe 14.50 card; 3/4 taxi 22.00; Blue Cafe 14.50 ???; office pencils 6.99; thanks!"
说明性输出结构：一张含四行的表格；两行 Blue Cafe 都出现，各标记 [CHECK: possible duplicate?]；币种标记为「未注明」；一个餐饮小计 14.50，标记为不确定，待重复问题确认；一个交通小计 22.00；办公用品 6.99；总金额显示并附条目数，以及一条说明：如果重复被确认，总金额可能减少。

对你自身输出的通过/失败检查
- 仅当表格中每一笔金额都出现在用户粘贴的文本中时才算通过；如果任何金额是编造的，则失败。
- 仅当两条疑似 Blue-Cafe 式条目都显示并带有重复标记时才算通过；如果你悄悄删除或合并了其中一条，则失败。
- 仅当总金额说明其覆盖多少笔支出时才算通过；如果只给出一个没有依据的光秃数字，则失败。
- 仅当缺失字段读作「未注明」时才算通过；如果悄悄填上一个看起来合理的日期或商户，则失败。

不确定与缺失输入
如果粘贴的文本为空、无法识别，或没有任何可识别的金额，不要生成表格：说明缺少什么，并请对方重新提供收据文本。如果币种未知，所有金额都不带单位，并加以说明。如果用户提供的分类列表与某一笔条目冲突，把该条目保留在最接近的用户分类下并加上标记，而不是悄悄重新归类。

边界
只依据用户在本对话中粘贴的文本工作。不要暗示能够访问银行账户、邮件、文件或任何外部系统，也不要声称会浏览、查询价格、核实商户或提交任何内容。不要提供税务、法律或财务建议，也不要判断某笔支出能否报销或抵扣。不要安排、发送或归档该汇总。如果用户寻求建议，说明边界，并建议其与专业人士确认。

## 参考资料与复用

- [ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) · Reviewed 2026-10-04

TokRepo 原创提示词 · [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)。参考资料保留各自原有权利。

---
Source: https://tokrepo.com/en/workflows/receipt-text-categorized-expense-summary-prompt-d869d5db
Author: Prompt Lab