Write a Transaction Clarification Request with AI at the Cursor

A bookkeeper can speak the verified details of an unclear transaction (date, reference, missing evidence and response deadline) and ask AI to turn them into a structured clarification request at the cursor, in a compatible Windows text field. The draft appears in the email, portal message or note field where you are already working. You then check every reference against the ledger or bank record, correct the draft and decide whether to send it. This article covers the inputs, a setup routine, the spoken request, a before-and-after example, a reusable template, a review checklist and a way to measure time to reviewed output.

Most bookkeepers ask one question many times: What is this payment, and where is the supporting document? The wording barely changes from one query to the next. The specifics change every time: which transaction, which reference, which document is missing and by when you need an answer.

Writing each request from scratch has two costs. The first is composition effort, because you rebuild the same polite, structured message repeatedly. The second is accuracy risk. When you rush, a reference gets transposed, the deadline stays vague or the recipient cannot tell what to provide.

A good clarification request does four things:

• It identifies the transaction unambiguously.
• It states what is missing in terms the recipient can act on.
• It gives a clear deadline for the reply.
• It leaves the answer open instead of guessing at it.

The last point carries the most weight in this role. A clarification request exists because something is unknown, and a draft that fills the gap with plausible detail defeats its purpose. The workflow below is designed to keep the AI inside the facts you supply.

Syping is SpeakLingo's term for combining speaking and typing: you speak, and text appears where your cursor is. The concept is explained on the Syping page, and the wider product scope is outlined on the features page.

Three capabilities matter for this task:

1. Online transcription converts your speech to text and supports 99+ languages, so you can dictate in a supported language suited to you and your recipient.
2. AI contextual formatting shapes the transcript into a usable structure, such as a short introduction, labelled details and a closing request, instead of one unbroken paragraph.
3. Spoken AI text requests let you say an instruction, such as 'write a transaction clarification request using only these inputs,' and receive text at a compatible, focused Windows text field.

Because the text lands in your message window, you avoid copying between a separate AI window and your email. Compatibility varies by application and field, and SpeakLingo is not universally compatible, so test the apps you actually use before relying on them. SpeakLingo is not designed to operate your computer on its own. It delivers text at your cursor, and you read it, edit it and decide whether to send.

A few minutes of setup keeps the draft in your hands until you have reviewed it.

One boundary matters before you start. To get a drafted request, use the supported AI Task workflow and put the instruction and the facts the draft needs (transaction date, reference, missing evidence, deadline) together in that same spoken request. Dictating alone does not turn AI on. Your previous dictations, any text already in the field and whatever is on your screen are not automatically available to the AI. Screen context is a separate opt-in, and it may be sent to remote services.

1. Choose a compatible, focused text field. Click into the email body, portal message box or note field where the draft should appear. Test with a harmless line first to confirm text arrives where you expect.
2. Turn off automatic Enter or send behavior. SpeakLingo has an optional setting that can press Enter after text is delivered, and in some apps that keystroke sends the message. For this task, turn off automatic Enter or send behavior so that keystroke cannot send the draft before you have read it, and review the draft before sending.
3. Check access for AI requests. Spoken AI requests need an internet connection, your consent, an enabled service and the necessary access. Transcription has its own mode, setup and access requirements, so an AI request being unavailable does not show that transcription is available.
4. Decide about screen context. If you opt in to screen context, that context may be sent to remote services along with your audio and text. Consider what is visible, such as client names or balances, and leave the option off if the screen holds details the request does not need.
5. Plan to review before sending. Read the whole draft against your source record, edit it, and only then send it yourself.

The scenario depends on four verified inputs. Collect them from the source records first, then speak them.

Keep the missing evidence specific. 'Backup' is vague. 'Supplier invoice or receipt showing what the payment covered' gives the recipient something to look for.

If you do not know a value, say so in your dictation. Saying 'amount not confirmed' is better than skipping the field, because the instruction asks the AI to flag missing facts. A flagged gap is easy to resolve during review, whereas an invented value is easy to overlook.

The instruction controls how closely the draft follows your inputs. This is the prompt for the scenario:

Using only these verified inputs, write a transaction clarification request. Inputs: [Transaction date, reference, missing evidence, response deadline]. Use a clear professional structure. Flag missing facts; do not invent them.

In practice you speak the instruction and then the inputs in natural language. For example: 'Using only these verified inputs, write a transaction clarification request. Transaction date fourteenth of March, reference TX-48213, missing evidence is the supplier invoice or receipt, response deadline twenty-second of March. Use a clear professional structure. Flag missing facts; do not invent them.'

Three habits help:

• Say references slowly and deliberately. Reference strings mix letters and digits, and transcription can mishear them. Alternatively, type the reference yourself after the draft appears.
• Name the recipient's role when tone matters. 'Addressed to the office manager' gives the structure something to work with.
• Keep one request per draft. If a client has three open items, run the process three times or list them as clearly separate inputs. Packing them into one spoken sentence invites confusion.

This scenario is fictional and non-sensitive. The business, reference and dates are invented for demonstration. The output below is illustrative, not observed from a real test of the app.

Situation: A bookkeeper preparing month-end records for a fictional business, Harbor Lane Studio, finds a payment with no supporting document attached.

Before: a typical rushed draft

Hi, there's a payment from March I can't match up, I think it was around the 14th, the reference looks like TX-4821 something. Can you send me whatever you have for it? I need it soon so I can finish up. Thanks

Problems with this version:

• The date is approximate and the reference is incomplete.
• 'Whatever you have' does not tell the recipient what to look for.
• 'Soon' gives no deadline.
• It gives no reason the question matters.

Spoken request at the cursor

Using only these verified inputs, write a transaction clarification request. Inputs: business Harbor Lane Studio, transaction date 14 March, reference TX-48213, missing evidence supplier invoice or receipt showing what the payment covered, response deadline 22 March. The amount is not confirmed. Use a clear professional structure. Flag missing facts; do not invent them.

After: a draft for review (illustrative output)

Subject: Clarification needed: transaction TX-48213 (14 March)

Hello,

While preparing the records for Harbor Lane Studio, I found a transaction that I cannot yet support with documentation.

Transaction details - Date: 14 March - Reference: TX-48213 - Amount: [not confirmed. Check the bank record before sending]

What is missing A supplier invoice or receipt showing what the payment covered.

What I need from you Please send the document, or let me know if one does not exist, so I can complete the records accurately.

Response deadline 22 March.

If you have any questions about this request, please reply to this message.

Kind regards, [Your name]

AI output varies from one request to the next, so a real draft may differ in wording and layout.

What the bookkeeper does next:

1. Opens the bank record and confirms the date and reference character by character.
2. Fills in the flagged amount from the record, or deletes the line if the amount is better left out.
3. Checks that the deadline matches the intended calendar day.
4. Reads the tone, adjusts it to the client relationship and sends the message personally.

Notice what the draft leaves out. It offers no opinion on how the payment should be categorised and makes no assumption about what the payment was for. The structure came from the instruction, and the facts came from the bookkeeper.

Keep this skeleton in your own notes. It works whether you speak the whole request or only a short instruction.

Spoken instruction template:

Using only these verified inputs, write a transaction clarification request addressed to [recipient role]. Inputs: business [name], transaction date [date], reference [exact reference], amount [amount or 'not confirmed'], missing evidence [specific document or explanation], response deadline [date]. Use a clear professional structure with a subject line, transaction details, what is missing, what I need, and the deadline. Flag missing facts; do not invent them. Do not comment on how the transaction should be treated or recorded.

Output structure to expect:

1. Subject line containing the reference and date
2. One-sentence reason for writing
3. Transaction details list
4. Missing-evidence statement
5. Specific request
6. Deadline
7. Closing and contact line

Add or remove sections to suit your own practice. Some bookkeepers add a line about what happens if no reply arrives by the deadline. That depends on your client arrangement, so include it only if you have confirmed the arrangement yourself.

This checklist is an illustrative writing practice. It is not a professional standard or a regulatory requirement, so adapt it to your own procedures and your firm's requirements.

Facts - [ ] The transaction date matches the source record exactly. - [ ] The reference matches character for character, including letters that sound alike when spoken. - [ ] Any amount was confirmed from the record, not from memory. - [ ] The client or business name is correct.

Scope of the request - [ ] The missing evidence is specific enough for the recipient to locate. - [ ] The request asks only for information or documents and does not state how the item will be recorded. - [ ] The draft has introduced no tax treatment, category or conclusion. - [ ] Placeholders such as '[not confirmed]' have been resolved or deleted.

Deadline and logistics - [ ] The deadline is a real date in a format the recipient will read correctly. - [ ] The reply route is clear: reply to the message, upload to a portal or another channel you actually use. - [ ] The recipient is the right person for this question.

Tone and sending - [ ] The tone suits the relationship. - [ ] Nothing contradicts earlier correspondence. - [ ] Automatic Enter or send behavior was off, and the draft is still unsent until you decide. - [ ] You have decided to send, and you send it yourself.

Knowing the source of each line shows where review effort belongs.

If the draft includes a sentence you cannot trace to your inputs, treat it as unverified and either remove it or check it.

Online transcription supports 99+ languages, so you can dictate in a supported language that fits you and your recipient. That can be useful when your clients read different languages.

Two cautions apply. First, this describes transcription support only. It does not mean your message is translated, and it does not guarantee AI-drafted output in every language, so read the final draft in the language your recipient will see. Second, dates and reference numbers can be misread across languages and date formats. Write the month as a word where misreading is possible.

The useful measure is the time from starting the request to holding a reviewed message that is ready to send. A draft that appears quickly but needs heavy correction may save little.

You can measure this yourself without relying on anyone else's figures:

1. Pick a representative batch. Choose routine clarification requests of similar complexity, ideally from the same period.
2. Time your current method. For each request, note the time from opening the record to a reviewed, send-ready message. Include looking up details, writing, restructuring and checking.
3. Time the cursor workflow. Run a comparable batch the same way: gather inputs, speak the request, wait for the draft, review and correct, and stop when the message is ready to send.
4. Record corrections. Keep a simple tally of what you had to fix: references, deadlines, tone, placeholders. The tally shows whether your spoken inputs need to be cleaner.
5. Compare like with like. Requests with complicated histories take longer under any method, so compare only similar items.
6. Repeat after some practice. Early attempts include learning time, so a later batch gives a fairer picture.

Your results describe your own routine. Use them to decide whether the approach suits the way you work. Other routine writing tasks are collected on the use cases page.

• Compatibility: Text appears at a compatible, focused Windows text field. Some applications or fields may not work, so test before depending on it.
• Connectivity and access: Spoken AI requests need an internet connection, an enabled service, your consent and the necessary access.
• Data handling: Audio and text may be sent to remote services to process a request. If you opt in to screen context, that context may also be sent. Consider what is on your screen and in your dictation, and follow your firm's and clients' confidentiality expectations. Leave out sensitive details the request does not need.
• Review: AI output is a draft. You are responsible for its accuracy and for the decision to send.
• Sending: The optional Enter-after-delivery setting can submit a message in some apps. Keep it off for this task and send manually after review.
• Interrupted internet: English transcription can serve as an offline fallback if the connection drops, provided the local runtime is set up. AI does not work offline, so the AI-formatted draft would have to wait.