A spoken workflow for searching, noting, and citing without losing sources.

Voice Commands for Browser research work best when the question is settled before the first tab opens. Write the main question in one sentence, then break it into three sub-questions that can each be answered by a specific source. Decide what would count as a good answer, such as a dated report, an official notice, or two independent explanations that agree. Set a time box for the search so the session does not expand into browsing. Keep the question and the sub-questions on a card beside the keyboard. A clear frame turns a wandering session into a short sequence of queries and makes the spoken commands easier to choose.
Identify the source hierarchy before you start. Name the kind of source that would settle each sub-question and the kind that would only provide context. For a policy question, the primary notice outranks a summary article; for a market question, a dated report outranks an undated blog post. Write the hierarchy on the card and refer to it when a result looks convenient but weak. If you cannot name the source that would convince you, the question is still too vague. A hierarchy prevents the common failure where the first readable page becomes the answer by default. Review it after the first three results and adjust if a stronger source type appears.
Choose the first three queries and say them aloud before you type or dictate anything. Each query should target one sub-question and use the vocabulary of the field rather than a casual paraphrase. Remove words that narrow the result too early, such as a specific year, unless the question requires it. Read the queries to a colleague and ask which one they would run first; their answer often reveals an assumption you missed. Keep the queries on the card and mark the one that produced the best result. The first three queries set the direction for everything that follows.
Set a stopping rule for the framing stage. Stop when each sub-question has a named source type, a first query, and a definition of a good answer. Do not open a tab until those three items exist for at least two sub-questions. If a sub-question resists framing, drop it from this session and note it for a later one. A narrow session that answers two questions well is more useful than a broad session that touches six. Write the stop condition on the card so you can recognize the moment instead of drifting into the search. If the rule is met early, start the search rather than adding another sub-question.
Dictate the query into the field where you normally search, then dictate a short note before you move to the next result. The note should record the query, the source name, and the one sentence that answers or challenges the sub-question. Keep the note in a document open beside the browser so the two stay visible together. Voiskey places text at the insertion point and adapts punctuation to the sentence, which means you can speak a full thought without stopping to format it. Review the note once and correct names or figures before continuing. One pass through the results produces a usable draft instead of a pile of unread tabs.
Separate what the source says from what you think about it. Start the quoted part with the phrase according to, then give the claim, then start a new sentence with your reaction. This habit makes later verification easier because the two kinds of statement never blend into one. If a source uses a term you do not recognize, dictate the term and a question mark rather than guessing a definition. Mark uncertain notes with a short label such as check this, then return to them at the end. Clear separation is more valuable than elegant prose, because the notes exist to support a decision.
Use voice for the parts that are hard to type and typing for the parts that need precision. A long explanation, a paraphrase, or a summary of a paragraph is a good candidate for speech. A URL, an exact figure, or a proper name may be faster and safer to type. If you dictate a number, read it back immediately and compare it with the source. Read any dictated figure twice before you move on. Keep a consistent place in the note for links so they do not get lost between sentences. Mixing the two input methods is normal; the goal is a complete note, not a demonstration of one technique.
Pause every few results and read the notes against the sub-questions. If a note does not help answer or refine a sub-question, it belongs in a parking area rather than the main draft. If two sources disagree, record both claims and the reason for the disagreement instead of choosing one early. Keep the notes in a fixed order that matches the sub-questions, so a reader can follow the reasoning later. At the end of the pass, count how many sub-questions have at least one usable note. That count tells you whether to continue searching or start verifying. Write that count on the note so the next pass starts with a clear gap.
Name each tab after the source it contains, not after the query that opened it. A tab called ministry notice is easier to return to than one called results page four. When you open a new source, say the name before you start reading, and close duplicates as soon as you notice them. Keep no more than seven tabs open at once; beyond that, the tab bar becomes a second search problem. If you need to keep more sources, record them in the note and close the tab. Rename a tab when the source changes, not only when you open it. A named, bounded set of tabs is easier to navigate by voice and easier to reconstruct after an interruption. Say the new name once after renaming so the change is easy to remember.
Keep a source log with four fields: the source name, the link, the date you accessed it, and the claim you plan to use. Fill the log while the source is open rather than at the end of the session. If a source supports two sub-questions, give it two rows so each claim has its own trace. Read the log aloud at the end of each pass and check that every row has all four fields. If a field is missing, mark it rather than leaving it blank. A complete log turns a pile of impressions into a set of citations you can defend. It also shows which sub-questions are still unsupported.
Use short spoken commands to move between sources, and keep the same command for the same action. If you ask the agent to open a new tab, name the tab in the same breath. If you ask it to return to the previous source, use the name rather than a number, because numbers change as tabs open and close. When a command fails, note the failure and use the mouse once rather than repeating the phrase five times. Consistency in navigation saves more time than speed in any single command, especially in a session with many sources. Keep a one-line navigation list beside the log so the names stay consistent.
Protect the trail from the two common losses: a closed tab with no note and a copied link with no source name. Before closing anything, check that the source log has a row for it. Before pasting a link into the note, add the source name and the date beside it. If you are interrupted, restart by reading the last complete row of the log and the title of the current tab. The log becomes the map back into the session. A trail that survives an interruption is the difference between a resumable research task and a restart. Do this check before any break, because a paused session is where trails usually break.
Verify every claim that will appear in the final summary. Open the original source again, confirm the date and the author, and read the sentence around the claim rather than the sentence alone. Check whether the claim is stated as fact, as an estimate, or as an opinion. Compare the exact wording with your note and correct any drift introduced during dictation. If the source has changed since you accessed it, record both versions and the access dates. Verification is slower than collection, but it is the step that turns notes into something another person can rely on. Keep the source open until the corrected note is saved.
Look for a second source when a claim carries weight in the decision. The second source does not need to agree in every detail; it needs to be independent and to address the same question. If two sources disagree, write a short note explaining the difference and decide which one to trust based on the source hierarchy on your card. Do not average the two claims or present the disagreement as agreement. A named disagreement is more useful to a reader than a false consensus, and it protects you when the question is revisited later. Record the second source in the log with the same four fields. Check that the second source is dated after the event it describes.
Ask a colleague to check one note against its source without seeing your summary. Their task is to say whether the note accurately represents the source, not whether they agree with the conclusion. If they find a mismatch, correct the note and check the other notes from the same source. Give them the source log as well, so they can see which claims are still unsupported. A short independent check catches the errors that repetition has made invisible to you. Record who checked what and when, so the verification has a trace. Keep their comments beside the note so a later reader sees the check. Give them a copy of the hierarchy so their check reflects your standard.
Decide what to do with unverifiable claims. If a claim cannot be traced to a source, remove it from the summary or label it clearly as an open question. If a source is inaccessible, note the access problem and the date, then look for a citable alternative. Do not promote a memory of a page into a citation. When a claim is important and unverified, say so in the summary and name the check that would settle it. Candor about an open claim is stronger than a confident sentence with no support behind it. An open question with a named check is a result, not a failure. Write that check as a task with an owner rather than a vague hope.
End the session by writing a summary in three parts: what you found, what remains open, and what you will do next. Limit the findings to three sentences, each tied to a source in the log. Keep the open questions to two, with a note about what would answer each one. State the next action as a verb and a deliverable, such as draft the comparison table or send the open question to the policy owner. Read the summary aloud once and check that every finding has a trace. A short, sourced summary is easier to hand over than a long narrative.
Before you close the browser, clean the trail. Archive the notes and the source log under a name that includes the question and the date. Close tabs that have a row in the log, and keep only the sources you still need for the next action. If you plan to continue tomorrow, write one sentence describing where to start. This small handover note is often more useful than the summary, because it preserves the exact place where the work paused. A clean trail makes the next session start with a query rather than a reconstruction. Name the archive so it sorts beside the question card.
If a colleague will continue the research, give them the question card, the source log, and the summary. Walk through one finding and one open question so they see how the evidence connects. Ask them to add one source of their own before the handover is complete; that small task confirms they can use the trail. When they want to run the same workflow, have them download Voiskey for Mac or Windows and start with their own question card. Keep the card format identical so the two sessions can be compared later. Note the date of the handover beside their name.
Close the loop with a decision about the session itself. If the summary answered the main question, record the answer and the evidence that supports it. If it did not, record the missing source or the unclear sub-question and schedule a narrower session. If the workflow was slower than manual research, note which step caused the delay and change one element next time. Archive the card and the log together so the next session starts from a known state. A research routine improves only when each session ends with a written judgment about what worked. Keep the judgment short enough to reread before the next session.
Yes, for the typing parts of the workflow. Voiskey places spoken text at your insertion point in the app you are already using, so you can dictate a query, a note, or a summary without switching tools. Keep a source log, name each tab, and verify every claim against the original before you cite it.
Log four fields for every source: name, link, access date, and the claim you plan to use. Fill the row while the source is open, give a source two rows if it supports two claims, and read the log aloud at the end of each pass. Close a tab only after its row is complete.
Open the original again, confirm the date and author, and read the sentences around the claim. Compare the source with your note and correct any drift. For weighty claims, find an independent second source. If the two disagree, record the difference rather than blending them into a false agreement.
Three findings, each tied to a logged source, plus two open questions and the check that would settle them. End with one next action written as a verb and a deliverable. Read the summary aloud once, confirm every finding has a trace, then archive it with the source log and the question card.

Download Voiskey for Mac and Windows and run one research question from query to summary.
Use it across macOS, Windows, iOS, and Android