Ten phrases, three routines, and a two-week review that makes them stick.

A Practical Starter Set of voice commands should be small enough to memorize and broad enough to cover a normal day. Ten phrases is the right size for most people: enough to cover capture, structure, prompts, translation, and a couple of device actions, and few enough to keep on one card. Write each phrase as you would say it, not as a description of what it does. Then write the result you expect beside it. A command you cannot describe in one result line is not ready for the set. Ten clear phrases beat forty impressive ones you have to look up.
Cover five categories and stop there. Capture commands turn a spoken thought into text at the insertion point. Structure commands create a heading, a list, or a table. Prompt commands turn a request into an AI-ready prompt with file names, paths, and commands. Translation commands produce a second language line. Device commands tell an agent which everyday task to run. Two phrases per category gives you a balanced set without turning the exercise into a catalog. If one category never appears in your week, drop it and add a second phrase to a category you use daily. Keep the two phrases per category equal in weight.
Choose by frequency, reversibility, and checkability. A phrase belongs in the set when you would use it at least once a day, when a wrong result costs seconds rather than hours, and when you can verify the result by reading it. A phrase that fails any of the three waits outside the set. This filter removes the commands that look clever in a demonstration and sit unused afterward. Keep the rejected phrases on a separate list with the reason, so you do not reconsider them every month. The set should reflect your actual week, not an imagined one. Write the reason in one line.
Put the ten phrases on a single card with three columns: phrase, expected result, and the place it lands. Keep the card visible for the first week and inside a document you already open for the second. Read the whole card aloud once before you start practicing, then read it again at the end of the week. The second reading tells you which phrases still feel unnatural; those are the ones to rewrite. The deliverable for this stage is a one-page card with ten phrases, their expected results, and their destinations. Read the card before the first practice session and again after the first week.
Learn one phrase per sitting, and keep the sitting to ten minutes. Say the phrase, watch the result, and compare it with the expected result on the card. If it matches, say it again in a slightly different context. If it does not match, change one word and try once more. Stop after three attempts; more attempts in one sitting teach frustration rather than fluency. Write one line about what happened and move to the next phrase tomorrow. Spacing the practice is what turns a card into a habit, and habit is the only thing that survives a busy day.
Record the exact wording that worked, not the wording you intended. Small differences in a name, a pause, or the order of two clauses change the result, and the exact version is the one you will need later. Write it beside the phrase on the card and date it. If two versions work, keep the shorter one and note the condition where the longer one is required. Delete the versions that failed, so the card does not fill with alternatives. A card that shows one working version per task is easier to trust than a card full of options. Then read the card aloud once.
Practice each phrase in its real destination, not in a blank document. A heading phrase behaves differently at the top of a page and in the middle of a paragraph, and a prompt phrase behaves differently when the editor already contains text. Open the app you actually use, put the insertion point where the text belongs, and speak. This small realism prevents the surprise of a phrase that works in a demo and fails in the work. The ten minutes are better spent in the real place, even if the document is a scratch copy. Change the document, not the phrase. Then try the same phrase in a message field.
Mark a phrase learned only after three clean uses in real work, on three different days. A clean use means the result landed correctly and you did not correct it. Until then, the phrase stays in the practice list even if it worked once. This bar is modest, and it prevents the common pattern where a good first attempt is mistaken for competence. Keep the practice list short by learning one phrase at a time; a long list of half-learned phrases is harder to maintain than a short list of solid ones. The deliverable is one learned phrase per day and a dated note of each.
Chain two phrases into a routine only after both are learned separately. A useful first chain is capture then structure: speak the note, confirm the text, then say the heading or list phrase that organizes it. A second chain is prompt then formatting: speak the request, confirm the result, then apply the structure you need. Keep each chain to two or three steps. Longer chains fail at the seams, where one step produces a result the next step does not expect. Two solid chains are more useful than one elaborate sequence you cannot debug. Write down the seam before you practice it. Keep the chain short.
Make the seam visible. Between steps, say the result you expect and confirm it on screen before continuing. If the first step produced a heading instead of a paragraph, the second step will attach to the wrong element. The confirmation takes a second and saves the minute you would spend untangling the result. When a chain fails, the useful question is which step produced the unexpected result, not which phrase is bad. Write the answer at the seam, then change one step at a time. Write the seam answer beside the chain, and change only the step that produced the surprise. Then rerun the chain.
Write a routine card for each chain with three lines: the steps in order, the check between steps, and the fallback if a step fails. Keep the card beside the phrase card. Use the routine card for the first week, then rely on the chain from memory and keep the card as documentation. If a chain needs a fourth step, split it into two routines rather than extending the card. Short routines are easier to remember, easier to check, and easier to hand to a colleague who has never seen your setup. Date each routine card when you first use it. Retire it when the chain changes.
Test each chain on a real task before you add another. Pick a task you do every day, run the chain from start to finish, and note the time and the corrections. A chain that takes longer than doing the steps separately is not ready. A chain that needs a manual move in the middle is not a chain. Keep at most three chains in the starter set, one for capture, one for prompts, and one for formatting. The deliverable for this stage is three routine cards with a named check at each seam. Compare it with the separate steps. Then keep the faster version.
Practice in the conditions where the commands will actually run. A quiet room teaches the phrases; a busy room teaches the corrections. Run the set once with background noise, once under mild time pressure, and once in a different window layout. Before the two-week practice block, download the current version of Voiskey on the computer you use every day, then confirm the phrases still behave the same way. If a phrase fails only under one condition, write the condition on the card instead of abandoning the phrase. Conditions are part of the command, not an excuse for it. Then note the condition in the practice log.
Change one condition at a time so the result stays interpretable. If you change the room and the layout on the same day, you will not know which one caused the difference. Keep a one-line practice log with the phrase, the condition, the result, and the correction. Read the log every few days and look for a pattern rather than judging each attempt. This is the same discipline you would use for any skill: controlled variation, honest notes, and a willingness to change the phrase instead of the goal. A pattern is worth more than a single good day. Then change the phrase.
Do not chase maximum speed. Voiskey supports up to 220 or more words per minute, but a starter set is about reliability, not velocity. Speak at a pace where the result needs no correction, then let speed arrive on its own. A fast phrase with three corrections is slower than a measured phrase with none. If you notice yourself speeding up to finish a task, stop and say the sentence again slowly. The habit you are building is a clean result, and speed is a side effect of repetition rather than a target. Say the last sentence again if it arrives wrong. Then move on.
Vary the destination as well as the room. Run one command in a message window, one in a document, and one in a note field. The insertion point is the same idea everywhere, but the surrounding behavior differs, and the differences are what a starter set must survive. Note which destinations feel awkward and why. Sometimes the fix is a shorter phrase; sometimes the fix is a different task for the same phrase. The deliverable is a practice log with conditions, results, and corrections for each of the ten phrases. Keep the log short enough to read. Then choose one phrase to rewrite.
Review the set after two weeks with the practice log open. Count how often each phrase was used, how often it needed a correction, and how often it was skipped. A phrase used daily with no corrections has earned its place. A phrase used once and corrected twice is a candidate for removal. A phrase never used is not a failure of the phrase; it is a sign that the task does not belong in your week. The counts turn a vague sense of progress into a short list of decisions you can make in ten minutes. Then make the three decisions in one sitting.
Replace at most two phrases at a time. The natural temptation is to rewrite the whole set after a mixed first fortnight, which resets the learning and hides which change helped. Pick the weakest phrase, write a replacement, and practice it for a week before touching another. If a chain is failing, fix the seam before replacing the phrase. Keep the retired phrases on a short list with the reason and the date. Two weeks is long enough to see a pattern and short enough that a bad choice has not become a habit. Keep the reason beside the retired phrase. Then date it.
Share the set with a colleague who does similar work. Give them the phrase card, the routine cards, and the practice log, then ask them to try three phrases in their own work. Their result will differ from yours, because a phrase depends on a voice, a vocabulary, and a set of habits. The differences are the most useful part of the review: they show which phrases are portable and which are personal. A team plan adds shared vocabulary and workflows, which gives a portable set a stable home when several people use it. Write their result beside yours. Then compare the two cards.
Close the review with a decision for the next fortnight. Keep the phrases that earned their place, replace the weakest one, and keep the set at ten. Before the next block of practice, download Voiskey on the computer you will use and confirm the phrase card matches the version you are running. Then schedule the next review and write the date on the card. The set is working when you reach for a phrase without reading the card. Start tomorrow with three phrases, and let the other seven arrive as the first three become automatic. Then put the card where you will see it.
Pick ten phrases across capture, structure, prompts, translation, and device actions. A phrase belongs when you would use it daily, a wrong result costs seconds rather than hours, and you can verify the result by reading it. Write each phrase with its expected result and the place it lands.
Plan one ten-minute sitting per phrase. Say it, compare the result with the expected result, change one element if needed, and stop after three attempts. Mark it learned only after three clean uses in real work on three different days. Speed comes from repetition, not from a longer session.
Learn both phrases separately first, then join two or three steps with a visible check at each seam. Confirm the result of each step before starting the next. Keep a routine card with the steps, the checks, and the fallback. If a chain needs a fourth step, split it into two routines instead.
Open the practice log and count daily uses, corrections, and skipped phrases. Keep what earned its place, replace at most two phrases, and fix a failing chain at the seam rather than rewriting the whole set. Ask a colleague to try three phrases, then keep the set at ten and schedule the next review.

Download Voiskey for Mac and Windows and start with three phrases tomorrow.
Use it across macOS, Windows, iOS, and Android