A 60-second pre-send dictation check catches misheard names, numbers, dates. Read the message aloud once before you hit send. Your ears catch what the screen hides.

Dictation errors are not spread evenly through a message; they cluster. The sentence about the schedule, the paragraph of commentary, the polite close, all of it transcribes cleanly most of the time, and then the specific name, the payment figure, or the meeting date arrives garbled. That clustering is the useful fact. It means the review does not have to be a full re-read; it can be a hunt for three kinds of word. A pre-send dictation check is a hunt, not a proofread, and a hunt takes a minute, not a scan. One real example: the client Ms. Cahn arrives as Ms. Kahn, the price $1,400 lands as $4,400, and the meeting on the 3rd becomes the 13th, all inside a paragraph that otherwise reads perfectly.
Why do these three fields fail? Name, numbers, and dates are the fields with the least context, and the engine predicts from context. A common name matches a common pattern, a rare name matches nothing, and the engine picks the neighbor. A figure depends on digits the ear shares with words like fifteen and fifty, and the context around them is thin. A date can be said in three different orders and dropped into a sentence full of other temporals. The engine is not bad; it is working on the most prediction-dependent words in the language. The fix is not better speaking but better looking: the passes below exist because the engine's guess is always the most probable version, and for a rare name the most probable version is the wrong name.
The professional reply makes the cluster worse: these are exactly the fields a reader cannot silently fix. A typo in a paragraph is interpreted, a misheard name in a signature is noticed, and a wrong date in a confirmation is acted on. The consequence is that the check protects the smallest and most expensive part of the message. The paragraph can carry a stray word all week; the figure cannot carry a wrong digit for an hour. A wrong date puts the reader at the wrong meeting, a wrong figure gets a sign-off on the wrong number, and a wrong name in a signature reads as carelessness. Each is a one-character fix, and each costs a conversation to repair.
The scan is four passes, one minute, in a fixed order. Pass one, names: every proper noun in the draft, said once again while watching the transcript, with the rare ones spelled. Pass two, numbers: every figure, read from the transcript against what you meant, digit by digit for the ones that matter. Pass three, dates and times: the day, the month, the year, checked in the form the reader will read. Pass four, the whole draft at half speed, reading only for the sound, not the sense. The order is fixed because the fields fail in a predictable order: names first, since they are rare and silent, then numbers, then dates, then the sound pass over everything that remains.
The passes run on a rule of three: if a field appears three times, check it all three times, because the third instance is the one the eye skips. If a name appears once, check it once and spell it once. The common failure of the check is the quick version, the glance that sees the name is spelled right and moves on, which is exactly the version that misses the right name spelled like a different name. The check is slow against the fields, and fast against the paragraph. The same rule applies to the two places a field appears: the greeting and the signature carry the same name, and the version checked at the top is not the version that ships at the bottom.
Do the check before the body edit, not after, because the body edit absorbs the attention. The order in a minute: dictation, scan fields, one pass for sound, then any content change. When the content edit comes first, the eyes are in the argument, and the fields pass without a glance; that order is how a wrong date ships in an otherwise perfect paragraph. The fixed order is the whole value of the minute, and the minute is small enough to keep. The one-minute budget matters for another reason: a check that takes three minutes gets deferred, while the minute finishes before the paragraph edit starts, and the message leaves checked and still fresh.
Homophones are the fourth cluster, and the one that reads right. Please and pleas, accept and except, its and it's; in a dictated draft the wrong twin can be grammatically complete, which is why the sound-alike check is a read for the ear rather than the eye. In a dictation, the mishear often arrives as the twin of the word you actually said, absent any cue that it is wrong, and only the sense catches it. That is why the pass reads aloud what is written, not what was meant. The class is wider than the schoolbook pairs: principal and principle, cite and sight, and the business pairs, lead and led, waiver and waver. A wrong lead in a delivery note changes the whole sentence's meaning.
The fix for a persistent pair is a spoken spelling, but the fix for the married pair, the one that recurs in your own writing, is a choice of words. If you always trip on between and among, or on the client name that sounds like a different name, rephrase the sentence rather than spell the word every time. The habit of swapping the sentence shape costs no time and removes the risk for good. Keep a list of the pairs that have bitten you, and scan those two words specifically in the pass. The list is three lines long in most cases, and the scan becomes a search for three words rather than a grammar review, which is the narrowing the check needs to stay under a minute.
Under time pressure the homophone check narrows to the dangerous few: the legal pair, accept and except; the quantity pair, less and fewer, only if the client cares; and the deadline pair, by and before, which is a business fact and not a style choice. For a deadline, say both: by Thursday, before the end of the day, and let the transcript show which you said. The check that defends the deadline is the check that protects the scheduling coordinator's week. The narrowing applies to the client's own vocabulary too: the word the client uses every week, the retainer, the turn, the close, gets spoken and confirmed in the pass, because those are the words the partnership runs on.
A mishear log is fifteen lines a month that pays like a department. Each time the check catches an error before a client does, write one line: the field type, the word misheard, the fix. Recruiters catch names, coordinators catch times, transaction staff catch figures; the log shows which cluster is yours. The personal pattern, once seen, narrows the check further: a coordinator who never catches a name can skip the name pass without risk, because the log says the risk lives in the times. The first month is usually the surprise: the field the writer feared most comes back clean, and the field that ships wrong is the figure at the end of the second sentence. That discovery is what the log is for, and it pays for the habit on its own.
The log also catches the tool's quirks, not just yours. If the same name has shipped wrong three times in two weeks, the engine has learned a competitor for it, and the answer is a phonetic spelling: say the name, spell a sound-alike variant that the engine owns, or switch to saying just the initials for that client. The log is the difference between checking every name forever and checking the five names that matter. The same fix covers figures: a number that has shipped wrong twice gets the digits said one at a time, seven, four, zero, which breaks the pattern the engine has learned. Each quirk is found once and fixed for good.
The discipline of the log is honesty about the near miss. An error caught by the log is an error that never left; count the near misses as assets, not as failures, because a proofing habit is exactly the sum of its catches. In the month the log is kept, most people find three errors a week they would have sent, and the discovery of that number, not the number of errors, is what turns the check from a suggested habit into a practiced one. The second value is the pattern the count exposes: three errors a week in the same field is not bad luck, it is a changed habit or a changed engine, and the log shows which one inside a month.
The check pays in two conditions worth naming. Weak reception, the car, the train tunnel, the hotel room, is the first, because the transcription loses quality where the speech loses its clarity, and the errors double exactly when the draft's stakes are highest. The rule: any draft dictated on a bad connection gets the full four passes, no shortcuts, before it leaves. The second is the late night, the draft dictated at the end of the day, when the eye reads what the mouth meant and supplies the missing word from memory. The two conditions share one cause: the signal is degraded, either in the air or in the writer. The check costs the same for both, and the draft's stakes are highest in both, so the rule is the four passes without exception.
The late-night draft is the one that fools the most people, because the error is more likely to be the engine's and the reader is more likely to be the writer's own tired idea of the text. The fix is scheduling, not willpower: draft late, check in the morning, or draft late and check before the send with the four passes run one at a time. The minute is most valuable at ten at night, and the minute is least likely to be taken at ten at night, which is exactly the gap the log is meant to close. For the draft that cannot wait, run the passes one at a time: the name pass, then the number pass, each a few seconds, and the check still holds at the hour it matters most.
The concluding rule is the rule of the second reader. When the stakes are high, the four passes are still the floor, but the floor is not the ceiling: one honest second reader, human, or the check done an hour later with fresh eyes, catches what the first minute cannot. The pre-send check is not a substitute for judgment, it is insurance on the words, and the words that hurt most are the ones that were nearly right. A second reader does not need the full context, only the four fields: the name matches the sender, the figure matches the deal, the date matches the meeting, and the sentence reads as the sender meant it. Thirty seconds of fresh eyes closes the gap the minute leaves.
Under a minute for a normal reply. Four passes: names, numbers, dates, then one read for sound. The common shortcut that fails is reading the whole draft looking for anything wrong; the pass structure is what keeps the check short and targeted at the fields that actually mishear.
Not in the minute. Full proofreading has its own slot, and it is a different skill that reads for grammar and flow. The scan is for the three clustered fields plus homophones. Trying to proofread everything in one pass is how the names and dates slip through.
Dates have thin context and several valid orders, and the engine guesses the nearest match. Say the date in the order the reader will see it, say the year out loud even when you think it is obvious, and on the check, read the transcript date against your calendar. The confusion lives in transition words like by, on, and before.
Spell them on first use, say the name after the letters, and add the role: Ms. Duarte, the finance lead. After two or three sends the engine usually learns it. If it does not, keep a spoken list of the five names you use most and spell them by habit.

Use the check on your next dictated client email.
Use it across macOS, Windows, iOS, and Android