Short answer: Dictate job descriptions by speaking each section as if explaining the role to a friend, then editing. Speak the prose parts (summary, responsibilities, success criteria) and type the exact data (salary bands, req numbers, locations). Record the hiring manager intake conversation and use it as your first draft.
Here is a test that almost everyone fails. Open the last three job descriptions your team posted. Read the opening paragraph of each. Then ask yourself whether a candidate could tell those three roles apart from the first hundred words alone.
Usually they cannot, because those hundred words were not written for that role. They were inherited. Someone opened the most recent similar posting, changed the title, swapped a few bullets, and shipped it. The rest of the document is sediment: language that has been passed down through so many copy-paste generations that nobody remembers who wrote it or whether it was ever true.
This is not laziness, and it is not a failure of care. It is an input cost problem, and once you see it that way the fix becomes obvious.
Why job descriptions get copy-pasted
Ask any recruiter or hiring manager to describe a role out loud and they will do it fluently in about ninety seconds. What the person actually does all day. Who they work with. The thing that makes the job harder than it sounds. The kind of person who burns out in it. The thing that would make someone great at it in a way the title does not capture.
All of that exists, fully formed, in their head. It comes out effortlessly in conversation. Then you ask them to write it down and it takes forty-five minutes, and what comes out is "The ideal candidate is a self-starter with excellent communication skills."
The gap is mechanical. Most adults type around 40 words per minute, and skilled typists reach 80 to 100. Speech runs at 130 to 150 words per minute without any practice at all. But raw speed is only half of it. The bigger cost is that typing interrupts thinking. You have a clear picture of the role, you start typing the first sentence, and by the time your fingers catch up the picture has gone hazy and you have started editing a sentence you have not finished. Three rounds of that and reaching for the old posting feels like relief.
Dictation removes the interruption. You describe the role the way you would describe it in a hallway conversation, at the speed you would actually say it, and you get a transcript that has your real thinking in it. The draft is messy. It is also specific, which is the one thing the copy-pasted version can never be.
What a job description is actually made of
Before dictating anything, it helps to notice that a job description is not one kind of writing. It is six, and they have different rules:
- The opening summary. Prose. Two or three sentences on what the role exists to do.
- Responsibilities. Prose, in list form. What the person actually does in a normal week.
- Requirements. Prose plus judgment. The part most likely to be wrong.
- Success criteria. Prose. What "going well" looks like at ninety days and at a year.
- Team and context. Prose. Who they report to, who they work with, where the team sits.
- Logistics and compensation. Data. Salary band, location, employment type, req number, closing date.
Five of those six are prose. Prose is exactly what dictation is good at. The sixth is data, and data is exactly what dictation is worst at, for a reason worth stating clearly.
Speak the prose, type the numbers
This is the single most important rule in this post, and it applies to every kind of dictated document.
Prose defends itself. If a sentence comes back garbled, you see it immediately, because garbled English looks garbled. A misheard word in a responsibilities bullet is obvious on the first read.
Data does not defend itself. If you say a salary band of one hundred and thirty thousand and the text reads one hundred and thirteen thousand, that number looks exactly as correct as the right one. There is no grammatical signal, no red squiggle, nothing your eye catches on a skim. It is a plausible number in a field that expects a number. It gets published, and you find out when a candidate accepts an offer at a figure you never approved.
So the rule is: dictate every sentence, type every value. Salary ranges, req numbers, start dates, headcount, the number of direct reports, application deadlines, office addresses. Those take four seconds each to type and they are the four seconds that protect the document.
A misheard sentence looks wrong. A misheard number looks correct. That difference is the whole rule.
The intake conversation is your first draft
The most useful change is not how you write the job description. It is when.
Most job descriptions are written after the intake meeting, from notes, at a desk, hours or days later. Which means the document is a reconstruction of a conversation, filtered through whatever the recruiter managed to scribble while also asking questions and reading the room. The specificity that made the conversation useful is exactly what gets lost.
Voice Keyboard Pro's Meeting Mode captures that conversation directly, with speaker detection so you can tell the hiring manager's answers from your questions, and AI notes that pull out the structure afterward. Run it during the intake call and you have the raw material with none of the reconstruction loss. The manager's own words about what makes this hire hard, the aside about the last person who did the job, the thing they said twice without noticing, which is almost always the actual requirement.
Two boundaries matter here. First, say out loud that you are capturing notes before you start. That is a courtesy, and in some jurisdictions and some companies it is more than a courtesy, so follow whatever your organization requires. Second, the capture is raw material, not a deliverable. Nobody should ever see the transcript. It is the thing you write the job description from.
The same approach helps for the other conversations around a role. A debrief with the outgoing employee. A conversation with the team member who will work most closely with this hire. Those produce better requirements than any template, and they cost nothing extra to capture.
Dictating each section
The opening summary
Do not write this first. Write it last, after you have dictated everything else, because the summary is a compression of the document and you cannot compress something you have not written yet.
When you do get to it, use a prompt that forces specificity. Speak the answer to: "Why does this job exist right now?" Not what the person does, but why the company decided to spend money on this seat this quarter. Roles get created for reasons. The reason is interesting to candidates and it is almost never in the posting.
Responsibilities
The standard failure here is a list of verbs. "Manage stakeholder relationships. Drive cross-functional initiatives. Own the roadmap." Those are not responsibilities, they are the shape of responsibilities with the content removed.
Dictate this section by narrating a week. Speak it as: "On Monday they would probably be doing this. Most of the middle of the week goes to that. Once a month there is this thing that nobody expects and it takes two days." Narrating a week is easy to say and impossible to fake, and the resulting bullets will be specific enough that a candidate can picture themselves in the chair.
One dictation per bullet works better than one long stream. Speak the bullet, release the hotkey, read it, move on. It keeps each item at a sensible length and stops you from producing one enormous run-on paragraph you then have to break apart by hand.
Requirements
This is the section that quietly does the most damage, because inflated requirements shrink your applicant pool before you ever see it.
Dictation helps here in an unexpected way. Speak each requirement as a full sentence answering one question: "Would I reject an otherwise excellent candidate for not having this?" Say the answer out loud. You will find yourself saying things like "No, honestly, we would train that" or "Yes, this one is real, they will be in the tool on day one." Speaking it makes the judgment audible in a way that ticking bullets off a template never does.
Then split the list into must-have and nice-to-have based on what you actually said, and cut anything you could not defend in a sentence. This exercise takes about four minutes when you speak it. It takes long enough when typing that most people skip it.
What success looks like
The most-skipped section and the highest-value one. Dictate the answer to "What will this person have done in ninety days for us to know this hire worked?" Then the same for a year.
This costs three sentences of speech and it does two jobs at once. It gives candidates something concrete to evaluate themselves against, and it forces the hiring manager to have an answer before the role opens rather than during the first performance conversation. If you also run structured reviews, the same clarity pays off later, which is the connection to dictating performance reviews from the same underlying language.
Team and context
Short, easy, and almost always missing. Speak who the person reports to, who else is on the team, and what the team is responsible for. Candidates read this section more closely than recruiters expect, because it is the only signal about who they will spend their days with.
The edit pass is mostly deleting
A dictated draft is longer than a typed one. This surprises people the first time and then stops mattering, because cutting is the fastest kind of editing there is. You read down the transcript and delete the repetition, the false starts, and the three different attempts at the same sentence, keeping the best one.
Compare that to the typed alternative, where you produce four hundred careful words and then have to generate more from nothing. Generating is slow. Cutting is fast. Dictation converts a generation problem into a cutting problem, which is the real reason it saves time. The raw speed difference is only the visible part.
A practical order for the edit: cut repetition first, then check every number against the source, then read the whole thing out loud once. That last step catches the sentences that read fine but say nothing.
Build a vocabulary for your company's proper nouns
Every company has words that no general transcription engine has ever encountered. Internal team names. Product codenames. The in-house title ladder. The tools in your stack. The acronym that means one thing at your company and something else everywhere else.
Smart Vocabulary in Voice Keyboard Pro handles this. It is a personal dictionary with replacement rules, so a term you say constantly gets written the way your company writes it, every time, without you fixing it by hand.
The important part is how to build it. Do not sit down and brainstorm a list in advance, because you will guess wrong about which words cause problems. Instead, dictate normally for two days and keep a note of every correction you make twice. Add those. Twenty to thirty entries usually covers an entire recruiting function, and it is the difference between dictation that feels smooth and dictation that feels like fighting.
There is a compounding reason to do this that goes beyond convenience. Job descriptions get searched, filtered, and reused. If your internal team name is spelled three different ways across postings, every future search for it finds a third of the results. Consistency in proper nouns is not cosmetic, it is what makes the archive usable later.
Everything else in the hiring workflow
The job description is the visible artifact, but it is a small fraction of the writing a role generates. Most of the hours go elsewhere, and all of it is prose:
- The sourcing note. The outreach message to a passive candidate, which works better when it is specific and specific costs typing time. Dictating it makes personalization cheap enough to actually do.
- The recruiter brief. What you tell an agency or an internal sourcer about the role. Usually a rushed paragraph. It deserves five spoken minutes.
- Interview scorecards. The place where "good communicator, seemed fine" gets typed because the real assessment takes four sentences. Dictate the scorecard within ten minutes of the interview ending, while the impression is still specific.
- Debrief notes. The reasoning behind a decision, which matters enormously three weeks later when someone asks why you passed on the second candidate.
- Rejection messages. The thing everyone agrees should be more human and almost nobody has time to make more human.
These are not glamorous documents, but they are where the week goes. This is the same pattern that shows up in voice dictation for HR work generally: the headline artifact is small and the surrounding correspondence is enormous. If you are on the other side of the table, the same logic applies to dictating cover letters and resumes, where the personalization that gets you a response is exactly the part people skip because typing it is slow.
Traps in ATS and document editors
A few practical things will bite you on the first attempt.
Selection is not insertion. Highlighting a field or clicking a card does not always give you a text cursor. Dictated text goes wherever the cursor is, so if there is no blinking caret, click into the field first. This is the single most common first-run confusion, and it is not a bug in anything.
Return may submit. In many applicant tracking systems, pressing Return in a single-line field saves the field or moves you to the next one. Dictate into the field, then look before you press anything.
Auto-numbering doubles up. If the editor turns a line into a numbered list automatically and you also say the number, you get "1. 1." Let the editor do the numbering and dictate only the text of the item.
One field, one dictation. Resist the temptation to dictate a whole form in a single pass. Fields lose focus, tab order surprises you, and a long stream that lands in the wrong box has to be redone entirely.
Hotkey collisions. Some tools claim single-key shortcuts. If your dictation hotkey conflicts with something the app wants, move it to one of the F13 through F19 keys, which almost nothing else uses.
Doing it from a phone
Intake conversations happen in hallways and after meetings, and the useful detail arrives at the worst possible moment. The Voice Keyboard Pro iPhone keyboard has a mic button built in, so you can dictate into any app on the phone: a note, an email to yourself, the ATS mobile view, a message to the hiring manager confirming what you just agreed.
Voice Edit is genuinely useful here, because fixing text on a touchscreen is miserable. Instead of tapping at a cursor, you speak the change you want and the text updates. And if you are hiring across regions, the two-way translation while dictating covers 24 languages, which is helpful for a first-pass draft. Anything that will be published or that carries legal weight needs a fluent human reader before it goes out. Treat a translated draft as a draft.
What this does not do
Being straight about the limits saves you from expecting the wrong thing.
It does not write the job description. It removes the friction between knowing the role and having it on the page. If you do not know what the role is, dictating produces a fast, fluent document about nothing.
It does not check your language for legal compliance or bias. Job postings carry real regulatory obligations that vary by country, state, and city, including pay transparency rules and prohibited criteria. Those are a question for your legal or compliance function, and no dictation tool substitutes for that review.
It does not drive your ATS. There are no spoken commands to navigate fields or submit a req. It types text where your cursor is, deliberately.
It needs a connection. Transcription happens in the cloud, so an offline flight is not the time to plan on drafting.
It is audible. If you sit in an open office and the role includes a compensation band, take it to a room. This is worth saying plainly: the privacy consideration that matters most day to day is the person sitting near you, not the network. On the network side, our servers store only operational pings, with no audio and no transcript content retained.
A realistic first week
Do not try to change your whole process at once. A sequence that works:
- Day one. Take one open req. Dictate only the responsibilities section by narrating a typical week. Compare it to what is currently posted.
- Day two. Dictate the requirements list by speaking the would-I-reject question for each line. Cut what you could not defend.
- Day three. Run Meeting Mode on your next intake call. Write the description from the capture the same day, while it is still fresh.
- Day four. Start the vocabulary list. Add every term you have corrected twice so far.
- Day five. Dictate the surrounding work: two sourcing notes, one scorecard, one rejection message. This is where the hours actually are.
By the end of the week you will have one posting that reads like a specific job at a specific company, and a working sense of which parts of your process were slow because the work was hard and which parts were slow because typing is slow.
Frequently asked questions
Will a dictated job description sound too casual?
The first draft will, and that is the point. Spoken language is more direct and more concrete than the register people default to when typing formal documents. Keep the specificity in the edit pass and tighten the register. What you should not do is edit it back into template language, because that was the problem you started with.
Can I dictate the whole posting in one go?
You can, but section by section works better. Each section answers a different question, and stopping between them gives you a natural moment to check what landed. A single long stream also tends to blur the boundaries between responsibilities and requirements, which are the two sections that most need to stay distinct.
What about salary ranges and req numbers?
Type them. Every time. This is the one rule to keep even when you are in a hurry, because those are the errors that reach a candidate looking perfectly correct.
Does it work with our applicant tracking system?
Voice Keyboard Pro types at the cursor system-wide on Mac, so it works in any text field in any application, including browser-based tools. There is no integration to configure and nothing to install into the ATS itself, which is usually what makes it viable inside locked-down IT environments.
Is the intake recording stored somewhere?
Your transcription history stays on your own machine. Our servers store only operational pings, not audio and not transcript content. Separately from the technology question, tell participants you are capturing notes and follow your organization's policy on it.
Try it on the req you have been avoiding
Everyone has one. The role that has been open for three weeks because writing the description properly requires an hour nobody has, so it sits behind a posting that was cloned from something adjacent.
Open a blank document and describe that job out loud, the way you would explain it to a friend who asked what you were hiring for. Do not organize it. Do not try to make it good. Just say what the job is for four minutes.
Then read the transcript. It will be messy, repetitive, and more honest about the role than anything you have posted this year. Cutting it into shape takes fifteen minutes. Voice Keyboard Pro has a free tier with daily limits, and Pro is $4.99 a month or $34.99 a year if the first req convinces you.