There are four realistic moments to disclose and they buy different things. Disclosing in the application is the only one that can change the format, because recording limits, countdown timers and slot lengths are set when your invite is generated. Disclosing in the first thirty seconds of a live call is the only one that reliably stops an interviewer finishing your words. Disclosing after a block repairs one moment but arrives when speaking is hardest, and never disclosing pushes the whole load onto word substitution, which is exactly where a technical round leaks marks. The wording for each is below.
This is wording and format help, not speech therapy and not legal advice. Stuttering is generally treated as a disability under the ADA in the United States and the Equality Act 2010 in the UK, which is what makes a format request a request rather than a favour, but what applies to you depends on your country, the employer and the role. One thing worth knowing before you tick anything: the voluntary self-identification disability question on many US applications exists for equal-opportunity reporting and generally does not reach the interview panel. Ticking it is not an accommodation request. A format change has to go to the recruiting coordinator, or a named accommodations contact, as its own message.
Four disclosure timings, side by side
Column three is generalised from how hiring processes are structured, not from any measured figure. Read it as where the leverage is, not as a probability.
| When | Say exactly this | What it typically gets you | What it costs |
|---|---|---|---|
| In the application or the scheduling email | "One note on format. I have a stutter. It has no effect on the work. In an interview the only thing that helps is not having my words finished for me. If any stage is a recorded video with a countdown, could I have extra recording time, unlimited retakes, or a live call instead?" | The only timing that can change the mechanics. Timers, recording windows and slot lengths are configured when the invite is generated, so a request arriving after the link does is often refused for reasons that are genuinely technical rather than hostile. It also puts the ask on record before any decision is made. | You cannot control who reads it, or how early in the funnel. If you are unwilling to say it before a human has spoken to you, this row is simply not available to you, and rows two and three still work. |
| In the first thirty seconds of a live call | "Before we start, one thing. I stutter. It doesn't change the answers, but if I stop in the middle of a word, just let me finish it rather than filling it in." | The highest value per second on this page. It stops word completion, and it stops a pause being read as not knowing the answer, for the whole hour, at a cost of about ten seconds of speech. Interviewers who have been told almost always comply, because most word completion is someone trying to be helpful. | You are speaking your hardest sentence cold, first, before you have warmed into the room. Rehearse this one line aloud more than you rehearse any answer. It buys nothing on format, because the clock is already running. |
| Only if a block happens | Mid-block or straight after it: "I stutter, one second." Then finish the same sentence. No apology, no explanation, no restart. | Keeps the top of the call neutral and spends the disclosure only when it is needed. Sensible for a twenty-minute recruiter screen, or a round with one interviewer you have already built some rapport with. | By the time you say it, four to eight seconds have gone and the person opposite has formed a theory: nerves, a dropped connection, or that you do not know the answer. On a phone line they will often have said "hello, are you still there?" over the top of you. You are also producing an unplanned sentence at the exact moment production is hardest. |
| Never | Nothing about the stutter, but pre-decide the neutral stall you will use instead, and pick one that starts easily for you: "Let me put that in order." or "Two parts to that." | Total privacy, no framing, no chance of being remembered as the candidate who stutters rather than the candidate who answered well. Perfectly reasonable for a short screen, a chat-first round, or an employer you have already decided against. | Every pause gets a default interpretation you did not choose. The larger cost is invisible: the work moves onto substitution, and in a technical round substitution is what quietly costs you the term the rubric is listening for. |
The accommodations that actually get granted
Send these to the coordinator who owns the calendar, at the moment you are being scheduled, not to the hiring manager. Use the word adjustment or accommodation once so it routes correctly, then go straight to logistics. Two asks per email, never five.
- No word completion, no interruption. "If I stop mid-word, please let me finish rather than filling it in." Not a process change, so nobody has to approve it and it cannot be refused. Say it aloud even if you disclosed in writing, because written requests routinely fail to reach the panel.
- A live call instead of a one-way recorded video, or extra recording time. The one that must be asked early. A recorded screen with a ninety-second cap and two retakes penalises a block directly and mechanically. "Could this stage be a live call, or could I have extended recording time on it?" Ask before the invite is issued, not after.
- Questions typed into the chat as they are asked. Cheap to grant on a video call. It means you can start choosing the words while they are still talking, instead of holding the question and composing the sentence at the same time.
- Permission to type when a block runs long. "If I get stuck on a word I may type it rather than lose the thread." Useful everywhere, essential in coding rounds: a term you cannot say, you can always spell into the editor or the chat, and the technical point survives intact.
- A longer slot with the same content. Ask for forty-five minutes of questions in a sixty-minute booking. Do not ask for "more time", ask for the same number of questions in a longer window. That is a scheduling change rather than a fairness question, and it gets a faster yes.
- Not narrating continuously in a live coding round. "Could I work quietly for a few minutes and then explain in chunks, rather than talking the whole way through?" Continuous think-aloud is the most demanding speech task in a technical loop, and most interviewers will take structured explanations at intervals instead.
- Video instead of a phone screen. Silence on a phone line reads as a disconnect, which is why phone screens produce the most talking-over. On video a pause looks like thinking.
- The attendee list in advance. Names are the one category you cannot substitute. Knowing who is in the room, and that you will be asked to say your own name, your last employer and possibly your email address out loud, lets you plan the introduction rather than meet it cold.
Re-confirm the day before. The commonest failure is not refusal, it is a granted adjustment that never reached the interviewer.
Why swapping the word costs you the mark
Substitution is invisible to the interviewer and expensive to you, because a technical rubric is often listening for a specific term rather than for the idea behind it. The answer is right, the word is missing, and the note that gets written down is about vocabulary.
| What you meant | What comes out instead | What that does to the scoring |
|---|---|---|
| Idempotent | "You can run it twice and it's fine" | Correct, and still commonly marked as not naming the property. The follow-up is often "is there a word for that?", which forces the exact word you avoided, now with one person waiting for one specific sound. |
| Amortised O(1) | "Usually fast, occasionally slower" | Complexity is scored on the term. A hedge reads as not having done the analysis rather than as an avoided onset. |
| Quorum, consensus | "When enough of them agree" | Fine in conversation. In a distributed systems round the vocabulary is the signal being tested, so the paraphrase scores as vague. |
| PostgreSQL, Kafka, Redis | "The database", "the queue thing" | Invites "which one?", so you end up saying the proper noun anyway, one level deeper into the conversation and with more attention on it than if you had led with it. |
| Eventual consistency | "It catches up after a bit" | Loses the trade-off you were actually making, which is the part the question existed to probe. |
| Your name, your last employer, a manager's name | A pause, or "my previous company" | No escape route exists. Proper nouns have no synonyms, which is why the introduction is the hardest ninety seconds of the hour, and why it arrives before you have warmed up. |
Where the word search starts, and how a drafted answer removes it
For a lot of people the effort is not the block itself. It is the scanning that runs three or four words ahead of the sentence, checking what is coming, deciding whether to swap it, then re-planning the rest of the sentence around the swap. All of that runs while you are also holding the question, the shape of the answer, the interviewer's face and a clock. The substitution decision gets made under that load, in about a second, which is why it so reliably trades the precise term for the safe one.
A sentence that is already drafted removes the composing half. You are not choosing words any more, you are following a route that already exists, and the decision that costs you the technical term never has to be made, because the term is written down and you saw it coming. That is the honest claim: it removes the search, not the stutter.
Most of this you can get with no software at all. Write out the twenty sentences you know you will need, the introduction, the two-line why-this-company answer, the three project summaries, and read them aloud beforehand so the terms are chosen in advance rather than at speed. The company Q&A generator exists to make that list finite: it builds a question bank for a named company and role, so you draft against real questions instead of guessing. In the room, Interview Copilot is the live version of the same idea, transcribing the interviewer on device and putting a speakable draft on your own monitor in about a second, with metering that counts listening time only, so a long pause on your side costs nothing on the clock. In a coding round Code Copilot returns the intuition, a dry run and the complexity in writing, which puts the complexity term on screen instead of making you retrieve it out loud while a timer runs.
Where this is the wrong tool
We publish this and we sell the software, so here is the boundary, plainly.
- It does not make you fluent, and it is not therapy. Nothing here is a technique, a target or a treatment. Speech and language therapy is a separate profession and this page is not it.
- Reading aloud is not automatically easier. For some people a fixed text is harder than spontaneous speech, precisely because the escape route is gone and there is now one specific word you are committed to. That is exactly why the free plan matters: five minutes a day, no card, enough to find out on a mock call whether a drafted sentence steadies you or corners you, before a real loop depends on the answer. Pricing only becomes a question after that.
- A one-way recorded video is the wrong place for it. The overlay is excluded from screen capture at the operating system level, so it does not appear in a Zoom, Teams, Meet or OBS share, or in a screenshot. That is a privacy property and nothing more. It does not defeat a webcam, a lockdown browser, eye tracking, a proctor watching you in person or a second-device camera, and reading off screen is visible as eye movement to anyone reviewing a recording. For a recorded stage, the right route is row one of the table: extra recording time or a live call.
- It is not a substitute for the ask. If being interrupted is your actual problem, the fix is the sentence in the first thirty seconds, not software. Solving it privately means going through the whole process without the change you were entitled to request, and the next round starts identically.
- Windows 10 and 11 only. No macOS, no Linux, no mobile, no browser extension. On a phone screen taken away from a desk, none of it applies.
If you take one thing from this page, take the thirty-second sentence. Ten seconds of speech changes how sixty minutes of pauses get interpreted, it needs nobody's approval, and it costs nothing.
Related guides
- Using AI in an Interview With a Disability: What to Disclose
- ADHD Interview Accommodations: 9 Asks and the Exact Wording
- Deaf Interview Accommodations: CART vs Captions vs Interpreter
- Dyslexia Coding Interviews: Cutting the Reading Bottleneck
- Job Interview Tips for Autistic Candidates: STAR Over Small Talk