What are the best job interview tips for autistic candidates?

Replace the parts that rely on reading a face with parts you control. Pre-write ten small-talk lines and an exit question for each. Cap every behavioural answer at four beats, Situation, Task, Action, Result, and keep the skeleton visible on your own monitor so you know you are finished when you have said the Result, not when the interviewer looks satisfied. Set eye contact as a cadence rather than a duration.

How do you handle small talk at the start of a job interview if you are autistic?

Prepare one line for each of the ten or so openers that actually occur, then end your turn with a question. A question is an unambiguous handoff, so the interviewer takes the floor and starts the real interview. Two exchanges is the correct length. Useful exits include "Where would you like to start?" and "Do you want my background first, or straight into questions?"

How much eye contact is expected in a job interview?

Less than most people assume, and in a rhythm rather than a hold. Roughly three seconds of contact, then look away while you build the next sentence, then return at the full stop. Looking away while thinking is normal and reads as thought. An unbroken stare is more conspicuous than breaking gaze. On video, aim at the interviewer's face on screen and move that window up under the webcam.

How do you know when to stop talking in an interview answer?

Use a positional signal instead of a facial one. Write the four STAR beats where you can see them, and treat the Result line as the end of the answer. When you have said the number and what changed because of it, stop, then add an explicit handoff: "That is the example, happy to go deeper on any part of it." That sentence replaces the nod you would otherwise be waiting for.

Why do my interview answers run four minutes long?

Usually because of accuracy habits, not waffle. The commonest additions are checking your example is representative, giving dates and chronology, defining terms, presenting the other person's case fairly, and stating your own confidence level and faults. Each one is correct in a design document and expensive in a 60-second answer. Cutting them removes no fact an interviewer was scoring.

Do you have to disclose autism to ask for an interview adjustment?

Usually not for a change to the format. Describe the barrier and the change you want: "I process questions more accurately in writing, so could each question be pasted into the chat as it is asked?" is a complete request without a diagnosis. Questions sent in advance, one interviewer instead of a panel, and a structured question list are all common asks. Rules vary by country and employer.

Job Interview Tips for Autistic Candidates: STAR Over Small Talk

Replace every part of the interview that depends on reading a face with a part you decided in advance. Pre-write ten small-talk lines, each ending in a question so the interviewer takes the floor back. Cap each behavioural answer at four beats and keep those beats visible on your own monitor, so the signal for "that is enough now" is you reaching the Result line, not a nod you have to interpret while talking. Eye contact is a cadence, roughly three seconds on and a look away while you build the next sentence, not a duration to endure.

Four minutes cut to sixty, with the edits marked

Here is the answer as people actually give it, to "tell me about a time you disagreed with a teammate". About 420 words, and with the pauses and restarts around it, close to four minutes. Everything ~~struck through~~ comes out, and each cut carries a number.

So, a time I disagreed with a teammate. I should say that I disagree with people fairly regularly, in a constructive way, so choosing the most representative example is slightly difficult, but I will go with the schema one because it is the clearest. [1] This was at my second job. I joined in the March, though not that particular team until roughly the July, and this would have been the following February, or possibly early March. [2] We ran a service that exported customer usage data into the billing system. A schema is the agreed shape of that data, so which fields exist and what type each one is, and a breaking change is one where something that was working stops working. [3] A teammate proposed dropping two fields from the export because nothing consumed them. To be fair to him, he had a reasonable basis for that. He had searched every internal repository and found no references, and in most circumstances that is sufficient evidence, and I would not want to give the impression that he was careless, because he was not. [4] I thought it was wrong, because the finance team ran their own spreadsheets against that export and those would never appear in a repository. Although I was not certain at the time. I would put my confidence at about seventy percent. I did consider that I might be remembering an older version of their process, and I went back and forth for two days over whether it was worth raising at all, because I did not want to be the person who blocks a change on a hunch, and I have had feedback about that before. [5] So I raised it on the pull request. I wrote quite a long comment with numbered points, and the thread got sharper than either of us intended, which is partly mine, because I have since learned that numbered rebuttals read as aggressive to some people even when nothing aggressive is meant. [5] That was the Tuesday. On the Wednesday he replied, and on the Thursday our lead got involved, and I should probably explain how our review process worked at that point, because we had only just moved from... [2]

Four minutes are gone and there is no Result, because the answer never reached one. The five cut types, in the order they tend to appear:

The representativeness check. The question said "a time", so you verify out loud that your example really is the most representative one before starting. Choose one story per theme the night before and never re-choose live.
The dateline. Months, sequence, what happened on which day. Nobody scores chronology. Allow yourself one time anchor for the whole answer: "at my last company".
The definition. Explaining what a schema is to a person who reviews schemas. Assume the interviewer knows their own field, and if they do not, they will ask.
The fair hearing. Giving the other person's case in full, because leaving it out would be unfair to them. It would be, slightly, and it also spends your Action time on someone who is not being interviewed. One clause discharges the obligation: "he had a reasonable basis".
The confidence disclosure. Stating your certainty level, your deliberation and your own faults, because a flat claim would feel like an overclaim. Say the flat version anyway. "I thought" already contains the hedge.

Those five account for roughly 300 of the 420 words. None of them is a character flaw. Every one is precision behaviour that is correct in a design document, a bug report or a code review, and expensive inside a 60-second answer. You are not cutting honesty, you are cutting the parts that were never being scored. Here is what survives.

Situation, 0:00. On a billing export at my last company, a teammate proposed dropping two fields, because he had searched the codebase and nothing referenced them. Task, 0:08. I thought finance read those fields from spreadsheets outside the repo, so I had to prove that or drop it, before the change shipped that week. Action, 0:16. Instead of arguing in the review thread, I added temporary logging on both fields. Within a day it showed reads from two finance service accounts. I took that to him on a ten-minute call rather than in writing, because the thread was getting sharper than either of us wanted. We agreed on a deprecation instead of a deletion: keep the fields, log usage for thirty days, tell the owners, then remove. Result, 0:45. The fields came out five weeks later and nobody's month-end broke. That temporary logging is now the standard check before any field removal, and the two of us default to a call once a review thread passes three replies.

About 165 words, 65 seconds. Every surviving sentence is a decision, a reason or a number. Time it once into your phone's voice recorder, and if the first thing you actually did arrives later than 20 seconds in, cut item 1 or item 2 again.

Ten small-talk openers, each with an exit line

Small talk is short, scripted and almost entirely unscored, which makes it the cheapest problem on this page to solve. The mechanic that matters is the exit: end your turn with a question, because a question is an unambiguous handoff and the interviewer will take it. Two exchanges is the right length. Learn these as lines, not as principles.

What they open with	Your line	Your exit
"How are you today?"	"Good, thanks, and glad the timing worked out."	"Where would you like to start?"
"Can you hear me okay?"	"Yes, you are clear. Can you hear me?"	"Great. Shall I start?"
"Did you find us alright?"	"Easily, thanks. I got here early and sat in the cafe opposite."	"Is it just the two of us today?"
"Awful weather today."	"It is. I gave up on the umbrella halfway."	"How are you planning to use the hour?"
"How has your week been?"	"Busy, mostly good. Yours?"	"Do you want my background first, or straight into questions?"
"Where are you based?"	"I am in Leeds, about twenty minutes from the office."	"Which team does this role sit in?"
"Have you been doing a lot of interviews?"	"A few. This is the one I have read most about."	"How long have you been on the team?"
Silence while they load their notes	"Take your time."	"Would a two-minute summary of my background be useful?"
The chat is still going after two exchanges	"That is interesting."	"I do not want to eat into your time. Shall we start?"
"So what made you apply?"	Not small talk. This is question one, answered in 45 seconds: one thing about the product, one about the role.	None needed. The interview has started.

Two notes on that last row. First, recruiters routinely deliver a scored question in a chatty tone, and the tone is not the signal. When you cannot tell which mode you are in, give the short structured answer: structure in response to small talk costs you nothing, while chat in place of an answer costs a line on a scorecard. Second, when the interview opens straight onto a real question, the small talk is simply over and does not need to be paid back later. Nobody is waiting for it.

Eye contact and pause settings you can actually dial in

Cadence, not duration. Roughly three seconds of contact, then look away while you construct the next sentence, then return at the full stop. Breaking gaze while thinking is universal and reads as thought. Compensating with an unbroken stare is more conspicuous than looking away.
Two deliberate hits per answer. The first sentence and the Result number. Those two land, and the rest of the gaze budget is yours to spend anywhere.
On video, the face is not the lens. Drag the interviewer's tile to the top of your screen directly under the webcam. Then look at their face rather than the lens: the offset is an inch or two and reads as ordinary eye contact. Anything you pin beside the webcam sits inside that same cone.
In person, use the bridge of the nose. At about a metre nobody can distinguish it from eye contact, and it costs far less if direct gaze is painful.
On a panel, answer the asker. One glance at each other person during the Action beat, then back to the asker for the Result. That is the whole rule.
Thinking silence: two to four seconds, narrated. "Give me a moment to pick the right example" turns silence into a stated action, and stated silence never reads as being stuck.
Do not fill the gap at the end. Video adds a few hundred milliseconds in each direction, so the pause after your last sentence runs longer than it feels and both people often start at once. Finish, then count three silently. The sentence people add during that gap is what turns a 65-second answer into a three-minute one.
Signal the end explicitly. "That is the example, happy to go deeper on any part of it." That sentence does the job the nod-and-eyebrow exchange would otherwise do, and it does it in words.
An interruption is information. If they cut in mid-answer, they have what they need. Stop mid-sentence and take the new question. It is not rudeness and it costs you nothing.
Keep one recovery line. If you lose the thread: "I have gone sideways, the short version is..." then jump straight to the Result. Interviewers hear that as self-awareness.

Put the STAR skeleton on your monitor, not in their face

The usual "is that enough yet" signal is facial: a nod that slows, a pen going down, a shift back in the chair. Reading that while composing speech is expensive, and misreading it costs you in both directions, stopping at 15 seconds or running to four minutes. So swap the social signal for a positional one that lives on your screen.

Four lines, nothing else

S: the setup in one sentence. T: what was specifically mine to fix. A: three decisions, in order. R: the number and what changed because of it. Full sentences are a trap, because you will read them aloud and it will sound read.

Two hard rules

You may not start R until three A items have been said. You are finished the moment R has been said. Finished is a position on a list, not an expression on a face.

Track it with a finger

Physically move a finger down the four lines as you speak. It gives you somewhere to look that is not a face, and it stops you restarting the Situation halfway through the Action.

Keep it inside the gaze cone

Directly beneath the webcam, so a glance at it looks like a normal break in eye contact rather than a look off to one side.

An index card taped under the webcam does all of that, on any operating system, in person included. The software version exists because a card cannot follow a question you did not prepare for. On Windows 10 and 11, the OfferPilot AI Interview Copilot listens to the interviewer's audio through system playback, so headphones are fine, transcribes on-device with Whisper, and drafts a skeleton grounded in your own CV in about a second. The overlay is excluded from screen capture at the operating-system level, so it is not in the Zoom, Teams or Meet share. That exclusion covers screen sharing and screen recording only. It does not defeat a webcam, a lockdown browser, an in-person proctor, eye tracking or a second-device camera, and none of this is about a monitored assessment. Use the four lines, not the sentences.

Whichever version you use, prepare against the questions you will actually be asked. The interview questions bank is a starting point, and the company Q&A generator builds a set for a specific employer and role, which is how you decide which four stories are worth skeletoning at all.

Where a copilot is the wrong tool for this

Worth saying plainly, given a vendor is writing it. A live draft on screen is a second stream of language arriving while you are producing language, and for some people that split is the exact thing that is hard. If it is, this makes the interview worse rather than better. Test it in a mock round before you test it in a real one. Hold P to ghost the overlay or hold M to pause the mic if it starts competing for attention, and if it still competes, run nothing at all.

Three further honest limits. Reading generated wording aloud flattens prosody, so if you already get feedback about sounding flat or rehearsed, the skeleton helps and the sentences hurt. No overlay changes a scorecard: where a company is grading warmth, handshake energy and banter, the fix is a more structured interview, and asking for the questions in advance or in writing is a real request that is granted more often than people expect and costs nothing to make. And it is Windows 10 and 11 only, no macOS, no Linux, no mobile, no browser extension, so on a Mac the index card is the whole product. The free plan is 5 minutes a day with no card, which is enough to find out whether the split attention works for you and not enough to carry a full round.

The through-line is the same in all four sections. Anything that depends on interpreting another person in real time gets replaced with something you decided in advance: a line, a beat, a count, a position on a list. That is not masking. It is removing a task from the interview that the interview was never testing.

Job interview tips for autistic candidates: structure over small talk