# The Live AI Interviewer Run, From Notice Check to a Logged Data Request

*Starting from one AI screen invite, you will decide in writing whether to take it, prepare stories that hold up under follow-ups, and log a dated data request within 48 hours.*

- Canonical URL: https://www.refolk.ai/candidates/guides/live-ai-interviewer-run
- Pillar: Interviewing
- Format: Playbook
- Published: 2026-11-17
- Last reviewed: 2026-11-17
- Reading time: 25 min
- Keywords: how to prepare for an AI interviewer, AI voice interview tips for candidates, can I ask for a human instead of an AI interview, what does an AI phone screen score, AI recruiter call follow-up questions, request my AI interview recording or transcript

## Key takeaways

- In the Jabarian and Henkel field experiment across 48 positions at 43 firms, AI voice interviews covered 6.8 topics against 5.5 for human-led ones and produced 12% more job offers, and human recruiters made every final hiring decision.
- New York City's Local Law 144 requires notice at least 10 business days before an AI screen and must tell you how to request an alternative process, but it does not require the employer to provide one.
- California's ADMT opt-out and Colorado's human-review right apply only to invites dated on or after January 1, 2027, and at least one legal source reports that Colorado's law is on hold with no firm enforcement date.
- AI interviewer follow-ups go after whatever your first answer left out: a specific example, your own role, your method, or the result. Prepare all four for every story, because one platform allows up to 5 follow-ups per topic and no retakes.
- In Refolk's index, 12 of 25 sampled profiles with "AI interviewer" in their headline hold founder titles in both the US and India, and no US or UK recruiter profile uses "AI screening" in its headline, so most public advice about these screens comes from vendors.
- Send your data request in writing on the day of the screen and cite the clock that applies: 30 days for NYC data information or Illinois video deletion, and 45 days for California access from January 1, 2027.

This guide is for a candidate who has been invited to a first-round screen run by an AI voice or video agent, the kind that listens to each answer and asks follow-up questions, and who has a deadline. It takes you from the invite email to a written decision on whether to take the screen or ask for a human. Then you prepare stories that hold up under follow-ups, check your own practice transcript, sit the screen, and send a dated request for your data, all inside 48 hours. Where the law or the evidence about the platforms is thin, I say so and tell you how to check it yourself.

## What a live AI interviewer scores, and who makes the decision

In the setups that are publicly documented, the agent produces a transcript plus scores with reasoning, and a human recruiter makes the hiring decision after reading them. You are producing a transcript for a human to read later. You are not performing for a camera.

That is the main difference from a one-way video interview, where you record answers to fixed prompts on a blank screen. A live agent changes its next question based on what you just said. Third-party candidate guides describe what the employer receives as "the transcript, the scores with their reasoning, and a recording when one is available."

The strongest public evidence on these screens is a field experiment by Jabarian and Henkel, "Voice AI in Firms." It covered 48 positions at 43 client firms. Applicants were randomly assigned to be interviewed by a human recruiter or an AI voice agent, and some summaries say a group was allowed to choose. Compared with applicants interviewed by humans, the AI-interviewed group had:

- 12% more job offers
- 18% more job starts
- 6.8 topics covered per interview, against 5.5 in human-led interviews

Final hiring decisions always rested with human recruiters, who reviewed transcripts, recordings, and standardized test scores.

**What this proves:** the agent collects more relevant ground per interview, and a human reads what it collected. **Where it can mislead you:** a 12% relative gain tells you nothing about your own odds. The summaries I checked do not give absolute offer rates or name the transcript features that predicted offers. If you want those numbers, read the paper itself on SSRN instead of a blog's retelling.

### What the platforms do, according to third-party guides

The table below comes from third-party candidate guides, mostly from a single site, not from the vendors' own help pages. Use it to know what to look for, then check the help link in your own invite.

| Platform | Length | Follow-ups | Retakes |
|---|---|---|---|
| Equip conversational | 15 or 30 min | Up to 5 per topic | None once submitted |
| Flowmingo | About 15 min | When an answer is incomplete | Up to 2 hrs after, unless turned off |
| TestGorilla | 10 to 20 min | Yes | Unverified |
| HireVue | Unverified | Probing | Unverified |

Equip's conversational mode is described as 5 to 10 questions across 3 to 5 topics. TestGorilla's one-way answers are capped at 30 seconds, 1 minute, or 2 minutes. Third-party guides describe both TestGorilla and HireVue as scoring from the transcript. Most AI interviews run 15 to 25 minutes. I could not establish any platform's policy on keeping notes in front of you during the screen.

### Does the agent score your voice?

That depends on the vendor, and the vendors disagree with each other.

- **HireVue** transcribes speech with Rev.ai. It stopped facial analysis in March 2020, announced that in January 2021, and later said it would also drop speech inputs such as "variation in tone or pauses." Its AI Explainability Statement is the document to read.
- **Talview's Ivy** is described as evaluating "communication traits like fluency, tone and relevance."
- **Filler words:** I found no independent evidence that they are scored.

So the honest answer to "what does an AI phone screen score" is: read the explainability page for the vendor named in your invite. Whatever that page says, the content of your transcript is the part you control. The rest of this guide is built around it.

### Who writes the public advice on AI interviews

Most of the public advice about these screens comes from the companies that sell them. In Refolk's index of professional profiles, 314 US profiles and 604 India profiles have "AI interviewer" in their headline. In a sample of 25 from each country, 12 hold founder or co-founder titles.

| Country | "AI interviewer" headline profiles | Founder titles in sample of 25 |
|---|---|---|
| United States | 314 | 12 (48%) |
| India | 604 | 12 (48%) |

**48% - Share of sampled "AI interviewer" headline profiles held by founders or co-founders, in both the US and India**

In the same index, no Recruiter or Talent Acquisition profile in the US or UK uses "AI screening" in its headline.

**What this proves:** the people talking publicly about these products are mostly the people building them. **Its limits:** the sample is small (25 per country) and covers headline keywords only. It does not show how employers use the tools. Treat any vendor blog's claims about tone, eye contact, or "keywords the AI wants" as marketing until the vendor's own explainability page confirms them.

## Which rules give you notice, a human alternative, or your data

For a US candidate, four places have rules that matter: New York City, Illinois, Colorado, and California. For an invite dated before January 1, 2027, only the NYC and Illinois rules apply. Colorado's start date is also contested. A rule applies because of where you live or where the job is, not where the employer is headquartered.

| Rule | Applies when | Alternative or opt-out | Data clock |
|---|---|---|---|
| NYC LL144 | You live in NYC; in force | You may request; employer may refuse | 30 days (data information) |
| IL Video Act | Illinois role, recorded video; since Jan 1, 2020 | Your consent is required | 30 days (deletion on request) |
| IL HB 3773 | Illinois employment; since Jan 1, 2026 | None | None |
| CO SB 26-189 | Colorado-resident applicant; from Jan 1, 2027, contested | Human review of adverse decisions | Access and correction |
| CA ADMT | California consumer; from Jan 1, 2027 | Opt-out or human appeal, with a hiring exception | 45 days (access) |

### New York City, Local Law 144

Candidates who live in the city must get notice at least 10 business days before an automated employment decision tool (AEDT, meaning software that substantially helps decide who advances) is used on them. The notice must say that a tool is used, which qualifications it assesses, and how to request an alternative process or an accommodation. The rules do not require the employer to provide the alternative.

The detail most candidates miss is that a notice posted on the careers site counts. Under the city's rules (6 RCNY 5-304), the employer may use the tool on candidates 10 business days after that page went up, regardless of when a particular job was posted. A 48-hour invite can therefore be lawful. Your check is whether the careers page carried the notice in time, not whether the invite email gave you 10 days. If the data details are not on the website, the employer must give you the type of data collected, its source, and its retention policy within 30 days of a written request.

### Illinois: two separate laws

- **AI Video Interview Act (820 ILCS 42).** In force since January 1, 2020, for Illinois positions. It covers AI analysis of recorded video interviews and requires notice, an explanation of how the AI works, your consent, and deletion within 30 days of your request. Whether it covers a voice-only agent has not been publicly established.
- **HB 3773.** Signed August 9, 2024. Since January 1, 2026 it has made discriminatory or undisclosed AI use in employment a civil rights violation. The Illinois Department of Human Rights withdrew its draft notice rules, but the statute still stands. It gives no opt-out and no deletion right.

### Colorado and California, from January 1, 2027

- **Colorado SB 26-189.** Signed May 14, 2026, with an effective date of January 1, 2027. Its definition of "consumer" includes Colorado-resident job applicants. It requires advance notice and disclosures after a decision is made. It keeps your rights to access and correct your data and to get meaningful human review of an adverse automated decision. Employers must keep records for at least three years. At least one legal publisher reports the law is on hold with no firm enforcement date because of litigation, so re-check its status before you cite it.
- **California CCPA ADMT rules.** ADMT means automated decision-making technology. Its duties (pre-use notice, opt-out, and access) apply from January 1, 2027. Opt-outs must be handled within 15 business days, and access requests within 45 days. There is a hiring exception: no opt-out is owed if the tool is used solely to assess ability to do the job and does not discriminate unlawfully. ADMT is also defined as technology that "replaces or substantially replaces human decision-making." In a setup like the Jabarian and Henkel one, where humans make the decision, the opt-out may not apply at all.

### Everywhere in the US: disability accommodation

The Americans with Disabilities Act (ADA) applies to AI screens like any other part of hiring. The EEOC removed its AI guidance documents from its website on January 27, 2025, but the ADA itself did not change. The archived 2022 guidance says you do not need to mention the ADA or use the phrase "reasonable accommodation," and that employers need to respond promptly. No law sets a specific response deadline.

> **Rule:** Quote the rule, or do not cite it
>
> Every right you rely on goes into your decision note as the exact rule name, what it says, and its effective date, next to the date of your invite. If you cannot quote it, leave it out of your request.

## Take it or ask for a human: making the written decision

Take the screen unless the format will distort your answers and you have a reason to ask for something else. If you do ask, ask in writing, before the screen, and keep the sent copy with full headers. In most cases the request is a well-documented ask, not an enforceable demand, so word it to keep the process moving.

Two things decide it: how much the format distorts what you can show, and how much a written rule backs your request. Distortion includes a speech or hearing disability, speech recognition that struggles with your accent, or a connection or setting you cannot fix before the deadline.

#### Take the screen or ask for an alternative

Horizontal axis runs from Format barely distorts your answers to Format distorts your answers. Vertical axis runs from No rule backs a request to A rule names a request route.

| Quadrant | What it means |
| --- | --- |
| Rule exists, low distortion | Take the screen and record the notice in your decision note for later. |
| Rule exists, high distortion | Request the alternative in writing before the screen, quoting the rule. |
| No rule, low distortion | Take the screen; no request needed. |
| No rule, high distortion | Ask anyway in writing; if it is disability-related, frame it as an accommodation. |

*The format's cost to you decides whether to ask; the rule behind you decides how to word the request, not whether you are allowed to make it.*

**On timing, the sources disagree.** The archived EEOC guidance says to ask before the assessment. A disability advocacy source says a request after you are screened out is still valid, and practitioner advice is to "ask the employer in writing for an alternative test format before or right after you're screened out, and keep a copy of the request." Ask before. It is cheaper for everyone, and it does not read as a reaction to a bad score.

**How to ask without stalling the process:** name the specific alternative you want, give a reply-by date, and say you will complete the AI screen by the original deadline if the alternative is not possible. That keeps you on schedule whatever they decide.

**Alternative format or accommodation request**

```
Subject: Request for an alternative interview format, [Role title], [Requisition ID if shown]

Hi [Recruiter name],

Thank you for the invitation to the first-round screen for [Role title], which I understand is an AI [voice / video] interview through [vendor] due by [date and time].

I would like to request an alternative format for this round: a [15 to 30] minute call or video interview with a member of the team. [Optional, one sentence on why: "The format relies on speech recognition, and a live conversation lets me answer accurately."]

[NYC residents only: As a New York City resident, I understand the Local Law 144 notice explains how candidates may request an alternative selection process or accommodation.]

So the process does not stall, I will keep the AI screen open and complete it by the original deadline if an alternative is not possible. Could you let me know either way by [date]?

Thank you,
[Your name]
[Phone]
```

*Send before the screen. Delete the rule line if no rule applies to you, and you do not need to name the ADA for a disability request.*

**What a reply proves, and when it misleads you:** a named person writing back with a yes or a no is an answer. An automated "we received your message" is not an answer, and it does not pause your deadline. If all you get is the auto-reply, take the screen on time and keep both emails.

## The 48-hour run, step by step

The whole run takes about six and a half hours of work, and it fits inside the 48-hour window that some invites set (Flowmingo's sample invites suggest finishing within 48 hours). Do the first three steps on the day the invite arrives, the preparation on day one or two, and the screen and data request last.

#### From invite to a logged data request

1. **Triage the invite (20 min)** - Write down the vendor, the deadline, the format (voice, video, or chat), the job location, and where you live. Done when all five fit on one dated line.
2. **Run the jurisdiction check (30 min)** - Check the job location and your residence against NYC, Illinois, Colorado, and California, counting the last two only for invites dated January 1, 2027 or later, and look for an LL144 notice on the careers page. Done when you have a list of the rights that apply, with dates.
3. **Write the take-or-ask decision (20 min)** - Accept, or email a request for an alternative format or an accommodation and keep the sent copy with headers. Done when there is a dated email or a dated note to yourself.
4. **Build the competency map (2 hrs)** - Pull each qualification from the posting and attach one story and one metric to it. Done when every qualification has a filled row.
5. **Add the backup layer (1.5 hrs)** - For each story, write the specific example, your own role, your method, and the result, because those are what follow-ups probe for. Done when every story has four detail lines.
6. **Run the recorded transcript check (1 hr)** - Record your answers, transcribe them, time each one against your chosen ceiling, and confirm every claim appears on your resume. Done when no answer runs over the ceiling and no claim is missing from your CV.
7. **Sit the screen (15 to 45 min)** - Lead each answer with the example and answer follow-ups from your backup layer. Done when the completion confirmation is saved.
8. **Log your data request (15 min)** - Email a written request for your transcript, recording, and data that cites the clock that applies to you. Done when there is a dated sent email and a calendar reminder for the day the clock runs out.

If the deadline is shorter than 48 hours, keep steps 1 to 3 and 8 at full length and shorten the preparation. A request sent late is weaker evidence than a story that is a little less polished.

## Building the competency map and the backup layer

A competency map pairs each qualification in the posting with one story and one number. The backup layer gives every story four prepared details, because those four are what the agent's follow-ups go looking for. Together they are the most important preparation for an AI interviewer.

Third-party guides say follow-ups "tend to ask for what your answer left out: a specific example, your own role, your method or the result." Equip's conversational mode allows up to 5 follow-ups per topic and no retakes. A vague first answer is not simply marked down. It sets off a run of follow-ups, and each one uses a turn while you rebuild facts you could have said the first time. In the field experiment, the agent's advantage was topic coverage (6.8 topics against 5.5), and a human later read that coverage in the transcript. Your backup layer is what the agent's coverage draws from.

#### What sits under each answer

1. **Posting qualification** - The competency the agent has been told to assess
2. **Story** - One situation you can tell completely in your first answer
3. **Four detail lines** - Specific example, your own role, your method, the result
4. **Resume line** - The claim the human reviewer will check your transcript against

*Each posting qualification gets a story, each story gets four details ready for follow-ups, and every detail traces back to a line on your resume.*

### Building the map (2 hours)

1. Copy every qualification line from the posting into a list. Split any line that combines two skills into two rows.
2. For each row, choose one story from your own work. Use each story for no more than two rows.
3. Attach one metric to each story: a count, a percentage, a time saved, or a scale.
4. Mark any row with no story. Those are the questions where you will improvise, so decide now what adjacent experience you will offer and say so plainly.

Pulling the qualifications out and finding the matching lines in your history is the slow part. [Refolk](/candidates) does it for you when it tailors your resume to a posting and scores your fit: the qualifications it flags as weak are the rows that need a story most.

### Writing the backup layer (1.5 hours)

For each story, write four lines, each one a fact you can say in a single breath:

- **Example:** the specific situation, with the date range and the team or customer.
- **Own role:** what you did yourself, written as "I" sentences, separate from what the team did.
- **Method:** the steps you took, and one alternative you rejected and why.
- **Result:** the number, plus how it was measured.

> A vague first answer does not just score low. It sets off follow-ups that use turns you cannot take back.

**What a full backup layer proves:** you can answer any of the four probes without stopping to think. **When it misleads you:** if every result line uses the same impressive number, or the "own role" line describes the team's work, the human reviewer will notice even if the agent does not.

## The recorded transcript check: length and resume match

Record yourself answering out loud, transcribe the recording, and check two things: whether each answer stays under a length ceiling you have chosen, and whether every claim in it appears on your resume. The agent scores the transcript, and the human reviewer compares it with your CV, so test the same thing they will see.

### Choosing a length ceiling

No standard length has been established. The 60 to 90 second figure that gets repeated widely does not appear in any vendor guide I found. Third-party advice conflicts:

- About one to two minutes per answer, from one guide.
- Two to three minutes per behavioral answer, from another.
- Twenty to forty-five seconds for a follow-up answer, from a third.

My working rule: give the first answer to a topic enough room to cover the example, your role, your method, and the result, and keep follow-up answers short and specific. Choose your ceiling before you record, then hold yourself to it. If the platform shows a timer or a cap, as TestGorilla's one-way answers do at 30 seconds, 1 minute, or 2 minutes, the platform's limit overrides your rule.

### Running the check (1 hour)

1. Read five likely questions aloud from your competency map and answer each one on a recording, one take only.
2. Have a colleague or yourself ask one follow-up per answer from the four probe types, and record those answers too.
3. Transcribe the recording with any speech-to-text tool you already have.
4. Time every answer. Flag any that runs over your ceiling, and any first answer short enough that it would invite a follow-up for a missing basic fact.
5. Highlight every factual claim in the transcript: titles, dates, numbers, tools, team sizes. Check each one against your resume.
6. Fix the resume or fix the answer. If a true claim is missing from your CV, decide whether to add it before the screen.

The resume match matters because the person who reviews your screen has your transcript and your CV side by side. If your resume was written from your own history, as Refolk writes it, this check becomes a line-by-line comparison instead of a hunt through old documents.

**What a clean check proves:** your answers fit the time and nothing in them contradicts your application. **When it misleads you:** a script you have read aloud ten times will pass this check and still sound recited. Do one final take answering from the four detail lines, not from a script.

## How this goes wrong

Most failures here come from three things: treating a request as if it were a right, citing a law before it takes effect, and believing vendor-blog claims about what gets scored. Each one has a quick local check, and you should run every check before you rely on the claim.

> **Watch out:** A request is not a right
>
> Under NYC's Local Law 144 you can request an alternative process, and the employer can refuse. If your email treats a request as a guaranteed right, it costs you goodwill and gains you nothing.

### 1. Statute changes that never happened

**What it looks like:** a blog claims a "February 2026 Amendment" to the Illinois Video Act with automatic deletion after 30 days. Other sources date the Act to 2020, with deletion only on request. **Check:** read 820 ILCS 42 itself before you put any clause from it in an email.

### 2. Treating the opt-out as a right where it is not one

**What it looks like:** "Under NYC law I am entitled to a human interview." The rules require notice of how to request an alternative, not that one be provided. **Check:** quote the exact rule in your decision note, as the rule callout above requires.

### 3. Citing Colorado or California rights too early

**What it looks like:** asking for a California ADMT opt-out on an invite dated before January 1, 2027, or citing Colorado's human-review right while its enforcement is contested. **Check:** put the invite date next to the effective date in your note. If the invite date is earlier, leave the right out.

### 4. Myths about voice and tone scoring

**What it looks like:** one interview-prep blog says filler words "are weighted negatively," which contradicts HireVue's own statement that it dropped tone and pause inputs. Another vendor, Talview, does say it evaluates tone. **Check:** read the explainability page for the vendor named in your invite, not a general guide. If the vendor publishes nothing, treat the claim as unknown, not as true.

### 5. Conflicting study details

**What it looks like:** citing 2,290 applicants from one metadata page, or a retention figure of 16% in one place and 18% in another. **Check:** use the SSRN abstract of "Voice AI in Firms" as the authority, and quote only the figures it states.

### 6. Answers that are too short, and follow-ups that are too long

**What it looks like:** a two-sentence first answer that gives the rubric nothing to score and sets off follow-ups, or a three-minute answer to a follow-up that only asked for the result. **Check:** time each recorded answer in the transcript check against the ceiling you set.

### 7. An accommodation request sent after scoring, with no dated copy

**What it looks like:** a phone call to the recruiter after a rejection, with nothing in writing. **Check:** send it by email, before the screen if you can, and keep the sent email with full headers. Practitioner advice says a request made right after you are screened out is still worth making, but only a written one counts as evidence.

## Sitting the screen, logging the request, and keeping the record current

Take the screen in a quiet room on a connection you have tested, lead every answer with the example, and save the completion confirmation. Then, on the same day, send a written data request that cites the clock that applies to you. A dated request is what lets you act later, whether you want to correct something, appeal, or just keep a record.

### During the screen

- Read the platform's retake rule before you start. Third-party guides say Equip allows none once submitted, while Flowmingo allows re-recording for up to two hours unless the recruiter has turned that off.
- Ask the recruiter beforehand if you want notes in front of you. No platform's policy on notes has been publicly established.
- When a follow-up arrives, work out which of the four probes it is (example, role, method, or result) and answer from that detail line only.
- Screenshot or save the completion page and any confirmation email.

### Logging the data request

Only some of these requests are backed by a rule. NYC gives you the type of data, its source, and the retention policy within 30 days. Illinois gives deletion of a recorded video interview within 30 days of your request. California gives access within 45 days for invites dated January 1, 2027 or later. No rule I found gives you a general right to a copy of your transcript, but asking costs nothing, and the date of your request is a record either way.

**AI interview data request**

```
Subject: Request for my AI interview data, [Role title], interviewed [date]

Hi [Recruiter name],

I completed the AI [voice / video] interview for [Role title] on [date] through [vendor].

I am requesting:
1. A copy of the transcript and any recording of my interview.
2. The type of data collected about me, its source, and your data retention policy.
3. [Illinois video interviews only] Deletion of my interview recording within 30 days of this request.
4. [California consumers, invites dated January 1, 2027 or later] Access to information about the automated decision-making technology used in my application, within 45 days.

[Rule line, if one applies: "As a New York City resident, I understand the information in item 2 is due within 30 days of a written request under Local Law 144."]

Please confirm you have received this request.

Thank you,
[Your name]
```

*Send on the day of the screen. Keep only the numbered items and the rule line that apply to you, then set a calendar reminder for the day the clock runs out.*

> **Tip:** Keep one folder for the whole run
>
> Save the invite, your one-line triage, your decision note, any request emails, the completion confirmation, and the data request in one folder named with the role and date. If you ever need to make a point later, the dates in that folder are your evidence.

#### Before you call the run done

- [ ] The one-line triage record names the vendor, deadline, format, job location, and your residence.
- [ ] Every right in your decision note has a quoted rule and an effective date earlier than the invite date.
- [ ] You have either a dated acceptance note or a sent alternative or accommodation request saved with headers.
- [ ] Every posting qualification has a story, a metric, and four backup detail lines.
- [ ] No recorded practice answer runs over the length ceiling you chose.
- [ ] Every claim in your practice transcript appears on your resume.
- [ ] The completion confirmation from the screen is saved.
- [ ] A dated data request is sent, and a calendar reminder is set for the day its response clock runs out.

### Keeping the record current

These rules are changing, so re-check the following before each new invite instead of trusting this page:

- **Colorado SB 26-189:** search for its enforcement status and any court ruling, because at least one legal source reports it is on hold.
- **California ADMT:** confirm the January 1, 2027 date still holds, and whether the hiring exception has been narrowed.
- **Illinois:** check whether the Department of Human Rights has issued new notice rules under HB 3773, and whether 820 ILCS 42 has been amended to cover voice-only agents.
- **Federal:** the EEOC's current enforcement plan omits AI, but the ADA still applies. Re-check whether new guidance has been published.
- **The vendor:** re-read the explainability page for the vendor in each new invite, because a vendor that dropped tone scoring can bring it back, and a new vendor may never have dropped it.

## Frequently asked questions

### Can I ask for a human instead of an AI interview?

Yes, you can always ask, but in most places the employer is allowed to say no. New York City's Local Law 144 says the notice must explain how to request an alternative process, but the employer does not have to provide one. California adds an opt-out from January 1, 2027, and it has a hiring exception. If a disability affects how you answer, ask for an accommodation instead. Either way, make the request in writing before the screen and keep the sent copy.

### What does an AI phone screen actually score?

It depends on the vendor, and the vendor's own explainability page is the only reliable source. Third-party guides say the employer receives the transcript, the scores with their reasoning, and a recording when one exists. HireVue said it would stop using speech inputs such as variation in tone or pauses. Talview's Ivy, by its own description, evaluates fluency, tone, and relevance. I found no independent evidence that filler words are scored.

### How long should my answers to an AI interviewer be?

Nobody has established a correct length. I found no vendor guide that states the commonly repeated 60 to 90 second figure. Third-party advice varies: about one to two minutes per answer, two to three minutes for a behavioral answer, and 20 to 45 seconds for a follow-up. Pick a ceiling, time your recorded practice answers against it, and keep each first answer complete enough that the agent does not have to use follow-ups to collect basic facts.

### How do I request my AI interview recording or transcript?

Email the recruiter on the day of the screen with a written request, and keep the sent copy. A New York City resident can require the type of data collected, its source, and the retention policy within 30 days. For Illinois video interviews, deletion is due within 30 days of a request. California access requests get a 45-day clock from January 1, 2027. No rule I found gives you a general right to a transcript copy, but asking costs nothing.

### Can I retake an AI interview if a follow-up went badly?

It depends on the platform, so check the invite before you start. Third-party guides say Equip's conversational mode allows no retakes once you submit. Flowmingo lets you re-record answers for up to two hours after you finish, unless the recruiter has turned retakes off. For other platforms I could not verify the retake policy. Prepare as if you get one take, because that is the safe assumption.

### Does the Illinois AI Video Interview Act cover voice-only AI interviews?

Nobody has publicly established whether it does. The Act, 820 ILCS 42, covers AI analysis of recorded video interviews for Illinois positions and requires notice, an explanation of how the AI works, consent, and deletion within 30 days of a request. Whether a voice-only agent falls under it is an open question, so read the statute text. Separately, Illinois HB 3773 has made discriminatory or undisclosed AI use in employment a civil rights violation since January 1, 2026.

---

*From the Refolk guide library. I revise these guides rather than replacing them, so the current version is always at https://www.refolk.ai/candidates/guides/live-ai-interviewer-run*
