An AI interviewer is software that conducts a structured first interview with an applicant, usually by voice, using one approved question guide for everyone. It records the answers, turns them into a transcript, summary and scorecard, and hands that evidence to a recruiter. AI voice interview software belongs after the application check and before the human interview. It gathers comparable facts at scale; the hiring decision stays with a person, which is also what GDPR and the EU AI Act expect.
That positioning prevents a common category mistake. A hiring workflow needs evidence about availability, experience and examples of work; it does not need an artificial substitute for rapport or judgment. A voice interviewer is a data-collection tool, not a judge of people. The broader guide to AI interviews and voice recruiting places spoken interviews within the full recruiting process.
How does AI voice interview software work?
A candidate receives an invitation and chooses a route: a web link, WhatsApp or a phone call. The invitation should say what the conversation is for, how long it takes, who reviews the answers and how to request another format. The interview starts only after that is clear.
Speech becomes a working transcript
Automatic speech recognition turns the audio into text in short sections. This is where real conditions matter: accents, industry vocabulary, a poor connection, pauses and background noise all affect the result. Test a provider with the voices and environments your applicants actually use, not only with a polished demo.
The transcript is evidence, not a verdict. Recruiters need a route from every summary line or score back to the answer behind it, so they can spot a recognition error and correct their reading. A concise summary that cannot be checked is convenient, but not dependable.
The interview follows a controlled guide
The dialogue layer uses an approved guide: which question comes first, which follow-ups are allowed, when an answer needs clarification and when a person takes over. That keeps the interview comparable across applicants without turning it into a script that ignores what was said.
A well-designed system asks for a concrete example after a vague answer. It does not invent criteria mid-call, and it does not infer personality, honesty or emotions from a voice. Inferring emotions of employees is prohibited under Article 5(1)(f) of the EU AI Act (in force since 2 February 2025), and the Commission's guidelines apply this to recruitment as well. Ask any provider how it handles that.
A scorecard makes the conversation usable
After the call, answers are grouped against the agreed criteria, for example shift availability, work authorization, location or a job-related scenario. A good scorecard separates three things: the criterion, the evidence the candidate gave and the reviewer's conclusion. It gives the next interviewer a focused starting point; it does not predict who will succeed in the role.
AI interview formats compared
"AI interviewer" covers several formats. Which one fits depends on the role, the audience and the effort you can ask of applicants.
| Format | Strength | Limit | Sprad credits per interview |
|---|---|---|---|
| Manual phone screen | Rapport, instant judgment | Does not scale, notes differ per recruiter | n/a |
| Chat interview (web or WhatsApp) | Low barrier, works in any noise, readable record | Less spontaneous than speech | 10 (WhatsApp: 15 invitation, 50 full conversation) |
| Voice interview | Natural for non-desk roles, quick, richer examples | Needs good speech recognition per language | 30 (by phone: 40) |
| Video interview | Closest to a live conversation | Highest effort for candidates, more sensitive under data protection rules | 80 |
Source: credit prices from sprad.io/pricing, as of 29 October 2026. The CV rating costs 2 credits per application and the candidate journey (updates, reminders, appointments) 9.
Where voice interviews help, and where they do not
The most useful placement is structured early screening. After a basic application check, a voice conversation collects what otherwise takes many similar recruiter calls. It helps most where volume is high, speed matters or a desk-based form is a poor fit for the audience.
- Before a recruiter call: collect recurring, role-relevant facts and examples so live time goes to deeper questions.
- For transparent minimum requirements: document answers to agreed conditions instead of making a broad claim about suitability.
- For multi-channel access: offer the route that fits the audience, as part of the candidate portal experience.
- Alongside application review: add context to structured CV screening where documents leave questions open.
It does not belong at the end of the decision. It cannot conclusively assess fit, experience chemistry, understand every personal circumstance or negotiate pay and expectations.
Is AI voice interviewing allowed? GDPR, AI Act and works council
Short answer: yes, with conditions. This is an orientation, not legal advice; involve your data protection officer and, where relevant, counsel.
- GDPR: Article 22 gives people the right not to be subject to a decision based solely on automated processing with legal or similarly significant effect. A scorecard that a recruiter actually reviews, with the real choice of next step, keeps a person in the decision. Applicants must also be informed about the processing (Articles 13 and 14). Whether a data protection impact assessment is needed, your data protection officer should decide before launch.
- EU AI Act: candidates must be told they are speaking with an AI system (transparency, Article 50, applicable since 2 August 2026). AI used to recruit or select people is a high-risk use under Annex III. Those obligations apply from 2 December 2027, after the Digital Omnibus Regulation (EU) 2026/1744 (in force 27 July 2026) moved them from August 2026. Planning for them now is cheaper than retrofitting.
- Germany, works council: if AI influences selection guidelines, section 95(2a) BetrVG gives the works council co-determination rights; technical systems that can monitor behaviour fall under section 87(1) no. 6. Our article on whether the works council has to approve AI in recruiting covers the details.
Common to all three: disclose, keep a human in the decision, offer an alternative route and keep the record explainable.
How to choose a voice-interview platform
Do not judge software by whether the voice sounds natural. Judge the evidence chain from a spoken answer to the record in your applicant tracking system. Six checks:
- Language quality: test with your languages, regional accents, role vocabulary and realistic call conditions.
- Guide control: questions, permitted follow-ups, stop rules and human handoffs can be approved, versioned and constrained.
- Traceability: every summary statement or score links back to an answer, and a recruiter can flag an error.
- Channels: match web, messaging and phone to the audience and offer an equivalent alternative.
- ATS write-back: define which fields, statuses, transcripts, summaries and scorecards return to the candidate record.
- Commercial model: model real volume and compare per completed interview, per minute, per credit or per licence, including repeat calls, languages and integration work.
The AI interview and voice tools category helps map the market. It cannot replace a role-specific pilot: a provider that shines in an English browser demo can still fail in a multilingual phone flow.
What does AI voice interview software cost?
Compare the cost unit, not the platform fee. At Sprad there are two ways. With credits (pay per use, company-wide) a voice interview costs 30 credits: €3.00 at the list price of €0.10, €2.70 on Atlas Base, down to €2.10 on Atlas Base Max. With the flat rate per application (billed yearly) CV screening, chat interviews and the candidate portal cost €0.49 per application; voice interviews for every fifth application add €0.20, so about €1.00 per voice interview.
Worked example (our assumptions): 1,000 applications a year, every fifth gets a voice interview, so 200 interviews. A manual phone screen with notes takes about 20 minutes, which is roughly 67 recruiter hours. On the flat rate the full package costs 1,000 × €0.69 = €690 a year, of which €200 is the voice part. Via credits the 200 interviews use 6,000 credits, €600 at list price. The saving is only real if recruiters use the scorecard to skip repeat questions.
Implementation: typical mistakes
- Starting with every vacancy. Start with one well-defined role and review the same calls with the hiring manager.
- Questions without a criterion. Each question must feed one scorecard criterion, otherwise it is noise.
- No alternative route. Candidates with technical, language or accessibility barriers need an equivalent path.
- Reading the score as a ranking. The score points to the next question; it is not a verdict.
- Skipping consent and notice. Say plainly that the conversation is with AI, what happens to the data and who reviews it.
Run test interviews before going live. If recruiters and hiring managers cannot explain their conclusions consistently, more automation will not fix the process.
Candidate experience affects the quality of answers
A good invitation tells candidates why they speak with software, the expected effort, what happens next and how to reach a person. Clear expectations help applicants prepare useful examples. Candidates who feel surprised, rushed or unable to use the channel give less useful information, which undermines the efficiency you wanted.
Where Sprad fits, and where it does not
Sprad runs CV screening, voice, chat and video interviews, the candidate portal and active sourcing in one pipeline with a talent network that stays yours. Concretely: every application is read and rated per criterion with the quote behind it, and nothing is rejected automatically. Voice interviews run in the portal, over WhatsApp or by phone, in more than 30 languages, and standard ATS integrations are available with more on request. The same account also sources candidates and reaches them by LinkedIn and email, so screening and sourcing share one record.
Limits we state openly: in the US, text chat works best today, and custom (for example US) phone numbers need an individual setup, usually in an enterprise plan. Hosting and security questions we answer per deployment; ask any provider the same.
A simple rule for deciding whether to deploy
Deploy AI voice interviews only when five conditions hold at once: the role has explicit criteria, every question serves one of them, outputs trace back to the conversation, a named person reviews the next-step decision and candidates have an understandable alternative. If one is missing, fix the hiring design first.
Frequently asked questions
What is an AI interviewer?
Software that runs a structured first interview with an applicant, mostly by voice or chat, and prepares a transcript, summary and scorecard for a recruiter. It collects evidence; it does not hire.
Can AI voice interview software replace recruiter interviews?
It can take over recurring evidence gathering before a recruiter conversation. Nuanced assessment, relationship-building, role explanation and the final decision stay with people.
Is an AI interview legal in the EU?
Yes, with conditions: inform candidates, keep a human in the decision (GDPR Article 22), avoid emotion inference, involve the works council where it applies and prepare for the high-risk duties from 2 December 2027.
Is an automated scorecard objective?
Not automatically. Its value comes from relevant criteria, comparable questions, visible evidence and routine human review.
Does it work for frontline or non-desk candidates?
Often better than a long form, because phone and messaging are accessible. It needs plain language, suitable timing and a real alternative route.
What should flow back into the ATS?
At least interview status, answers to agreed criteria, summary, scorecard and open questions. Whether recordings are kept follows your privacy and retention design.
What is the safest way to begin?
One role, a controlled guide and a mandatory human review. Before scaling, compare the evidence, reviewer conclusions and candidate feedback, not only completion rates.
Read next: Does the works council have to approve AI in recruiting?
Jürgen Ulbrich
Jürgen Ulbrich has more than a decade of experience in developing and leading high-performing teams and companies. As an expert in employee referral programs as well as feedback and performance processes, Jürgen has helped over 100 organizations optimize their talent acquisition and development strategies.
