Questions to Ask Before Buying an AI Interview Tool in India

Before you buy an AI interview tool in India, get written answers to three questions: what does it do when it suspects impersonation, what can a person see and override behind each score, and what does one role cost in total? Then check DPDP data handling, ATS fit and exit terms.

Infosys postponed online assessments for more than 20,000 applicants in June 2026 after detecting multiple instances of impersonation and assessment malpractice, according to People Matters. Gartner predicts that one in four candidate profiles worldwide could be fake by 2028. An interview tool bought without a question about that is a tool bought for a different market.

This page gives eight questions for an engineering or volume-hiring lead walking into vendor calls. Each comes with what a good answer sounds like and what a red flag sounds like.

What do generic AI interview checklists leave out for an Indian IT team?

They leave out impersonation, India's data law and the cost of one role at your volume. When we put the buyer question to Perplexity in October 2026, its answer covered volume, scoring oversight, bias, low-bandwidth and language fit, privacy and a pilot, and never mentioned a candidate who is not who they claim to be.

Two vendor-written interview-tool checklists we read the same month cover ATS integration, bias monitoring and return on investment. Neither names the DPDP Act or impersonation. The questions below keep the standard ones and add those gaps.

Which questions test whether the interview is real?

Ask how the tool establishes that the person on screen is the applicant, and who decides when something looks wrong. Those two answers decide whether your engineers' hours go to real candidates. In Gartner's 2024 survey of recruiting leaders, 54% agreed AI had made candidate fraud a significantly larger problem, while only 7% were using identity document validation software, as cited by Greenhouse in June 2026.

What does the tool do when it suspects impersonation or live AI help?

Good answer: it says what its integrity safeguards are for, says a flag sends the interview to a person for review, and says one weak signal never decides an outcome. A vendor may keep detection mechanics private, and that is reasonable. Red flag: a headline detection accuracy figure with no method behind it, or a tool that rejects a candidate automatically on a flag. For one vendor's approach, see how AgentR handles AI interview cheating.

What does a recruiter see after a flag, and what can a candidate say in reply?

Good answer: a clear category the recruiter can act on, and a way for a candidate to explain an odd setup such as a shared laptop or a weak connection. Red flag: a bare "suspicious" label with no route to reply. A tool that treats an unusual setup as cheating fails honest candidates, so ask how it avoids that.

Which questions test whether the score can be trusted?

Ask what evidence sits behind every score, whether a person can overrule it, and whether the tool has been tried on a role where you already know the answer. A score you cannot inspect is an opinion you cannot argue with. Structured interviews, where every candidate gets the same questions scored against a rubric, had the highest mean validity of the selection methods in Sackett and colleagues' 2022 meta-analysis, at r = .42. Ask whether the tool scores that way.

What evidence does each score rest on, and can our engineers change the rubric?

Good answer: a score per criterion with the candidate's own answer beside it, and a rubric your engineers can edit before the role goes live. Red flag: one overall "fit score" with no rationale, or a rubric you cannot see.

Can it reject or advance anyone without a human decision?

Good answer: every decision stays with a person, written into the contract, and the shortlist is a recommendation. Red flag: auto-reject thresholds switched on by default. India's DPDP Act gives candidates no right to demand human review of an automated decision, as our DPDP explainer sets out, so this safeguard is yours to require. The law will not require it for you.

Can we pilot it on a closed role and compare it with what our interviewers decided?

Good answer: a free or capped trial on one role you have already filled, with your engineers scoring the same candidates blind. Red flag: a pilot offered only after an annual contract is signed.

Which questions test the cost, the data and the exit?

Ask for the total price of one role, where candidate data lives and when it is deleted, and how you leave. Buyers tend to discover the answers after signing, so get them first.

What does one role cost in total, including GST and longer interviews?

Good answer: a written estimate for a role of your size, for example 100 applicants and 25 interviews, with GST included and a spend cap you can set. Vendors charge per interview, per credit, per seat or per annual contract, so a headline price cannot be compared until it becomes the cost of one role. Red flag: a per-interview figure that leaves out review, proctoring or tax. The arithmetic is worked through in how much AI interviewing costs in India.

Where is candidate data stored, how long is it kept, and who answers for it?

Good answer: a named storage location, a retention period tied to the hiring purpose, deletion on request, breach-notice commitments in the contract, and no training on candidate data without consent. Under the DPDP Act you remain the data fiduciary and the vendor is your processor. The government's November 2025 notification of the DPDP Rules sets an 18-month phased compliance period and a penalty of up to ₹250 crore for failing to keep reasonable security safeguards. Red flag: "we are DPDP compliant" with no document behind it.

How does it fit our ATS, and what happens to our data if we leave?

Good answer: a named integration or a documented export, a stated window for exporting your data, and written confirmation of deletion at exit. Red flag: "integrates with everything" and no exit terms.

Where do buyers usually get this wrong?

They test the demo and skip the failure. A demo shows a cooperative candidate on a good connection. Ask the vendor to show you a flagged interview, a rubric edit, a deleted record and an estimate for your volume.

The second mistake is reading "verification" as background verification. A tool that checks claims for consistency is not a BGV vendor, and nothing in an AI interview replaces the formal check that comes after an offer. What to send into that check first is covered in what to check before background verification in India.

Questions people ask

What evaluation criteria should I use to compare AI interview tools for volume hiring?

Use the eight questions above, and convert every price into the cost of one role at your own applicant and interview counts. Add peak load: ask what happens in the week a campus drive lands.

How do I assess bias and compliance in an AI interview platform in India?

Ask for the vendor's bias testing method and the date of its last test, then ask for its DPDP documents. The most developed bias audit rule is a US one, covered in what NYC Local Law 144 actually checks, so an Indian buyer has to ask for the evidence directly.

What data security measures matter for AI interviewing tools?

Ask where recordings and transcripts are stored, who can open them, how long they are kept and whether they are used to train models. Then ask what the vendor commits to in writing if there is a breach.

How do I integrate AI interviews with my ATS and HRIS?

Ask whether the integration is a named connector, a documented export or custom work, and who pays for custom work. Ask to see the integration run on your own ATS before you sign.

How AgentR answers this checklist, and where its answers are weak

AgentR reads every application in full, checks candidates' claims against the public record, runs structured interviews scored against a rubric written before anyone applied, and hands the team a ranked shortlist with the evidence attached. Here is how it answers the eight questions, including the weak ones.

• Impersonation: AgentR has integrity safeguards for live AI help, and its interview integrity page says a flag sends the interview to a person and that no single signal decides an outcome. The mechanics are unpublished and there are no performance benchmarks.
• After a flag: recruiters see a category rather than a raw signal log. That page does not describe how a candidate replies to a flag, so ask us.
• Score evidence: answers are scored against a written rubric, and the shortlist carries the evidence.
• Human decision: nobody is auto-rejected, and every decision stays with a person.
• Pilot: the first role is free, up to 50 applicants reviewed and 5 interviews, with no card. Beyond that it is paid.
• Cost: ₹25 per applicant reviewed and ₹10 per interview minute, before GST, as of October 2026. The reference role comes to ₹10,000, with an estimate before launch and a spend cap per role.
• Data: the pricing page links a privacy policy and a security page. Read those against your own questions rather than relying on this post.
• ATS and exit: this is the weakest answer. Custom ATS and HRIS integrations belong to larger plans, and the pages we checked name no standard connectors and no public API. Unused paid balance can be refunded on request.

AgentR is in private beta, so there are no customer references and no outcome data to offer. Its verification checks claims for consistency against public and organisational signals, does not contact employers or institutions, and is not background verification. If your risk is more about screening volume, how AgentR screens high application volume is the closer page.

See how AgentR handles interview integrity, and what it leaves to your people.

Read the details on the AI interview integrity page.