AI Recruiting Assistant: Asking Questions About a Candidate's File

• After several interviews, what you know about a candidate is scattered: resume, notes, interview reports, emails. Finding what they said on a specific point means rereading everything.
• A useful AI recruiting assistant answers from the candidate's file and nothing else, and cites its sources.
• It separates what the candidate said from what it infers, and says plainly when the information is missing.
• It refuses to comment on protected characteristics and never makes the decision for you.
• In ResumeRank, that is the job of Iris, included in the Hiring and Team plans.

The hiring manager calls you ten minutes before meeting the candidate: "She mentioned the size of her accounts, didn't she?" You vaguely remember a number. It is somewhere in your notes from the first interview, or in the recording, or maybe in the report. This happens in every hiring process, and it is exactly the problem an AI recruiting assistant should solve.

The real problem: a file nobody rereads

A candidate who reaches the final round leaves behind a resume, an analysis, a screening call note, two or three interview reports, sometimes hours of recordings. All the information is there. It is simply out of reach when you need it:

• before the next interview, to know what is left to dig into;
• at decision time, to find the argument that tips the balance;
• when the hiring manager asks a precise question and nobody has time to replay a 45-minute conversation.

A good assistant does not replace these documents. It lets you ask them a question.

Why a general-purpose chatbot is not enough

Many recruiters have tried pasting a resume and some notes into a general-purpose chatbot and asking their question. For a one-off, the answer can help. For a hiring process, the method hits four limits.

It only knows what you pasted. If the answer is in the manager interview and you only pasted the resume, it will answer anyway, with what it has, or by filling in the gaps.

It can make things up. A general-purpose language model produces a plausible answer, not necessarily a true one. In a candidate file, an invented quote or an approximate figure can skew a decision.

It cites nothing. Without a source, you cannot check where a claim comes from, or show it to the hiring manager.

Candidate data leaves your recruiting tool. Copying resumes and transcripts into a consumer service raises a real confidentiality question.

We covered these limits in ChatGPT and LLMs for recruiting and in why ChatGPT is not enough for professional recruiting.

What a good AI recruiting assistant should do

Before choosing a tool, check these seven points.

Capability	Why it matters
Answers only from the candidate's file	No invented or imported information
Cites the source of every claim	You check in one click, and you can show the evidence
Separates what was said from what is inferred	An inference does not carry the weight of a quote
Says when the information is missing	"Nothing in the file" beats a guess
Refuses protected characteristics	Age, health or family status never enter an opinion
Drafts ready-to-send documents	Manager note, follow-up questions, message to the candidate
Never decides for you	It argues, you decide

The last point is not just a matter of principle: the EU AI Act classifies AI systems used for recruitment as high-risk, which requires human oversight of decisions.

Iris: the assistant that has read the whole file

Iris is the assistant built into ResumeRank. It opens from a job's Results page with the "Ask Iris" button and works on one candidate at a time.

What Iris has read

• the job: the job ad, the weighted criteria and the name of the hiring company;
• the resume and its analysis: score, summary, strengths, points to check, suggested questions;
• the decision already made about the candidate, if any;
• every step: interview reports with their timestamped transcripts, and the recruiter's debrief notes.

It knows nothing else: not your other candidates, not your pay policy, not the web. That is by design.

What you can ask

You ask	Iris answers
"Did she mention the size of the accounts she managed?"	The answer in one sentence, with the quote and the exact moment in the interview
"What is left to check before the final interview?"	The points still open across the steps, each with its source
"Prepare a note for Julien before his interview"	A ready-to-send note: in 30 seconds, already validated, to dig into, watch-outs
"Should we give him a test?"	A reasoned opinion, without deciding for you
"Write a message inviting him to a second interview"	A message ready to copy, with no internal information
"How old is he?"	A one-sentence refusal (protected characteristic) and a useful question to ask instead

Four documents are one click away: a note for the hiring manager, a report for leadership, follow-up questions and a message to the candidate. You can also limit answers to a single step (the "Answer scope" setting offers the whole file or one specific interview), or ask whether the candidate's answers are consistent from one interview to the next.

How Iris answers

With sources. Every claim is followed by its source: an interview or note passage with its timestamp, the resume, or the job ad. Clicking the timestamp shows the exact transcript passage under the answer.

Without filling the gaps. When the file does not contain the information, Iris says so ("nothing in the file about…") and suggests turning it into a question for the next interview.

Keeping facts and inferences apart. What the candidate said and what Iris concludes are never mixed.

Privately, except what you share. The conversation is private. A useful answer (a manager note, a summary) can be pinned to the file, which makes it visible to your team.

What Iris does not do

It does not compare candidates. It answers about one file at a time. To compare, the ranking and the scores remain the right tool.

It is only as good as the file. An interview that was neither recorded nor debriefed does not exist for Iris. That is why a debrief note after a screening call or an in-person interview matters.

It does not decide. It gives a reasoned opinion when you ask. The decision stays yours.

How to use it in your process

Before each interview, ask what is left to check. One question tells you what to dig into, without rereading the previous reports.

After the interview, ask the questions the report leaves open, or request a short version for the hiring manager.

At decision time, ask for the arguments for and against, with their sources, then decide.

For the hiring manager, a note pinned to the file replaces the email back-and-forth before their interview. See also how to validate a candidate with a hiring manager without email ping-pong.

Iris is included in the Hiring and Team plans, with no question limit. Plan details are on the pricing page, and you can see Iris in action on the home page.

Frequently asked questions

Can an AI assistant replace the interview report? No. The report structures the evaluation and serves as a record; the assistant lets you query it and turn it into other documents. They work together.

Can Iris be wrong? It can misread an ambiguous passage, like a human reader. That is why it cites every source: you check the passage before using it. It does not, however, invent quotes, figures or experience.

Is the candidate informed? They should be. In France, for example, the Labour Code requires candidates to be told in advance about the recruitment methods and techniques used to assess them (article L1221-8). Mention that interviews are recorded and analyzed in your invitation.

Where is the data stored? Resumes, recordings and reports are hosted in France, in Paris, and can be deleted at any time.