# AI recruiting agents: only 0.1% ask for a human, and that may be their mistake

> Source : https://stubby-ai.com/en/blog/agent-ia-recrutement-01-demandent-un-humain-et-cest-peut-etre-leur-erreur
> Publié le 2026-11-13 par Stubby AI (https://stubby-ai.com)

**En bref.** According to Workable, only 0.1% of candidates ask its AI recruiting agent for a human, yet Greenhouse reports that 46% would like to be able to. Does this figure measure acceptance, or unawareness of a right? According to stubby-ai.com, these agents turn every message into a ticked box: a simple "no" or a bare number can close your application before a recruiter ever opens it. In this article, you will learn how AI pre-screening works, the 5 riskiest knockout questions with "before/after" answers, the 4 situations where you should ask for a human, a 5-step method, your transparency rights (CNIL, the French data protection authority) and a checklist to go through before you reply.

## AI recruiting agents: the figure that should make you wince

According to the software vendor Workable, in a press release dated 10 November 2026, only **0.1% of candidates** who interacted with its AI recruiting agent asked to speak to a human. Workable sees this as proof that candidates accept automation. I mainly see a question: did they know they could ask?

A vendor press release is a source with a vested interest, and that should be said without assuming bad faith. The figure measures behaviour, not satisfaction. Between the two lies everything candidates don't know about those few messages.

The [Greenhouse 2026 report](https://www.greenhouse.com/blog/2026-candidate-ai-interview-report), based on 2,950 jobseekers across five countries, tells a different story. **46%** of them want to be able to request an interview with a human. Wanting an option and using it are two different things, especially when nobody tells you it exists.

## What an AI pre-screening agent actually does

Picture Léa, a project manager, applying for a job on a Sunday evening. Three minutes later, a friendly WhatsApp message arrives: "Hi Léa! A few quick questions to move your application forward 😊". She replies in a couple of words, as she would to a friend, between two episodes of a series.

On Monday morning, a polite email tells her that her profile isn't a match. No recruiter has opened her CV. Léa didn't have a conversation: she filled in a **knockout form disguised as a dialogue**.

In practice, a pre-screening recruitment chatbot:

- asks closed questions, often knockout ones: availability, salary, willingness to relocate or commute, right to work, driving licence, qualifications, years of experience;
- checks that your answers are consistent with the job ad and with your CV;
- books an interview if everything ticks the boxes, or closes the application if not;
- turns every message into a screening field, and therefore into a score or a decision.

The investigation by [Privacy International](https://privacyinternational.org/long-read/5798/humanless-resources-uncovering-ai-recruitment-software/) tested two platforms, Manatal and Talenteria, in January and February 2026. A "knockout question" feature immediately disqualified candidates judged unsatisfactory. The investigators also found that identical CVs could receive different scores.

To be fair: the investigation covers two tools, tested in a trial environment. Talenteria has since changed some elements. You can't generalise to every agent, but the mechanism described deserves your attention.

## The knockout questions that rule you out silently

The most telling example comes from the same investigation. A candidate listed Rust, JavaScript, C++, Python and SQL: the answer was judged non-compliant because Java and C# weren't on the list. A human recruiter would probably have recognised transferable experience.

Here are the five riskiest questions, with the answer that rules you out and the one that keeps the door open.

### Salary expectations

**Before:** "€55,000."

**After:** "I'm aiming for a range of €50k to €56k depending on the scope and benefits. I'm happy to discuss this with the recruiter."

A bare figure outside the range is enough to close the application. A reasoned range leaves room to manoeuvre, and asking for a conversation signals that the topic deserves a human.

### Availability and notice period

**Before:** "3 months."

**After:** "My contractual notice period is 3 months. An earlier departure can be discussed with my current employer."

Only write the second sentence if it's true. But if a negotiation is possible, keeping quiet about it amounts to ruling yourself out.

### Required years of experience

**Before:** "4 years." (for a role asking for 5)

**After:** "4 years as a management controller, preceded by 18 months as a work-study apprentice on monthly closes. That's over 5 years of practice in the tasks described."

### Location and remote working

**Before:** "No, I'm looking for fully remote."

**After:** "I can be on site 2 days a week. I'm 50 minutes from the office by train."

### Degree or certification

**Before:** "No."

**After:** "Not the master's degree requested, but a professional bachelor's degree and 6 years in these tasks. I'd like this point to be reviewed by a recruiter."

The golden rule: **never lie**. The agent cross-checks your answers against your CV, and an inconsistency will cost you more than a missing criterion. The aim is to be complete, not creative.

## An ally for some profiles, a trap for others

Let's be fair: these agents have real advantages. Instant replies, quick interview booking, and an end to ghosting at this stage. For a profile that ticks every box, it speeds things up.

Greenhouse confirms it: candidates don't reject AI as such. They object to processes that are opaque, inconsistent or lack visible human accountability. The report also notes that **38%** have already dropped out of a process because it included an AI interview.

The problem starts when your background doesn't fit a box. Privacy International points out that knockout questions suit objective criteria: right to work, driving licence, certification, place of work. They work far less well for equivalent experience or unconventional career paths.

Career change, a gap in your CV, specific mobility constraints: the agent can't interpret these. If that's you, also work on [your transferable skills for a career-change CV](/en/blog/cdd-de-reconversion-2026-ats-cv-competences-transferables), because the same translation problem arises with ATS.

## The 4 situations where you should ask for a human

Asking for a human isn't an admission of weakness. It's a strategic choice, provided you do it at the right moment and politely.

### 1. Your profile is unconventional

Career change, equivalent experience, non-linear path: your value lies in the context, which the agent doesn't read.

> "My background includes equivalent experience that's hard to summarise here. Would it be possible for a recruiter to review my application before a decision is made?"

### 2. The question is ambiguous and a yes/no puts you at a disadvantage

"Do you have 5 years of project management experience?" If the true answer is "yes, but not under that title", don't tick "no".

> "The answer is nuanced and a yes/no wouldn't reflect my experience. Could I clarify this point with someone from the team?"

### 3. The topic is sensitive

Salary, disability or a need for adjustments, a period of personal transition: these topics deserve someone able to listen.

> "I'd prefer to discuss this point directly with a recruiter. Could you tell me how to contact them?"

### 4. The agent misunderstands or goes round in circles

The same question asked three times, an off-topic reply, an absurd rephrasing: stop there. Persisting risks recording a wrong answer.

> "I think my answer has been misinterpreted. I'd like to be put in touch with a human."

If no option appears, simply type "speak to a human" or "adviser". If there's no response, contact the company by email, quoting the job ad and the date of the exchange.

## A 5-step method to prepare for pre-screening

Most candidates discover the questions live. You're going to do the opposite.

1. **Spot the likely knockout criteria** in the job ad: "mandatory", "required", "minimum", qualifications, location, languages, driving licence.
2. **Prepare a written answer for each**: short, factual, nuanced, following the "after" examples above.
3. **Align these answers with your CV**. If you mention 18 months of work-study, they must appear clearly on the document.
4. **Set your red lines**: the topics on which you'll ask for a human, whatever happens.
5. **Keep a record**: a screenshot of the exchange, then a final question about the next steps in the process.

Step 1 takes the longest when done by hand. That's where an analysis report like the one from [Stubby](https://stubby-ai.com) becomes useful: it cross-checks the job ad against your CV to flag the criteria on which you risk being filtered out.

Once you've cleared this stage, another one often awaits. Prepare for it too with [the interview that follows the AI recruiting agent: facing a conversational AI](/en/blog/premier-entretien-ia-conversationnelle-reussir-2026).

## Transparency, a right to exercise

According to Greenhouse, **70%** of US candidates assessed by AI say they weren't clearly informed that it was being used. Only **18%** of respondents believe most employers have clear AI policies. And **38%** want to know that a person will review the result before any decision.

In France, the [CNIL has made recruitment a priority of its 2026 inspections](https://www.cnil.fr/fr/controles-prioritaires-2026). It is targeting in particular automated decision-making, the information given to candidates and data retention periods. Its [recruitment guide](https://www.cnil.fr/fr/le-guide-du-recrutement) points out that the data collected must be relevant and that a rejection should not be based solely on automated processing.

This doesn't guarantee you a human interview in every situation. But you can legitimately ask these questions:

- Am I talking to an AI?
- Which criteria are being assessed at this stage?
- Will a human review my application before a decision?
- How long will my answers be kept?
- Which channel should I use to request a review?

A serious recruiter will have no trouble answering. Silence, on the other hand, tells you something about the company.

## Checklist: before replying to an AI agent

- [ ] I've identified the knockout criteria in the job ad
- [ ] I've written my answers on salary, notice period, location, experience and qualifications in advance
- [ ] My answers are consistent with my CV
- [ ] I'm replying calmly, not on the go between tube stops
- [ ] I've never replied with a bare "no" when reality is nuanced
- [ ] I know which topics I'll ask a human about
- [ ] I've taken a screenshot of the exchange
- [ ] I've asked about the next steps and whether a human will review my application

## Conclusion: taking back control

The 0.1% figure doesn't say candidates are satisfied. It says they don't yet know how to play this stage. Léa didn't have a weak profile: she simply didn't know she was being screened.

With an AI recruiting agent, every message counts as a ticked box. Prepare your answers, set your red lines, and ask for a human at the right moment: it isn't a weakness, it's taking back control.

## Synthèse

| Section | Key messages |
| --- | --- |
| The 0.1% figure | Workable data (according to the vendor): measures behaviour, not satisfaction. 46% of candidates want to be able to ask for a human (Greenhouse 2026). |
| Role of the AI agent | Closed questions, cross-checking against the job ad and CV, interview booking or closing the application: a knockout form disguised as a dialogue. |
| Knockout questions | Salary, notice period, experience, location, qualifications: favour complete, nuanced answers, never untrue ones, over bare replies. |
| Ally or trap | Fast and ghosting-free for profiles that tick every box; unsuited to career changes, equivalent experience and unconventional paths. |
| 4 situations to ask for a human | Unconventional profile, ambiguous question, sensitive topic, agent that misunderstands or goes round in circles. Polite wording ready to copy. |
| 5-step method | Spot the criteria, prepare answers, align them with your CV, set your red lines, keep a record. |
| Transparency | CNIL: 2026 inspections on automated decision-making and information given to candidates. Ask whether a human will review your application. |
| Checklist | 8 points to check before replying to an AI recruiting agent. |

---

Stubby AI analyse une candidature précise : collez une offre, déposez votre CV, obtenez un diagnostic puis un rapport complet. Premier rapport offert : https://stubby-ai.com/
