# AI-optimised CV in 2026: the defensibility test that keeps you out of interview traps

> Source : https://stubby-ai.com/en/blog/cv-optimise-ia-en-2026-le-test-de-defendabilite-qui-evite-les-pieges-en-entretien
> Publié le 2026-10-24 par Stubby AI (https://stubby-ai.com)

**En bref.** Tailoring your CV with generative AI has become second nature, but the tool has no idea what you actually did. It can inflate a verb, add a tool or round up a figure to match the job ad. Your CV then gets through the ATS filter, only to fall apart at the recruiter's first question in the interview. According to stubby-ai.com, a CV can be too well optimised: the more closely a line matches the job ad, the more it will be questioned. In this article, you will discover the 4 types of indefensible lines (the bolted-on keyword, the oversized verb, the figure you cannot reconstruct, the team result presented as individual), the 5-step defensibility test, three copy-and-paste prompts for using AI as a sceptical recruiter, and a checklist to run through before every application.

## AI-optimised CV: the scene nobody tells you about

Thomas applies for a project manager role. The day before, his AI assistant suggested adding "managing data projects" to match the job ad. His AI-optimised CV gets through the filter, and he lands the interview.

Ten minutes in, the recruiter asks him: "What was the budget? How many people? Which tool?" Thomas stumbles. In reality, he built Power BI dashboards for a project led by his manager.

He did not lie deliberately. He let a phrase that was too big for him settle onto his CV. At the first follow-up question, it turned against him.

### Why the launch of Careerz Copilot is a signal

A new tool, Careerz Copilot, has just launched with an unusual promise. It does not promise to "beat the ATS", but to help build applications the candidate can defend. Its bet: consistency between real experience, the CV and the job ad.

I have not yet been able to test the tool properly, so I will not comment on its performance. But its positioning says something important after years of chasing keywords. The challenge is no longer being read by the machine; it is holding your own in front of a human.

## ATS in 2026: why getting through the filter is no longer enough

Tailoring your CV to a job ad with generative AI has become second nature. The tool rephrases, generalises and embellishes to maximise the match. Yet it has no way of knowing what you actually did.

[France Travail](https://www.francetravail.fr/candidat/vos-recherches/bien-vous-organiser/comment-utiliser-l'ia-pour-optim.html) (the French public employment service) says so plainly in its guide on AI. A tool can "hallucinate" and introduce a skill simply because it appears in the job ad. The advice is clear: check that the changes "remain faithful to your own experience and know-how".

So the automated filter approves lines you have sometimes never checked. Then comes the job interview, which works as a **consistency test**. The recruiter digs into precisely the items most aligned with the job ad, since they are what justify your being there.

Here is the paradox: a CV can be *too well* optimised. The more closely a line matches the job ad, the more it will be questioned. If it is inflated, it is the line that will make you fail.

The [AI Agents Simplified](https://aiagentssimplified.substack.com/p/prep-for-your-next-interview-with) blog suggests a very sound approach. An AI first reviews the CV to spot the claims a recruiter is likely to question. Consistency between CV and interview then becomes the real criterion: what is written must be something you can talk about.

## The 4 types of indefensible lines (and how to rewrite them)

The examples below are fictional, but based on very common situations. In each case, the "after" version stays aligned with the job ad. It simply becomes defensible.

### 1. The bolted-on keyword (developer)

A tool or method mentioned when you have only used it superficially.

- **Before:** "Expert in Docker, Kubernetes and CI/CD pipelines."
- **After:** "Daily use of Docker in development; involvement in Kubernetes deployments managed by the infrastructure team."

Kubernetes is still there for the ATS. But you will no longer be caught off guard by "How did you configure your cluster?".

### 2. The oversized verb (communications officer)

"Led", "directed" or "designed" instead of "contributed", "took part" or "coordinated".

- **Before:** "Led the brand's social media strategy."
- **After:** "Coordinated the LinkedIn and Instagram editorial calendar, working with the head of communications."

The second version is more precise, and therefore more credible. Above all, it opens the door to questions you know how to answer.

### 3. The figure you cannot reconstruct (sales)

A percentage or amount whose source and calculation you cannot explain.

- **Before:** "+35% revenue."
- **After:** "Portfolio of 40 SME accounts: revenue grew from €410k to €520k between 2023 and 2024 (CRM data)."

The first figure immediately prompts "compared with what?". The second answers that before the question is even asked.

### 4. The team result presented as individual (administrative assistant)

A team result attributed to you alone.

- **Before:** "Reduced supplier payment times by 20%."
- **After:** "Invoice digitisation project (team of 4): responsible for matching purchase orders and invoices, helping to cut payment times by 20% across the department."

The result is still visible. Your personal role becomes explicit, and that is what the recruiter will explore.

## The 5-step defensibility test

The idea comes from a question raised by the [University of Arizona's Center for Career Readiness](https://career.arizona.edu/resources/using-ai-tools-in-your-career-development/). Before sending a document prepared with AI, ask yourself whether you can explain and defend every line in an interview. In practice: **"If the recruiter asks me to prove it, can I explain this line without improvising?"**

### Step 1: cross-check each line against the job ad

Highlight the lines in your CV that directly address the requirements in the ad. These are your high-risk lines. They will be probed first.

### Step 2: apply the 3 follow-up questions test

For each highlighted line, answer three questions out loud:

1. **"How?"** The steps, the tools, your decisions.
2. **"How much, and with whom?"** The volume, the budget, the team size, your place in it.
3. **"What would you do differently?"** The question that exposes those who were not in charge.

If you hesitate on even one of the three, the line is fragile.

### Step 3: rebuild the evidence

[JobShinobi](https://www.jobshinobi.com/blog/ai-resume-builder-hallucinations-how-to-fact-check) offers a very practical method. Create a reference document containing only verified facts. Then extract every claim from your CV into a checking table.

For each figure, note its source, its time period and how it was calculated. For each achievement: the context, your personal action, the result, the deliverable. No evidence you can rebuild, no line.

### Step 4: recalibrate the verb

Choose the verb that matches your real role, not the one described in the job ad. Here is a simple scale, from most modest to strongest:

- **Took part in**: you were part of the project.
- **Contributed to**: you delivered an identifiable part.
- **Took charge of**: an entire strand was entrusted to you.
- **Coordinated**: you organised the work of several people.
- **Led / directed**: you owned the decisions, the budget or the results.

An accurate verb is not a weak verb. It is a verb that does not collapse at the second question.

### Step 5: own the gaps

A partial skill is not a flaw to hide. Turn it into an honest area for development: "Working knowledge of SQL (simple queries), currently building on it". Then prepare your spoken answer: what you can do, what you are learning, and how.

JobShinobi in fact recommends doing this defensibility check *before* checking ATS compatibility. I fully agree: substance first, then form.

## Using AI as a challenger, not just a writer

The AI that inflated your CV can also help you deflate it. You just need to change its role. France Travail presents these tools as able to generate questions and mock interviews, without replacing your experience or your judgement.

Three prompts to copy and paste:

```
Play a sceptical recruiter. For each line of my CV related to this job ad, ask me three precise follow-up questions.
```

```
Find every claim in my CV that a recruiter might doubt: figures without a source, strong verbs, tools mentioned without context.
```

```
Do not add any new facts. Rewrite my CV for this job ad using only the information in my reference document.
```

Two precautions, highlighted by the University of Arizona: do not share any confidential data, and respect your employer's rules. Anonymise client names, sensitive figures and internal projects.

Your interview preparation really starts here: answer the questions *yourself*, out loud. [Interview answers that are too perfect get you rejected](/en/blog/simulation-entretien-ia-2026-reponses-parfaites-recaler) just as surely as an inflated CV. To stay genuine from CV to interview, this [strategy for using AI without sounding robotic](/en/blog/ia-sans-paraitre-deshumanise-2026-strategie-entretiens) complements the method well.

This is also where a specialist tool saves time. The [Stubby](https://stubby-ai.com) analysis report cross-checks your CV against each job ad and separates your genuine strengths from your areas for improvement. You know which matches are solid and which need preparation, before you even apply.

## Your checklist before every application

The best CV is not the one that matches the job ad most closely. It is the one where **every line stands up to the question "prove it"**.

- [ ] I have identified the lines most aligned with the job ad.
- [ ] Each one passes the 3 follow-up questions test, out loud.
- [ ] Every figure has a source, a time period and a calculation.
- [ ] Every verb reflects my real role.
- [ ] Team results specify my contribution.
- [ ] The tools mentioned match a use I can describe.
- [ ] Gaps are owned, with a spoken answer ready.
- [ ] No confidential data has been shared with the AI.

An AI-optimised CV is still an excellent idea, as long as you can defend it line by line. To find out which lines to prepare first, try Stubby: the CV/job ad compatibility analysis shows where you are strong and what to work on before the interview.

## Synthèse

| Section | Key messages |
| --- | --- |
| The trap | A line inflated by AI gets through the ATS but collapses at the recruiter's first follow-up question. |
| The paradox | The more closely a line matches the job ad, the more it will be probed in the interview: the interview is a consistency test. |
| 4 indefensible lines | Bolted-on keyword, oversized verb, figure you cannot reconstruct, team result presented as individual. |
| Defensibility test | 1. Cross-check with the job ad. 2. 3 follow-up questions test. 3. Rebuild the evidence. 4. Recalibrate the verb. 5. Own the gaps. |
| Verb scale | Took part, contributed, took charge, coordinated, led or directed: choose according to your real role. |
| AI as challenger | Ask the AI to play a sceptical recruiter, spot fragile claims and add no new facts. |
| Precautions | Check every fact, anonymise confidential data, answer the questions yourself out loud. |
| Key takeaway | The best CV is not the one most aligned with the job ad, but the one where every line stands up to the question "prove it". |

---

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/
