Blog 6 November 2026 · 6 min
Preparing for your interview with AI: when ChatGPT has you reciting facts about a company that no longer exists
More than one in two candidates use AI in their job search, including to research the company before an interview. But a generative model produces a probable answer, not a verified one: knowledge cut-off dates, hallucinations, companies with the same name and recycled marketing copy can have you mentioning a discontinued product or a director who has left. According to stubby-ai.com, a candidate's credibility depends on accuracy, not on the amount of information recited. In this article, you will find the 4 riskiest types of information to quote, a 5-step verification protocol, 3 prompts designed to limit fabrication and a checklist to tick off before the interview.
Preparing for your interview with AI: the scene nobody talks about
An imaginary scene, but an entirely plausible one. Léa decided to prepare for her interview with AI the night before. ChatGPT gave her a flawless summary: figures, directors, flagship products.
The next day, she opens with a smile: "Congratulations on the recent launch of your premium offer!" The recruiter pauses. "We discontinued it last year."
Silence falls. Léa didn't lie. She recited an answer she had never checked.
She is far from alone in using these tools. According to PeopleScout's Inside the Candidate Experience 2026 study, 52% of candidates surveyed worldwide use AI in their job search. Among the uses mentioned: optimising their CV, writing applications, researching companies and preparing for interviews.
That's the paradox. The tool meant to make you better prepared can make you less credible. The problem isn't AI: it's reciting without checking.
Why ChatGPT "gets the company wrong"
The CNIL (France's data protection authority) makes the point in its FAQ on generative AI: a generative model is not a knowledge base. It produces a probable answer, not a verified one. Four mechanisms explain most errors.
- The knowledge cut-off date. The model learned up to a certain date. Without web search, it knows nothing of the chief executive's departure or last spring's acquisition.
- Web search is no guarantee. When AI searches online, it selects its sources in an opaque way. An old blog post can carry as much weight as an official press release.
- Hallucinations. NIST calls it confabulation: false content presented with confidence. When information is missing, the model fills the gap with a plausible figure, name or date.
- Namesakes and groups. Two companies with the same name, a subsidiary mistaken for its parent company, a brand taken for a company. AI happily mixes them up.
- Recycled marketing copy. AI reproduces the values displayed on a careers page, sometimes an old one. It tells you nothing about what the team is really like.
Keep one useful distinction in mind. Outdated information was true at some point; hallucinated information never was. In an interview, both cost you exactly the same.
One last trap, and not the least. The NIST generative AI risk management framework points out that errors can affect the facts, but also the citations meant to back them up. A "sourced" answer is therefore not automatically reliable.
The 4 most dangerous types of information to quote without checking
Your recruiter knows their company better than any language model. Every error sends them a clear signal about the way you work.
1. Figures
- The risk: wrong turnover, headcount, funding round or growth. The person in front of you spots the error instantly.
- What they conclude: a lack of rigour. That's a deal-breaker if the role involves handling data.
2. Names
- The risk: a director who has left, a wrongly identified manager, an organisation chart two years out of date.
- What they conclude: superficial preparation. It's the equivalent of mispronouncing someone's name on a first date.
3. News
- The risk: a launch, acquisition or reorganisation that never happened, or is no longer happening. That was Léa's trap.
- What they conclude: you read a summary, not the company. In short, a dependence on AI.
4. Culture and values
- The risk: praising the "startup spirit" of a team in the middle of restructuring. The mismatch sounds hollow, even clumsy.
- What they conclude: a script pasted on. It's one of the pitfalls of overly perfect answers generated in interview simulations.
Researching a company before an interview: the 5-step protocol
Good news: five habits are enough to turn a shaky summary into solid preparation. Allow thirty to forty-five minutes.
Step 1: demand sources and their dates
Ask the AI where each piece of information comes from, and how old it is. Discard anything unsourced without hesitation.
Then actually open the links. A cited source is not a checked source, and the CNIL specifically recommends verifying accuracy before reusing anything.
Step 2: cross-check with recent primary sources
A primary source comes directly from the company. It's your reference point against any summary.
- The official website and its newsroom.
- The annual report, for figures.
- The company's LinkedIn page and recent posts by its directors.
- Job adverts published in recent months.
That last point is underrated. Current recruitment reveals present priorities better than any press release: learn to decode a job advert and its hidden clues.
Step 3: apply the 12-month rule
Any news more than a year old must be reconfirmed. A product, a director or a strategy can change within a quarter.
Step 4: turn the statement into a question
Replace "You launched X" with "I saw you were working on X, how is the project going?". If the information is wrong, you're protected: the recruiter corrects you, with no awkwardness.
If it's right, you open a real conversation. It's the best way to build your questions to ask the recruiter, the ones that make you stand out.
Step 5: build a one-page company sheet
One page, no more. For each line, note:
- the verified information;
- its source and date;
- your confidence level;
- its link to a specific responsibility in the job advert.
This is the traceability logic recommended by NIST, applied to your interview. Information that serves no responsibility of the role has no place on the sheet.
ChatGPT for job interviews: have it organise, not assert
Using ChatGPT for a job interview is still an excellent idea. Provided you give it what it genuinely does well:
- structuring information you provide;
- cross-referencing the job advert with your background;
- generating relevant questions;
- simulating exchanges so you can practise.
The difference fits in one sentence. An AI that asserts teaches you things you can't verify. An AI that organises works on documents you control.
Here are three "safe" prompts, designed to reduce fabrication.
Prompt 1: cross-reference the job advert and CV, without inventing anything
Here is the job advert and my CV. Cross-reference them using only these two documents.
List the 5 key responsibilities, my evidence for each, and my 3 weak points.
Do not add any outside information.
Prompt 2: dated research, with a confidence level
List the news about [company] over the last 12 months.
For each item: link to the source, publication date, confidence level (high, medium, low).
If you are not certain, write "unverified". Do not invent any figures or names.
State whether [company] belongs to a group or has namesakes.
Prompt 3: questions, never statements
Based on my verified company sheet below, suggest 6 open questions to ask the recruiter.
Each question must be linked to a responsibility in the job advert.
Phrase them as questions, never as statements.
Even with prompt 2, the rule doesn't change. The output remains a lead to follow up, to be confirmed at step 2 of the protocol.
Checklist before the interview
Checklist before the interview
- Every figure quoted comes from a dated primary source.
- The names of the director and my future manager are confirmed on LinkedIn or the official website.
- No news older than 12 months is quoted without reconfirmation.
- Subsidiary, parent company and namesakes are clearly distinguished.
- Uncertain information is rephrased as questions.
- My company sheet fits on one page and links each item to a responsibility.
- I have prepared 3 questions to ask the recruiter, based on verified facts.
Credibility comes down to accuracy
In an interview, a single false piece of information can wipe out ten minutes of good answers. Conversely, a precise, well-dated question shows genuine curiosity. That's exactly what recruiters are looking for.
That's the approach we champion at Stubby AI: preparing for your interview with AI based on the real job advert and your CV. Strengths, areas for improvement, likely questions, company information: everything starts from concrete documents, not assumptions.
Let's be honest: no tool replaces a glance at the newsroom the day before the interview. Keep that habit, whichever tool you choose. Prepare for your interview with AI, yes; trust it with your credibility without checking, never.
« Prepare for your interview with AI, yes; trust it with your credibility without checking, never. »
| Section | Key messages |
|---|---|
| The paradox | 52% of candidates use AI in their job search (PeopleScout). The tool meant to help you prepare better can damage your credibility if you recite without checking. |
| Why AI gets it wrong | Knowledge cut-off date, opaque web sources, hallucinations (confabulation), confusion between namesakes and subsidiaries, recycled marketing copy. |
| The 4 risky types of information | Figures, names, news, culture and values. Each error signals to the recruiter a lack of rigour, superficial preparation or a dependence on AI. |
| 5-step protocol | Demand sources and dates, cross-check with recent primary sources, apply the 12-month rule, turn the statement into a question, build a one-page sheet. |
| Using AI well | Have it organise your documents (job advert, CV, verified sheet) rather than assert facts. Demand links, dates and a confidence level. |
| Checklist | Sourced figures, confirmed names, news less than 12 months old, entities distinguished, uncertainties rephrased as questions. |
| Conclusion | Credibility comes down to accuracy. AI structures your preparation, but checking recent news remains an essential habit. |
