Toto-Rex's story
Toto-Rex submits a CV at 9:00.
At 9:04, an email arrives.
'Dear [Name], we are sorry to inform you that we will not be progressing your application.'
Four minutes. And [Name].
Toto-Rex asks: was that an AI, or just a template?
Toto-Rex's story is made up. The cases and studies in this video are real.
The short answer
Possibly. A quick rejection alone does not tell you whether AI was involved.
The details
Screening tools can use fixed rules, AI, or both.
Different studies look at different tools, settings, and outcomes. Here are four.
Study 1: a regulator's audit.
The ICO, the UK's data protection regulator, audited several AI recruitment tool providers. It published its findings in November 2024.
Some tools let recruiters filter out candidates with certain protected characteristics.
Others inferred candidates' gender and ethnicity from their names.
The ICO made almost 300 recommendations. The providers accepted or partly accepted them.
Study 2: a simulated test of AI models.
University of Washington researchers tested three text-embedding models (AI that turns text into numbers so it can be compared) in simulated CV screening, changing only the names.
In the main race tests, white-associated names were significantly favoured in about 85% of test settings.
That is a share of research tests, not a hiring rate.
When the researchers used names with roughly equal frequency in a large text corpus, the result flipped: Black-associated names were favoured in about half the tests (and white-associated names in about a fifth).
The study did not test any employer's real hiring decisions.
Study 3: a field experiment at one company.
A 2020 working paper studied software-engineer hiring at one company.
Candidates picked by an algorithm but missed by human screeners were about 14% more likely to pass interviews and get an offer than candidates selected by both the algorithm and human screeners.
The algorithm also increased hiring of non-traditional candidates, including women and racial minorities.
The same paper warns that algorithms can copy past bias if the human decisions they learn from were consistently biased.
Study 4: a real-world application test.
A study published in 2022 sent over 83,000 fictitious applications to 108 large US employers.
Applications with distinctively Black names were 2.1 percentage points less likely to receive employer contact than those with distinctively white names.
That study did not separate the roles of human reviewers and automated systems.
Separately, Workday is defending a US age-discrimination collective action about its screening tools. The claims are unproven, and Workday denies them.
Useful to know: when you hear a claim about AI screening, ask what was tested, who it was compared with, and whose outcomes changed.
Real-life examples
Real life: a rule, not a robot.
In 2023, iTutorGroup agreed to pay 365,000 US dollars to settle an age discrimination lawsuit brought by the US Equal Employment Opportunity Commission.
According to the lawsuit, the company programmed its tutor application software to automatically reject women aged 55 or older and men aged 60 or older.
The agency said more than 200 qualified applicants in the US were rejected because of their age.
According to a law firm's summary of the complaint, one applicant was offered an interview after reapplying with a more recent birth date and an otherwise identical application.
iTutorGroup denied wrongdoing.
It also agreed not to ask tutor applicants for their birth dates before making a job offer.
Real life: the AI interviewer.
In January 2021, HireVue, a video interview company, published an outside audit of one of its products.
It said HireVue no longer built visual features, such as facial movements, into new models.
Those features were being removed from existing models as each came up for its annual review.
HireVue said advances in analysing language meant visual analysis no longer added much value.
The privacy group EPIC had filed a complaint about HireVue with the US Federal Trade Commission about a year earlier.
HireVue continued to analyse candidates' spoken answers.
The auditors said their review covered one use case, not all of HireVue's algorithms, and that they had not independently checked HireVue's fixes.
Real life: the Workday case so far.
In May 2025, a US judge let the age claim go ahead as a nationwide collective action for applicants aged 40 and over.
That is an early procedural step, not a finding that Workday discriminated.
As of October 2026, the case was still going: more than 15,000 people had opted in, and the plaintiffs had asked the court to certify further classes.
Back to Toto-Rex
So, back to Toto-Rex's four-minute email.
The email cannot tell us how the decision was made.
It could be a template sent by a person, a fixed rule like iTutorGroup's age filter, or an AI tool.
Toto-Rex can ask the employer whether automated tools were used, and ask for a copy of personal data.
In the UK, if the rejection was a significant decision made solely by automated means, the employer must give information about it.
Summary
In short: a quick rejection does not tell you whether AI was involved.
The four studies tested different things.
Some tools filtered by protected characteristics. Some AI models favoured certain names.
One company's algorithm found candidates human screeners had missed, and a field study found names still affected employer contact.
Three key terms
Inferred data
Information derived about you from other data.
For example: estimating your ethnicity from your name.
The inference may be wrong.
Disparate impact
A US legal concept: a practice that seems neutral but disproportionately disadvantages a protected group.
A claim does not require proof of discriminatory intent.
Whether the practice is unlawful depends on the legal test and the available defences.
Correspondence study
Researchers send fictitious applications to real job vacancies.
They systematically vary details, such as names, and compare employer responses.
Further reading
Start with the ICO's audit report: what the regulator found and recommended.
The University of Washington paper includes the authors' correction of 29 August 2026.
Then the Cowgill working paper and the Kline, Rose, and Walters study.
For the iTutorGroup case, see the EEOC's press release of 11 September 2023.
Sources
ICO, AI tools in recruitment: audit outcomes report, November 2024. UK GDPR, Article 22C, inserted by the Data (Use and Access) Act 2025, section 80. Wilson and Caliskan, 'Gender, Race, and Intersectional Bias in Resume Screening via Language Model Retrieval', AIES, 2024, arXiv version 3 with correction dated 29 August 2026.
Cowgill, 'Bias and Productivity in Humans and Algorithms: Theory and Evidence from Résumé Screening', working paper, 21 March 2020. Kline, Rose, and Walters, 'Systemic Discrimination Among Large U.S. Employers', Quarterly Journal of Economics, 2022. Mobley v. Workday, Inc., No. 3:23-cv-00770 (N.D. Cal.).
EEOC, 'iTutorGroup to Pay $365,000 to Settle EEOC Discriminatory Hiring Suit', 11 September 2023. Greenberg Traurig, 'EEOC Secures First Workplace Artificial Intelligence Settlement', August 2023. EPIC, 'HireVue, Facing FTC Complaint From EPIC, Halts Use of Facial Recognition', 12 January 2021. ORCAA, 'Description of Algorithmic Audit: Pre-built Assessments', 15 December 2020, published January 2021.
Bloomberg Law, 'Workday AI Bias Suit to Go Forward as Age Claim Class Action', 16 May 2025. Mobley v. Workday, discovery order, 6 October 2026. EEOC v. iTutorGroup, consent decree.