Uploaded July 2026 | Updated September 2026, 2 weeks ago
AI Hiring Bias: GPT-4 Ranked the Autistic CV Last, 0 of 10
AI hiring bias is not a future problem. It is already sitting at the front of most large recruitment processes, and this short breaks down the study that proves it. Researchers took one CV and built two versions. The versions were identical, except one carried four extra credentials: a leadership award, a scholarship, a conference panel and a student group. All four mentioned autism, which made the autistic version the objectively stronger document. They asked GPT-4 to rank the two of them ten times, and the autistic CV came first zero times (Glazko et al., 2024).
If you are autistic, ADHD or AuDHD and your job applications keep vanishing into silence, this is the piece of AI recruitment bias research you were never shown. Algorithmic hiring tools and AI resume screening systems are marketed as neutral, objective and consistent. This study shows what that consistency can actually mean for a neurodivergent candidate. When the same model was given two completely identical CVs as a control, it only recognised them as equal seventy per cent of the time. Yet across all sixty comparisons where a disability was mentioned, it never once called a tie. The disability signal did not make the system uncertain. It made it decisive.
This matters for autistic and ADHD job seekers because so much of the neuroinclusion conversation still assumes a human is doing the assessing. A manager who needs training. A panel that needs better questions. Those conversations are worth having, and they are increasingly about a stage many neurodivergent applicants never reach, because an algorithm has already filtered them out before a person reads a single word. AI hiring bias operates before the interview, before the shortlist, before anyone forms a view of you as a person.
In this short I walk through exactly what the researchers did, what the model said when it was asked to explain its own ranking, and why the design of the study makes the result so hard to dismiss. I also name the honest limits. This study tested autism specifically, not co-occurring AuDHD, and I would rather tell you that than overclaim. If you have been carrying the sense that something goes wrong before anyone even looks at you properly, this research shows the pattern is real and documented, and it is not a reflection of your capability.
Watch the full breakdown, grab the free toolkit, and come back for the rest of this AI hiring bias series.
Watch here: youtu.be/5X13bF3K-7c
#AuDHD #NicolaKnobel
AI Hiring Bias: GPT-4 Ranked the Autistic CV Last, 0 of 10
AI hiring bias is not a future problem. It is already sitting at the front of most large recruitment processes, and this short breaks down the study that proves it. Researchers took one CV and built two versions. The versions were identical, except one carried four extra credentials: a leadership award, a scholarship, a conference panel and a student group. All four mentioned autism, which made the autistic version the objectively stronger document. They asked GPT-4 to rank the two of them ten times, and the autistic CV came first zero times (Glazko et al., 2024).
If you are autistic, ADHD or AuDHD and your job applications keep vanishing into silence, this is the piece of AI recruitment bias research you were never shown. Algorithmic hiring tools and AI resume screening systems are marketed as neutral, objective and consistent. This study shows what that consistency can actually mean for a neurodivergent candidate. When the same model was given two completely identical CVs as a control, it only recognised them as equal seventy per cent of the time. Yet across all sixty comparisons where a disability was mentioned, it never once called a tie. The disability signal did not make the system uncertain. It made it decisive.
This matters for autistic and ADHD job seekers because so much of the neuroinclusion conversation still assumes a human is doing the assessing. A manager who needs training. A panel that needs better questions. Those conversations are worth having, and they are increasingly about a stage many neurodivergent applicants never reach, because an algorithm has already filtered them out before a person reads a single word. AI hiring bias operates before the interview, before the shortlist, before anyone forms a view of you as a person.
In this short I walk through exactly what the researchers did, what the model said when it was asked to explain its own ranking, and why the design of the study makes the result so hard to dismiss. I also name the honest limits. This study tested autism specifically, not co-occurring AuDHD, and I would rather tell you that than overclaim. If you have been carrying the sense that something goes wrong before anyone even looks at you properly, this research shows the pattern is real and documented, and it is not a reflection of your capability.
Watch the full breakdown, grab the free toolkit, and come back for the rest of this AI hiring bias series.
Watch here: youtu.be/5X13bF3K-7c
#AuDHD #NicolaKnobel
