Methodology

How this tool works

A plain-English guide to what the Neuromine job advert analyser does, why it does it, and what it can't tell you.

What this tool is for

This is a free tool built for candidates — particularly (though not exclusively) neurodivergent candidates — who are deciding whether to apply to a job, how to tailor their application, and what to ask in an interview.

It's not a tool for grading employers. There is no inclusivity score the employer could defend or game. What you get instead is a qualitative reading of the advert, with specific guidance on how to weigh what's in front of you.

We built it because most existing tools in this space serve employers — helping them write more inclusive ads. Nothing equivalent existed for candidates, who often have to read between the lines of an advert without a translator. This is that translator.

What we believe

A few principles shape every part of how the tool works.

The candidate is the user. Every output is framed around your decision: should you apply, how should you apply, what should you ask. We don't speak to employers in this tool. We don't tell you what an employer "should" do. We tell you what we see in their advert and what that might mean for you.

A flag is a signal, not a verdict. When we highlight language in an advert — "this looks gendered," "this looks ableist," "this is vague" — we're saying the language pattern matches something research has found can deter applicants. We're not saying the employer is biased. The same words might mean a clumsy template, an outdated style guide, or a real cultural problem. The tool tells you what to ask to find out which.

Evidence first. Every category we flag is anchored to published research, UK regulatory guidance, or recognised practitioner frameworks. Where the evidence is strong, we say so. Where it's weaker, we say that too.

Apply anyway, more often than not. The strongest single finding in the literature on job adverts is that vague, intimidating, or overly demanding language causes qualified candidates — particularly from underrepresented groups — to take themselves out of the running before they apply. Our default posture is to help you apply better, not to talk you out of applying.

"Neuroinclusive" is a label we don't give lightly. Many tools call any inclusion signal "neuroinclusive." We don't. We reserve that label for the specific markers — interview questions in advance, alternative application formats, sensory considerations, named adjustment processes — that current neurodiversity research and UK guidance (CIPD, Acas, the Buckland Review) actually identifies as helpful. Pay transparency is great, but it isn't neuroinclusion. We label it as what it is: general inclusivity.

What we look for

The tool reads an advert and looks for three kinds of things: language that research links to bias, language that's vague in ways that disproportionately affect applicants, and concrete signals that the employer has thought about inclusion.

Language linked to bias

We check for patterns in these categories. Each one is grounded in published research:

  • Gendered language. Words and phrases research has shown to deter applications from people whose self-identity is less aligned with stereotypically masculine traits.
  • Age-coded language. Phrases that signal a preference for younger or older candidates, sometimes unintentionally.
  • Ableist language. Words used as metaphors that can be exclusionary, or physical/cognitive criteria not actually required by the role.
  • Class-coded language. Phrases that filter implicitly for particular social backgrounds.
  • Heteronormative language. Gendered family or relationship assumptions in role descriptions or benefits.
  • Culture-fit language. "Good fit," "team fit," "our tribe" — phrasing that research links to affinity bias in hiring.
  • Gendered job titles. "Salesman," "foreman," "waitress" — flagged as potential discrimination under UK law.
  • Pre-employment health questions. Questions about disability, illness, or sickness absence in job ads — generally unlawful before an offer under section 60 of the Equality Act 2010.
  • Intensity-culture language. "Fast-paced," "high-pressure," "wear many hats," "thrive in ambiguity" — phrasing that signals a demanding environment without describing what the workload actually looks like. Often filters out candidates who need predictability, including many neurodivergent applicants.
  • Hero-leadership language. "Rockstar," "10x," "own everything," "wear the cape" — framing that treats individual heroics as the norm. Research links this to exclusionary team cultures and to bias against candidates who work collaboratively or need clear scope.
  • Meritocracy framing. "Pure meritocracy," "results speak for themselves," "we only hire the best" — claims that paradoxically correlate with more biased hiring decisions in published research, because they suppress the reflection that catches bias.
  • Assessment accessibility. Timed tests, whiteboard coding, group exercises, or unspecified assessment formats — features that disadvantage candidates with processing differences, anxiety, or sensory needs unless adjustments are offered up front.

Vague or missing information

Some of the most powerful effects in this research come not from biased language, but from unclear language that causes qualified candidates to self-screen out.

  • Vague qualification requirements. "Several years of experience," "demonstrated excellence," "strong background" — without specifics. Research shows this disproportionately deters women and underrepresented candidates from applying.
  • Inflated requirements. Long must-have lists, arbitrary numeric experience thresholds, degree requirements that aren't justified by the actual job tasks.
  • Vague role descriptions. Phrases that could mean very different things ("lead the delivery and growth," "set best practice"). We don't flag these as bias — we surface them as things to ask about, with multiple possible interpretations.
  • Hidden or implied requirements. Expectations the advert doesn't state directly but clearly assumes.

Signals of inclusion practice

On the positive side, we look for concrete things — not boilerplate diversity statements, but specifics:

  • General inclusivity: pay transparency, flexible working, hybrid/remote specifics, enhanced parental leave, enhanced holiday schemes, Friday early finishes and other schedule-clarity signals, employee resource groups, named recruitment partnerships.
  • Neuroinclusion (tightly scoped): interview questions provided in advance, alternative application formats, strengths-based assessment, sensory considerations, explicit adjustment process descriptions.
  • Disability inclusion: Disability Confident scheme (with level), guaranteed interview scheme, Access to Work mentions, BSL or accessible formats, named partner organisations.
  • Process transparency: stages described, interview formats, decision timelines, named contacts.

The four bands

We summarise an advert with one of four bands, chosen to support your triage across a job search:

  • Worth considering — No serious concerns. Specific qualification language. Pay disclosed. Some signs the employer has thought about inclusion.
  • Worth considering with caveats — A few medium-strength signals worth weighing, or notable absence of inclusion markers, but nothing that should put you off applying.
  • Significant signals to weigh — Multiple medium signals, or one high-severity signal, or a clear absence of inclusion markers across several dimensions. Apply if it's a strong fit and you're ready to ask hard questions early.
  • Major concerns — Legal-risk patterns (potential Equality Act issues), gendered job titles, or multiple high-severity signals. Still your decision, but worth investigating carefully before investing time.

The band is a triage signal, not a recommendation. A "Major concerns" advert might still be the right job for you. A "Worth considering" advert might still be the wrong employer. The band tells you how much attention the detail deserves, not whether to apply.

The band is computed deterministically — given the same flag pattern, you get the same band, every time. We don't ask an AI to decide your band based on vibes.

How detection actually works (in general terms)

We use two kinds of detection, working together:

Deterministic detection is the backbone. We use word lists, phrase patterns, and structural rules drawn from published research and recognised practitioner sources. Gendered language, age-coded language, ableist and class-coded terms, gendered job titles, pre-employment health questions, pay disclosure, and the inclusion signals above are all detected deterministically. When a deterministic detector fires a flag, that flag is stable — re-running the same advert produces the same result.

LLM-augmented explanation sits on top. Once a flag has been deterministically detected, we use AI to generate the explanation, the suggested rewrite, and the contextual interpretation. The AI doesn't decide whether something is a flag; it explains what we've already detected.

This split matters because AI alone is unreliable for detection — it produces different results on different runs, which would undermine the tool's credibility. Deterministic detection makes the rubric auditable and stable. AI augmentation makes the explanations natural and contextual.

(We've published a detailed technical specification of this approach. If you're a researcher, academic, journalist, HR/DEI professional, or potential partner who'd like to see it, contact us and we'll send it.)

The evidence behind the categories

A short, plain-English tour of the research we rely on most.

Gendered language in job ads. Danielle Gaucher and colleagues (2011, Journal of Personality and Social Psychology) showed that job adverts using masculine-coded language reduce women's interest in applying — even when the language has nothing obvious to do with gender. More recent work by He and colleagues (2025, PNAS) replicated this across four studies with nearly 38,000 participants and showed that replacing masculine words with neutral synonyms increases application rates from women and from men whose self-identities are less aligned with stereotypically masculine traits.

Vague qualification language. Katherine Coffman, Mark Collis, and colleagues (2024, Management Science) ran field experiments showing that when qualification language is vague, only about 42% of qualified women applied to a job, compared to 56% of men. When the same job was advertised with specific qualifications, women's application rate jumped to 62%. This is one of the strongest findings in the field, and it's why our default advice for vague-qualification flags is "apply anyway."

Reproducing labour-market segregation through language. Yuanyuan Hu and colleagues (2024, PNAS Nexus) analysed 28.6 million UK job ads from 2018–2023 and found distinct ways in which job-advert wording sustains both gender and racial segregation in the labour market — the most comprehensive UK study to date.

Cultural matching and affinity bias. Lauren Rivera (2012, American Sociological Review; 2015, Pedigree: How Elite Students Get Elite Jobs) showed how "cultural fit" works as a vehicle for affinity bias in elite hiring, filtering candidates by class, race, and gender rather than capability. This is why we flag "culture fit" language and recommend "culture add" as the contemporary alternative.

Neuroinclusion at work. Our neuroinclusion category draws on the CIPD's Neuroinclusion at Work report (2024), Acas guidance on neurodiversity at work (2025), the Buckland Review of Autism Employment (2024), and McDowall, Doyle and Kiseleva's Neurodiversity at Work: Demand, Supply and a Gap Analysis (Birkbeck, 2023).

UK regulatory framework. Our legal-risk flags reference the Equality Act 2010 (particularly section 60 on pre-employment health questions, Schedule 9 on lawful occupational requirements, and sections 158/159 on positive action), and the EHRC's updated 2024 guidance on discriminatory adverts.

Class background in hiring. Sam Friedman and Daniel Laurison's The Class Ceiling: Why It Pays to Be Privileged (2019) and Social Mobility Commission UK reports inform our class-coded category.

Age-inclusive recruitment. Centre for Ageing Better research and campaigns ("Happy to Talk Flexible Working") inform our age-coded detectors and our recognition of certain phrases as positive inclusion signals.

Intensity culture, hero framing, and meritocracy claims. Emilio Castilla and Stephen Benard's "paradox of meritocracy" work (2010, Administrative Science Quarterly) showed that organisations claiming to be pure meritocracies made more biased decisions, not fewer. Alongside that, the CIPD's Working Lives reports and Acas guidance on workplace wellbeing inform our intensity-culture and hero-leadership categories — both flagged because they correlate with burnout and with cultures that disadvantage neurodivergent, disabled, and caregiving candidates.

Assessment accessibility. The Buckland Review of Autism Employment (2024) and CIPD neuroinclusion guidance specifically identify timed tests, unspecified assessment formats, and group exercises as points where neurodivergent candidates are disadvantaged without proactive adjustments.

This isn't an exhaustive list — the full bibliography is in the detailed specification — but it's the spine of what the tool relies on.

What this tool is not

A few things we want to be clear about:

  • Not legal advice. We flag potential Equality Act issues to help you spot them, not to give you a legal position. If you believe an advert is unlawful, the EHRC, Acas, or a solicitor can advise you on what to do about it.
  • Not career advice. We don't tell you whether to take a job. We give you information to help you decide.
  • Not a verdict on the employer. An advert is a signal about how an employer has chosen to communicate. It isn't proof of what they're like to work for.
  • Not infallible. We use research-anchored detection, but no automated tool catches everything or interprets every context perfectly. Treat our output as informed input to your judgement, not a substitute for it.
  • Not a substitute for your knowledge of your own situation. Your priorities — sector, location, life circumstances, career stage, what you need from an employer — are yours to weigh. We don't know them.

What we don't yet do

We're being upfront about gaps:

  • English only, for now. Adverts in other languages aren't reliably analysed; we'd rather decline than mislead.
  • UK regulatory focus. Our legal-risk flags reference UK law. Adverts under other jurisdictions get the same detection but the legal commentary may not apply.
  • No certainty about employer culture. We read what the advert says. We don't see what's behind it.
  • AI-generated explanations can vary slightly between runs. The detection (whether a flag fires) is deterministic and stable. The wording of the explanation can vary because it's generated by AI — the substance stays the same, but the phrasing might differ.

How we keep this current

The research base in this field is moving. Lexicons change as language evolves. New evidence appears regularly.

We review and update the rubric on a planned cycle, and tag every analysis with the rubric version it was generated under. When we update, the previous behaviour stays available as a versioned reference. We publish a summary of changes when we update the rubric.

If you spot something we should be detecting and aren't, or something we're flagging that we shouldn't — tell us. The tool gets better when candidates tell us what we're missing.

Who built this

Neuromine was built by people who care about how neurodivergent candidates experience the job market, including some who've navigated it themselves. The full rubric is the work of practitioners drawing on published research, UK regulatory guidance, and recognised inclusive-recruitment frameworks.

If you're a researcher, academic, journalist, HR/DEI professional, or potential partner and you'd like to see the detailed methodology specification — including category definitions, detection criteria, severity logic, and the full evidence bibliography — get in touch and we'll share it.

A final note

This tool is meant to take some of the cognitive load off applying for jobs — particularly when adverts are vague, jargon-heavy, or quietly excluding. It can't read your situation or know what you want. But it can flag what's worth a second look, suggest what to ask, and remind you that more often than not, you should apply.

Who built this

Neuromine is built by Theo, who delivers neurodiversity speaking and training through Neurodiversity at Work. The rubric draws on that practice — what employers actually ask about, and what candidates repeatedly report — alongside the research cited above.