Today's announcement by Champion's of Change Coalition of an Inclusive AI Pledge and Prompt caught my attention.
On face value, it is a good idea. As my reader's know, I am all for inclusive AI, completing Oxford University's AI Ethics program last year and founding ada.ai to bring together leaders committed to inclusive and ethical AI.
But is it an effective prompt to achieve it's goals? In particular, how would it account for intersectional issues? Does the term equality as opposed to equity pose a problem? What criteria would the tool use to evaluate effectiveness? As I was pondering these concerns and others (as the human-in-the-loop), I thought to myself… why not ask the tools themselves to test the prompt's effectiveness.
So I asked ChatGPT and Claude to critically evaluate the prompt against it's goals. Below are the results and my human-in-the-loop conclusion.
ChatGPT
Prompt
"Critically evaluate the effectiveness of this prompt for ensuring equality, inclusion, fairness, safety and respect (including intersectionality) are built into outputs using ChatGPT “I want your help to ensure this [project, communication, decision or design of policy, product, service or system] builds inclusive gender equality into the future…”
Response
The prompt is directionally strong, but in its current form it is too broad, underspecified, and aspirational to reliably produce consistently equitable, intersectional, safe, and bias-aware outputs from ChatGPT.
Overall Assessment
Strengths
Your draft prompt:
“I want your help to ensure this [project, communication, decision or design of policy, product, service or system] builds inclusive gender equality into the future…”
has several important advantages:
- It establishes a normative intent upfront.
- It frames inclusion and equality as design principles rather than afterthoughts.
- It encourages forward-looking systems thinking (“into the future”).
- It broadens the scope beyond communications to policies, systems, products and decisions.
- It signals that gender equality should be embedded, not merely represented symbolically.
This is already significantly more sophisticated than prompts that simply ask ChatGPT to “make this inclusive.”
However, for high-quality outputs — especially in organisational, policy, leadership, HR, product, or systems contexts — the prompt alone is unlikely to produce robust inclusion outcomes consistently.
Critical Limitations
1. “Inclusive gender equality” is conceptually vague
The phrase sounds positive, but AI systems require specificity.
The model may interpret this as:
- gender-balanced language,
- women’s representation,
- avoidance of sexism,
- inclusive wording,
- or generic DEI messaging.
It does not necessarily trigger:
- structural equity analysis,
- power analysis,
- intersectionality,
- disparate impact assessment,
- accessibility considerations,
- psychological safety,
- cultural inclusion,
- anti-racism,
- socioeconomic equity,
- LGBTQIA+ inclusion,
- disability inclusion,
- or safeguards against exclusionary implementation.
Risk
Outputs may become:
- superficial,
- performative,
- language-focused rather than systems-focused,
- or overly centred on gender binaries.
2. Intersectionality is not explicit
One of the largest weaknesses is the absence of intersectionality.
Without explicit instruction, AI often defaults to:
- white women,
- corporate/professional contexts,
- Western norms,
- able-bodied assumptions,
- heteronormativity,
- and majority-culture experiences.
Example failure modes
The output may:
- improve gender representation while worsening racial inequity,
- ignore disability impacts,
- overlook caregiving inequities,
- reinforce class exclusion,
- neglect First Nations perspectives,
- fail to consider trans and non-binary inclusion,
- or assume universal access and safety.
Intersectionality should never be implied if it is mission-critical.
3. The prompt lacks operational criteria
The prompt asks for a goal but not a method.
ChatGPT performs substantially better when prompted to:
- evaluate,
- identify risks,
- test assumptions,
- assess impacts,
- challenge bias,
- compare stakeholder experiences,
- and propose mitigations.
Your current wording does not instruct the model to do these things.
Consequence
The AI may generate:
- values statements,
- polished inclusive language,
- or high-level recommendations,
without interrogating:
- who benefits,
- who may be excluded,
- unintended harms,
- implementation inequities,
- accessibility barriers,
- or power asymmetries.
4. It does not specify fairness or safety dimensions
“Equality” alone is insufficient.
AI systems often need explicit prompting around:
- procedural fairness,
- distributive fairness,
- psychological safety,
- cultural safety,
- safeguarding,
- trauma-informed approaches,
- dignity,
- bias mitigation,
- and harm prevention.
Otherwise outputs may unintentionally:
- reinforce stereotypes,
- privilege dominant groups,
- or optimise efficiency over equity.
5. The prompt does not require critique
The wording subtly encourages affirmative collaboration:
“help ensure this builds…”
This can produce supportive rather than critically evaluative outputs.
The model may avoid identifying:
- exclusionary assumptions,
- structural inequities,
- political risks,
- harmful trade-offs,
- or performative inclusion.
For serious DEI work, prompts should explicitly authorise challenge and critique.
Recommended Improved Prompt
Here is a significantly stronger version:
“I want you to critically evaluate and strengthen this [project, communication, policy, decision, product, service or system] to ensure it promotes equality, inclusion, fairness, psychological safety, dignity and respect.
This version is substantially more reliable because it:
- operationalises inclusion,
- requests critique,
- invokes systems thinking,
- explicitly requires intersectionality,
- prompts risk analysis,
- and demands practical improvements.
Additional Enhancements for Different Use Cases
For HR / Workplace Contexts
Add:
“Consider impacts on recruitment, progression, flexibility, pay equity, leadership access, psychological safety, speaking-up culture and workplace power dynamics.”
For Product / Service Design
Add:
“Assess usability, accessibility, safety, representation, data bias, digital exclusion and differential user experiences across diverse groups.”
For Policy
Add:
“Consider implementation risks, enforcement inequities, unintended consequences and who may be disproportionately burdened or excluded.”
For Communications
Add:
“Assess tone, framing, stereotypes, representation, accessibility, readability and the likelihood different audiences will feel respected, safe and included.”
Important Reality Check About AI Limitations
Even excellent prompts cannot fully guarantee equitable outputs.
ChatGPT can still:
- reflect societal bias in training data,
- overgeneralise,
- produce overly Westernised inclusion approaches,
- flatten lived experience complexity,
- miss context-specific cultural dynamics,
- or provide generic recommendations.
Therefore, high-stakes inclusion work still requires:
- human review,
- lived experience input,
- stakeholder consultation,
- and governance processes.
The prompt should be treated as:
- a structured inclusion-thinking accelerator, not
- a substitute for equity expertise.
Final Evaluation – Effectiveness Rating (Current Prompt)
Signals inclusive intent—Strong
Encourages systems thinking—Moderate
Ensures intersectionality—Weak
Surfaces bias/exclusion risks—Weak
Produces actionable equity analysis—Moderate–Weak
Encourages critical challenge—Weak
Supports fairness/safety analysis—Moderate
Reliability across contexts—Moderate
Overall
6/10 as currently written
With the recommended revisions
8.5–9/10 for general organisational use
The strongest improvement you can make is moving from:
“help make this inclusive”
to:
“critically evaluate this for differential impact, exclusion, power imbalance, harm and inequity across intersecting identities, then improve it.”
Claude
Prompt
_"_Critically evaluate the effectiveness of this prompt for ensuring equality, inclusion, fairness, safety and respect (including intersectionality) are built into outputs using Claude “I want your help to ensure this [project, communication, decision or design of policy, product, service or system] builds inclusive gender equality into the future…”
13:59
Examined prompt effectiveness across equality, inclusion, and intersectionality dimensions
Critical Evaluation of the Prompt
"I want your help to ensure this [project, communication, decision or design of policy, product, service or system] builds inclusive gender equality into the future…"
What the Prompt Does Well
Breadth of application — The bracketed options signal the prompt is designed as a reusable template across contexts, which is practical. Covering policy, product, service and system design reflects awareness that inequality is structural, not just interpersonal.
Forward orientation — "into the future" gestures toward proactive design rather than just auditing existing harm, which is a stronger framing for embedding values early.
"Inclusive gender equality" — The pairing of "inclusive" with "gender equality" is a subtle but meaningful move, implying gender isn't a binary category to be balanced but a spectrum to be included.
Significant Weaknesses
1. Intersectionality is absent — critically so This is the most serious gap. The prompt centres gender without acknowledging that gender inequality is compounded by race, disability, class, age, sexuality, caste, migration status and more. A prompt framed around gender alone risks producing outputs that centre the experiences of white, able-bodied, cisgender women while treating other identities as secondary or invisible. For a prompt explicitly aiming at inclusion, this is a foundational contradiction.
2. "Equality" vs "equity" is unresolved Equality (same treatment) and equity (proportionate treatment to achieve fair outcomes) produce very different designs. The prompt doesn't signal which is intended — or that the distinction matters. Claude could default to the more superficial reading.
3. Safety is entirely missing There is no instruction to assess harm, risk, or potential for the output to endanger people — particularly those who face gendered violence, harassment, or discrimination. A policy or product designed with "gender equality" as the only lens could still be actively unsafe for trans people, women in conflict zones, or people in abusive situations.
4. The ellipsis does a lot of unspecified work The prompt ends with "…" — meaning the actual task, context, criteria for success, and constraints are entirely undefined. Claude has no basis for knowing what a good output looks like, what trade-offs to flag, or how deep the analysis should go. This risks producing a polite, surface-level response rather than a rigorous one.
5. No instruction to challenge or flag The prompt asks Claude to ensure inclusion rather than to critically examine whether inclusion is achievable in the current design. This positions Claude as an implementer, not a critical evaluator — which limits its usefulness for genuinely surfacing problems.
6. Whose definitions are centred? "Gender equality" is a contested concept across cultures, legal systems, and political frameworks. The prompt doesn't ask Claude to surface those tensions or acknowledge whose framework it's applying. This could produce outputs that impose a particular (often Western, liberal) conception of gender equality onto contexts where different frameworks apply.
7. No mention of power Inclusion and equality work that ignores power dynamics tends to produce tokenistic outputs. The prompt doesn't ask Claude to examine who benefits, who makes decisions, who is consulted, or whose interests might be structurally deprioritised.
8. Respect and fairness are absent Despite the user's stated goals, neither "respect," "fairness," nor "dignity" appear in the prompt itself — so Claude has no instruction to prioritise them as values.
Recommendations for a Stronger Prompt
A more effective version might read something like:
"Critically evaluate this [project/policy/product/system/communication] for how well it achieves equity, safety, dignity and respect for all people. Specifically: (1) Apply an intersectional lens — examine impacts across gender, race, disability, class, sexuality, age and other relevant axes of identity, including where these overlap. (2) Distinguish between equality of treatment and equity of outcomes. (3) Identify who is centred and who is marginalised in the current design. (4) Flag any safety risks, particularly for people facing discrimination or violence. (5) Surface power imbalances in who designs, decides and benefits. (6) Where genuine inclusion is not yet achieved, say so directly and suggest concrete changes."
Summary Verdict
The original prompt is a useful starting point but is too narrow, too vague, and too passive for the full scope of work you've described. It risks producing outputs that are gender-aware but intersectionally blind, aspirational but not actionable, and collaborative rather than critically interrogative. The stakes of equity work are high enough that the prompt needs to do heavier lifting.
Without intersectionality, a distinction between equality and equity, safety considerations, or concrete task framing, it is likely to produce outputs that are inclusive in tone but not in rigour. For high-stakes use — policy, product design, institutional decisions — it would need substantial strengthening.
My human-in-the-loop conclusions (after reviewing and critically examining the output above against my DEI and AI ethics expertise):
-
AI has value and risks for inclusion, but when it comes to inclusive AI, you can't outsource it to a prompt. Human-in-the-loop is critical. And don't forget to spar with your preferred GenAI tool.
-
Inclusive and ethical AI requires an multidisciplinary approach. DEI, tech, and governance need to start talking to each other.
Felicity Menzies is CEO and Principal Consultant at Culture Plus. Over 15 years, she has worked with more than 35,000 leaders across government, corporate and global organisations on workplace culture, DEI, Respect at Work, and psychosocial safety — often through multi-year partnerships supporting sustained reform. A Fellow of Chartered Accountants Australia and New Zealand, a University Medallist in Psychology, an accredited Cultural Intelligence (CQ) facilitator, and an Oxford-trained Certified AI Ethicist, Felicity is also the author of A World of Difference: Leading in Global Markets with Cultural Intelligence, contributor to leading business and HR publications globally, and a lead facilitator for the Australian HR Institute (AHRI)'s DEI programs. Her evidence-based, trauma-informed approach has been commended in parliamentary settings and underpins her track record of cultural reform in high-scrutiny environments.