Every Rapid Gender Analysis (RGA) begins at a desk.
Before anyone can say how a crisis affects women, men, girls, boys, and gender-diverse people differently, someone has to find the situation reports, check each figure against its source and date, look for sex-disaggregated data that often does not exist, and fit what they find into a standard structure.
That is true for the international analyst who deploys in week two and for the practitioner who has been on the ground since day one, including those who belong to the women-led organisations responding to the crisis.
Teaching that desk work used to require a full course of four to six weeks.
A typical curriculum opens with an introduction to the RGA, then moves through finding existing data in a secondary data review, collecting new primary data, analysing the gender data, writing the RGA, and a final week of peer review and sharing.
Apart from primary data collection, almost every week teaches work done at a desk.
Today we are releasing an AI skill that does that work.
The Rapid Gender Analysis Skill researches the emergency, assembles and verifies the evidence, and drafts the RGA in a standard structure.
It works for all languages.
Learn more: see examples of real-world RGAs produced with the Skill.
It can save time and money, and it can raise the quality of the review, because it applies the same verification rules to every source.
Practitioners keep the judgement, the relationships, the accountability, and the decisions about who receives resources and credit.
Above all, they get back time for the work no AI model can do: engaging affected people, the groups most exposed to harm, and the women-led groups within those communities.
We built the skill on a decolonial critique of the RGA
An RGA exists so that aid workers can understand quickly how a crisis affects people differently by gender, using “good enough” information so that aid is safe, fair, and effective.
TGLF’s Reda Sadki and Charlotte Mbuh set out that purpose in Reimagining Rapid Gender Analysis as decolonial practice.
In the same article, he warned that prioritising speed and standardisation can reinforce hierarchical knowledge practices that centre the international NGO as the expert and the local population as the subject of analysis.
Our alternative, which we call decolonial rapid gender analysis (dRGA), starts from the fact that women, and the men who support them, are already analysing the context.
The international specialist becomes a “translator of systems”, who receives intelligence from women-led organisations and repackages it into the language donors require.
Seen this way, tools such as the RGA build analytical muscle, as Ogochukwu Udenigwe and her colleagues argue.
The skill is one more of those tools, and it has to meet the same test.
An AI model also speaks from a position
The dRGA critique rests on an idea from intersectional feminist scholarship: all knowledge about a crisis is situated.
It comes from particular people in particular places, and gender is only one of the systems of power that shape what they know and what they risk.
That idea applies to the model as much as to the consultant.
A language model reflects the texts that trained it, and most of those texts were written in English, in the Global North, by institutions.
Situation reports share that origin.
International agencies produce most of them, they often describe affected people in gender-neutral terms, and they tend to cite the same few visible organisations.
Anyone who automates the secondary data review without questioning it moves the hierarchy that the dRGA critiques from the consultant to the machine.
We wrote rules into the skill to counter this.
It must never invent a figure or a quotation.
It must mark every claim as verified, single-source, or unverified.
It must explain risk through conflict, money, and the collapse of services, and never through culture alone.
These rules reduce the harm, but they cannot remove the bias in what the model has read.
So the skill also gives its users questions to ask of every draft.
Catherine D’Ignazio and Lauren F. Klein built the seven principles of Data Feminism on the intersectional thought of Kimberlé Crenshaw.
Five of those principles translate directly into checks on the AI output.
Users can put these questions to the skill directly.
It will answer them, and the user decides whether its answers hold up.
These questions may also be used when reviewing a draft RGA prepared by a human.
| Principle | Question to ask the AI draft RGA |
|---|---|
| Examine power | Whose sources dominate the evidence base, and whose are missing? |
| Consider context | Where did the model describe people in gender-neutral terms, or treat a figure as more certain than its collection allows? |
| Rethink binaries | Does the draft see only women, men, girls, and boys, and erase other dimensions of lived experience? |
| Embrace pluralism | Does the draft treat local and experiential knowledge as evidence, or as colour? |
| Make labour visible | Who did the work the model is summarising, and does the draft credit them? |
How to use a skill for the first time
A skill is a folder of instructions that an AI assistant reads before it does a task, much like the method binder a new colleague receives on day one.
- The main file, SKILL.md, tells the AI when to use the skill and which steps to follow.
- A references folder holds the rules, the research checklist, the document structure, and the quality checks.
The model predicts plausible text from what it has read and from the instructions in the folder.
It does not know your context.
Once you understand that, you can challenge its output with the confidence you would bring to a colleague who has never visited the area.
You do not need to write code.
Follow these steps.
- Download the skill from GitHub and add it to your AI tool. In Claude, you upload it as a skill. In other assistants, you can add the files to a project or paste SKILL.md into the instructions.
- Fill in the input template with the emergency, the location, your audience, the language, your documents, and the distribution format (web page or PDF).
- Type “Run Phase 1 of the rapid-gender-analysis skill” and review the evidence file it returns.
- Ask the evidence file the questions in the table above, take its gaps to your local partners, and give the skill your corrections.
- Type “Run Phase 2 of the rapid-gender-analysis skill using the corrected evidence file”. Then review, edit, and finally sign the RGA as the human accountable for it. State in the document how AI assisted its production.
An example of how the RGA Skill can help
Here is how a first week might look.
In this invented scenario, a government declares an outbreak on a Monday.
That afternoon, the skill returns an evidence file.
It shows that no sex-disaggregated case data has been published, that maternity services were understaffed before the outbreak, and that a local women’s network is running burial and caregiving support.
Asked whose sources are missing, the skill reports that all of its sources come from international agencies.
On Tuesday, you call the network, or raise the gaps at its own meeting if you are a member.
You ask what its members see, what they need, and what they want an AI to understand about gender in their context.
You agree how they will be credited and whether anything they share may go into the tool.
On Wednesday, the skill turns the corrected evidence into a draft that names the network in the funding recommendation and gives it a seat in the coordination structure.
The operator manual inside the skill asks you to record your own times, so that the gains can be measured.
Women’s knowledge belongs to the women who hold it
Step 4 is where the skill can do the most good and the most harm.
Voice notes, WhatsApp summaries, and call notes from local partners can make the analysis far better, and they can also expose the people who shared them.
The Feminist Data Manifest-No commits its signatories to working with and for minoritised people and to making “no” a real option.
In practice, that needs to include practices such as:
- Ask before you upload anything a partner shared, and accept a refusal without penalty.
- Remove names, locations, and any detail that identifies a person.
- Keep survivor testimony and gender-based violence case details out of commercial AI tools.
- Agree how partners want to be credited, including as co-authors.
- Delete local inputs from the tool when the analysis is done.
Free AI tools usually offer weaker privacy protections than the enterprise tools that large agencies buy, and many local organisations can only afford the free ones.
That gap comes from how AI is priced and sold, and it leaves the person at the keyboard with the responsibility for protecting the data.
The time saved belongs to engagement
The skill depends on labour it does not show.
- Local staff wrote the sitreps.
- Women gave the testimony that fills them.
- Workers whose names nobody will see labelled the data that trained the model.
Credit the first two wherever you can, and remember the third.
The hours the skill returns should go to paid, reciprocal work with women-led organisations, on terms those organisations set.
They should not turn into more reports for donors.
Listening to women during a crisis also costs them, and the listener, emotional energy that no budget line records.
The time the skill saves gives no licence to rush that listening.
Peer networks build the trust that connects analysts across the North and South
Engagement depends on competencies that are relational: listening to people in crisis, earning their trust, and sharing power and credit with them.
People develop those competencies in relationships with other people.
We build those relationships online, in peer networks where practitioners learn from and support each other over weeks and months.
Our Gender in emergencies programme works this way.
When we announced it in September 2025, we started from the firsthand experiences of practitioners in our network and from intersectional, decolonial, and feminist scholarship, with no pre-packaged curriculum.
In that programme, we say, “your frontline experience is the textbook”.
Our learning-to-action model, the Impact Accelerator, runs on a weekly rhythm.
- On Monday, participants commit to one concrete action and share it with their peers.
- On Wednesday, they check in about what is working and what is difficult.
- On Friday, they report what happened, and their peers ask how it turned out.
“Guides on the side” offer expert support without taking over.
Participants describe networks that outlast the programme.
One of them said:
“I used to feel like I was the only one facing these problems. Now I have colleagues across my country who understand and support me.”
Week after week of shared action and mutual accountability builds bonds of trust.
Those bonds can connect an analyst in the Global North, who may have funding, technical skills, access to AI tools, and fluency in donor language, with a counterpart in the Global South who lives in the affected community and knows it from the inside.
Our aim is a site where global strategic knowledge and “authentically intelligent” local experience can meet as equals, and where a practitioner’s first responsibility is to the community served, which we call “relational accountability”.
In an RGA, that relationship lets the analyst bring the AI-assembled evidence and the counterpart bring what the documents miss, with credit shared between them.
In the Gender in emergencies programme, we counted 6,592 practitioners on 21 January 2026, and most of them are leaders based in local communities of Asia, Africa, and Latin America.
When we opened a call for facilitators, applicants identified as survivors of conflict and displacement and described their commitment to gender equity as “painfully and personally lived”.
Members who join meet and learn with people who live and work for gender equity in the communities affected by crisis, and whose lives are shaped by age, disability, displacement, ethnicity, and class as much as by gender.
We released the skill openly so that others can write the next version
We should not write the next version of this skill alone.
That is why we are releasing it as an open skill under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International licence (CC BY-NC-SA 4.0).
A commercial enterprise that wants to use the skill needs to request a licence from us.
Anyone working for a not-for-profit purpose can use it, adapt it, translate it, fork it, and contribute improvements, as long as they credit TGLF and share their versions under the same terms.
Women-led organisations can rewrite its rules so that it reasons about gender the way their contexts require, and their versions stay open to everyone else in turn.
The Feminist AI Research Network (f
The skill cannot consult affected people.
An RGA drafted from documents is a secondary data review, and the skill says so in the document itself.
It cannot verify what nobody has published, it cannot detect every bias it inherited, and it cannot move money.
Only the people who use it can turn its recommendations into funding, decision-making seats, and authorship for local women-led organisations.