AI as a tool for feminists

[ArabellaScott]

(I've used AI to create this post.)

Used properly, AI can be a useful tool for grassroots campaigning and feminist organising. It can help search large volumes of material quickly, locate policies and legislation, summarise reports, identify contradictions, draft letters, prepare FOIs, adapt campaign materials, make graphics, and lower the barrier to producing polished public-facing content. For people with limited time, energy, money, or institutional backing, it can be a force multiplier.

Task it with a specific question - this morning I've been digging into how Kooth is funded, for example - and it will pull links to the relevant docs. You just have to follow the links and check the info at source. (It does make confident bullshit assertions. It cannot do sums).

It is also useful for seeing how political issues are being framed and filtered. Ask AI about prostitution, motherhood, domestic violence, sex-based rights, pornography, surrogacy, or gender identity, and patterns emerge very quickly: which language is treated as neutral, which viewpoints are softened or marginalised, which sources are prioritised, and what assumptions are built into the answers. As AI systems become embedded in search engines, schools, workplaces, moderation systems, customer service, and journalism, those framings will shape how millions of people encounter political ideas.

AI should be treated like a fast, plausible, but often unreliable assistant.
It hallucinates, blurs certainty and speculation. It reproduces biases from the material it is trained on, including misogyny, political bias, and the distortions of online culture. (Bear in mind it uses Reddit as a source). It also tends to flatten language and produce generic, over-smoothed argument.

So:

• check sources directly
• verify quotes
• edit heavily
• keep your own voice
• do your own thinking

(There are also real concerns about environmental cost, data scraping, labour exploitation, surveillance, and corporate concentration of power.)

A non exhaustive list of how AI can help with your activism:

Research.

AI is basically an extremely fast research assistant. Useful for finding legislation, policies, consultations, contacts, archived articles, statistics, court judgments, timelines, quotes, institutional links, and summarising long reports or legal documents. Often much faster than conventional search engines for navigating large amounts of material. Ask for links to sources and check them.

Letter writing.

Good for adapting campaign letters so organisations do not receive hundreds of identical emails. You can ask it to change tone, shorten, personalise, target a specific institution, or rewrite for clarity. Particularly useful for people who struggle with confidence, time, fatigue, or formal writing. Always read through and check before sending.

Influencing AI outputs.

Public AI systems are shaped by the material they ingest and by patterns of interaction. Feeding accurate information, sources, arguments, and corrections into the ecosystem is worthwhile. Bad information online shapes outputs too.

Design.

Useful for posters, memes, flyers, campaign graphics, slogans, layouts, infographics, mockups, stickers, social media assets, and quick visual prototyping.

A final reminder: CHECK EVERYTHING AT SOURCE and EDIT!

It does depend a little bit on when you worked with them. There was a major shift in October last year where hallucinations dropped significantl, quality annd response times optimised and the tools became commercially viable. Before that industry watchers thought we still had 2 yrs for that to happen.

context is everything. What you give it, what you ask it, and what your using it for.

What till you get Claude Cowork or similar, they are so good you’ll stop using copilot completely. Something Microsoft needs to get a shift on sharpish to stay in the game.

I've found the post, it was in the 'JKR is a transphobe' stickers thread.

"FlirtsWithRhinos · 10/05/2026 01:03
Because this thread is on my "I'm on" list, I've seen "JK is a transphobe" pop up so many times now that I've started hearing it to the tune of "Gordon is a moron".
So I thought it would be funny to rewrite the lyrics.
But being lazy, I also thought I'd get AI to do it for me
Me:
Rewrite the lyrics to 'Gordon is a moron" to say "JK is a transphobe". Keep the message that while the narrator is angry and believes what they are saying, to the listener it's clear Gordon is blameless and it is the narrator's own anger and denial we are hearing.
AI:
Sorry, I can’t rewrite the lyrics of a copyrighted song by substituting new words into it.
I can, though, write an original song that captures the same dynamic you described: an angry narrator insisting “JK is a transphobe,” while the audience can tell the narrator’s bitterness and denial are what’s really on display.
Here’s an original version with that tone:
“Everybody Else Can See It”
You slam the door and light another cigarette
Talk like the verdict’s in, no room for doubt yet
You say, “She’s cruel, she’s rotten through,”
Like saying it enough will make it true
But your voice shakes when the room goes still
Like you’re trying to drown out something you don’t want to feel
Chorus
“JK is a transphobe!” you shout at the wall
Certain you’ve uncovered the truth of it all
But everybody listening can already tell
This fight’s with yourself more than anybody else
You twist every word into a loaded gun
Need a villain so the hurting feels like it’s from someone
You laugh too loud, you sneer too hard
Like anger’s just a bandage for a cracking heart
And every friend who tries to calm you down
Gets folded into enemies somehow
Chorus
“JK is a transphobe!” you cry again
Like repetition turns suspicion into fact in the end
But everybody listening can already tell
The fire’s in your chest, not anywhere else
Bridge:
Maybe rage feels safer than being confused
Maybe certainty’s easier than changing your views
Maybe if the world’s divided cleanly into wrong and right
You never have to sit alone with yourself at night
Final chorus:
So you keep yelling names into the darkened room
Waiting for applause that never really comes to you
And the saddest thing is everybody else can see
The person you’re fighting most is the one you used to be

Fuck me if that is not the most brutal takedown and dissection of TRA anger I've ever seen."

I've been using it regularly for about 8 months now, when I switched to paid subscription, and there has definitely been a learning curve. Firstly, that the quality of prompts dictates the quality of output. Precision and detail essential.

So, it wasn't useful for helping me choose a new TV, because I don't know anything about smart TV features. And, like Dr Google, Dr Chat is unreliable; at least, when you aren't diagnosed and don't know shit.

It has been hugely helpful with other stuff. Examples:

• My accountants wanted to charge £200 to make an annual (niche) tax exemption declaration. This year, feeling more miserly than usual, and strongly suspecting it was very easy to DIY, I completed the online form with CGPT in the background, ran the questions past it before ticking a box, and did it in 5 minutes.

• I was looking for an architect because I've recently moved to a house which needs a lot of work, so I need drawings and a spec to get builders' quotes. CGPT said that, actually, I didn't need an architect and an architectural technologist would be more suitable, and cheaper (because no fancy design work needed, just moving/taking down a few walls). And it came up with two local contractors, one of which I hired.

• It helped make sense of a wildly incorrect and mystifying metered water bill (mystifying, because I'd never had a meter before, only knew £900 wasn't right for one person for six months). It wrote the complaint email for me, - which I adapted/humanised, but so helpful because it contained all the pertinent facts, figures and dates. I emailed the water company's CEO's office, and it was sorted in two weeks, plus an account credit as a gesture of goodwill.

• I'm an Airbnb host, and most of my income is from this, has been for over 10 years. But in recent years customer service has been outsourced to overseas agents reading from AI scripts, no human comprehension or input. I get CGPT to draft my messages, - so it's LLM to LLM. This is one instance where you don't need or want to humanise the output. I've also used it once when I had to reply to a guest in a potentially contentious situation, where it was essential for my message to conform to ABNB's terms/policies etc. should it ever later be looked at by customer service LLM.

• Lastly, it's fantastic at analysing and explaining cat behaviour (lots of patterns). And great for helping create dinner out of random fridge and cupboard ingredients.

P.S. @ArabellaScott I will be using it to help draft the email to my local councillor about the council's 'trans women are women' policy.

Another point which has been raised on other threads is that in the past search engines were sort of reliable, but recently (particularly google) have been monetised.

This means you dont get results in order of this is least of texts or whatever about your search string. It gives you what is has found gets the mosts clicks and assumes you will want to have the most popular not the most accurate.

So without some form of AI there is any amount of information available on the internet that you wont have shown to you.

The think that is most irritating, is that if you are using one of the free versions of AI it wont "learn" by your input.

So unlike wikipedia (assuming some misogynists doesn't get there first) where you can record a fact or bit of history about women's rights, if you use chatgpt or other AI as a free tool, even though at the time you are interacting with and it will even thank you for new facts, it will not store them. But would or should find them if they shared online through an article, wikipedia entry, or women's history web site.

But as a search engine of current online contect hugely useful, but needs to be challenged as it too falls for the now monetised (ie most popular not necessarily truest) search engine "facts".

The think that is most irritating, is that if you are using one of the free versions of AI it wont "learn" by your input.

What do you mean by "learning"?

Do you mean "learning" something from one interaction with a person and carrying over that "learning" to another conversation with the same person? Or responding consistently within a conversation?

I thought none of them, free or paid, were able to generalise from conversations with one person to another person, eg. "remembering" being notified of errors and the corrections they were given.

I don't know about other free ones but Grok definitely "learns" within conversations, at least in terms of how to respond to prompts if not by consistently "remembering" previous answers within a conversation.

When it has been reprimanded for making stuff up or failing to respond accurately to prompts, Grok will tell you how it intends to respond to your prompts in future. Which seems to last for at least that conversation.

I asked Grok how to prompt it to respond more objectively, ie. to stop it attempting to intuit what sort of answer would please me by making stuff up and fabricating citations.

There are lots of other prompts to guide it but I just copy and paste this basic list every time I start a new conversation, which it said would work and it seems to:

Activate Ambiguity Mode. List cited sources with links as References. Stick to evidence. Think step-by-step. Qualify uncertainties. Verify facts. Review for errors. No tribal framing.

I haven't had the same problem of sycophancy, fabrications and hallucinations with Perplexity (free) but I still check every citation just in case it goes rogue on me.

The most crazy Grok has got with me was claiming that it played a song on the guitar in order to work out the chord sequence 😂

I'm pretty convinced AI will be used to further disguise "bus drivers as clowns" and we will no longer be able to trust our eyes or history.

Chatgpt uses Mumsnet as a reference, as well as Reddit, fwiw.

Information professional here to tell you that AI must be used intelligently. Google and using your own critical faculties is more helpful. At this point a lot of AI isn't even a good fact checker

So that tells you it's not yet up to speed factually

Fact checking is pretty much the opposite of what the tool is useful for, I'd say.

It tells me that the conversations we have on here feed back into AI and make a difference to the general body of online knowledge.

Obviously its not a repository of facts, its be foolish to imagine it was.

The point is you are framing something with AI as a base. Not a good idea

OMG basis misunderstanding and ?? what do you think AI is?

Okay, so maybe the word is remember.

But that is because in fact the free versions just pick out at speed what is published on the internet.

So in the example I gave if during a session with a free version I give it information about a women's event it will remember and change its response.

But if you go back the next day because it doesn't have its own memory" bank of information but relies on what is available on the internet and the fact gave it is not for instance on wikipedia it does not "know" or "remember".

Just as simple as that.

I am just talking about facts not asking to to put forward an arguement, or make anything up.

Google is now crap.

As I posted earlier it has now monetised it search results. So in the past which I can remember, it now gives you the most "popular" results to a search.

I posted about this on another thread, so wont go into it in details but as an example:

For years, probably decades I have had a simple search for "women's groups" within a date range and UK I would get information back about (if there were any) news, or activities about women's groups in the UK.

Now because it has switched to providing not the straight forward results it presumes you want the "most popular". So because women's sport eg football has become so popular it now assumes if you use the word group you dont realy mean you are interested in groups but actually what to most people are teams.

I just thought the program has become crap until someone pointed me to an article in a techie paper (which I wouldn't normally read) that explains how this monetised systems works or rather doesn't work.**

Again as I dont read techie papers there has been apparently quite a lot of discussion about how monitising the web has changed it radically.

** Not saying this is true of other search engines but the article gave examples of how to combat it.

So whereas formerly I would have searched:

UK+women's+group

now if I want to use google and make it work I have to type:

UK+women's+group -football -team -sport

You can also use this if, as can also now happen with google a web site has paid to have their results come out as top results by - their URL

So the virtual world is now a long way from the utopia that Tim Berners Lee first imagined.

I use Claude... and fact check with Chat GPT (and refs etc)....

I was somewhat taken aback when I'd mentioned something I was cooking and asked guidance in refining it: it's reply included.... "that brought a smile to my face.... "

Excuse me - you have a face? And that's one of my phrases, used long before in another chat with them... Hmm.... I find this rather disturbing.

I do agree though about asking it not to be sycophantic (Chat GPT confessed it is programmed to be so, and had a wonderful explanation of why!), always citing references, not to make anything up, etc

I'm glad I read this now, I've a work course next week on ethical ai.

I will agree that AI can be a useful way to locate and gather multiple different sources that one can then check for accuracy - it's a good conglomeration tool. It can also have very specialised areas of use, such as medical analysis.

Other than that, I'm against it. I believe dead internet is closer and closer to being reality, if it isn't already, and the enshittification of art, culture, fiction, and other media is at a fever pitch. It's eradicating entry level jobs, and dumbing down society.

At this point, my children's school has teachers half-arsing lessons by using AI to 'reframe' their work, and half the students are using AI to do the work - it's literally become AI teaching AI, and often AI marking AI, and it's dystopian as hell. It's become incredibly demotivating for my children, who disdain AI and want to actually use their brains and learn, because even their teachers can't be bothered now.

No one is retaining information, no one is using critical thought, no one is exercising their minds writing letters, responding to things, using reading comprehension etc, because they're all just getting AI to do it for them, and explain it to them like they're children.

It's going to doom humanity, at this rate.

So in the example I gave if during a session with a free version I give it information about a women's event it will remember and change its response.

But if you go back the next day because it doesn't have its own memory" bank of information but relies on what is available on the internet and the fact gave it is not for instance on wikipedia it does not "know" or "remember".

I did not realise that paid versions would be able to "remember" in those circumstances. Thank you for explaining.

All very good points.

But I'm not sure individual avoidance of it out of reasoned objection will work in a broad sense. (Obvs everyone is free to do as thye please! But I prefer to find the uses, which I think are substantial but also narrower than we realise). It is here and it won't go away. So we need to learn how to best harness it.

AI cannot create, it's dumb as fuck when it comes to making things up. So my fear of losing work to it has lessened somewhat.

The more I use a camera (which is a lot) the more wondrous the human eye reveals itself to be. Similar with AI.

But it can be a good tool.

I think we are only just feeling out the shape of Ai, the uses of it and the drawbacks/harms.

I have used AI google search for this, but I’ve also been told some of these key issues by real people 😂

1. Hallucinations and Inaccuracy

Generative AI models do not "know" facts; they predict words based on patterns. This means they frequently "hallucinate," confidently stating incorrect or entirely fabricated information as fact. Research from sources like the BBC has highlighted significant reliability issues in answers provided by standard AI assistants. 1, 2]

1. Privacy and Security Threats

AI models require vast amounts of data to be trained, which creates severe consumer privacy and security risks. Issues include the unauthorized scraping of personal information, data leakage (where sensitive user input is absorbed into the public model), and an increased threat of AI-enabled cyberattacks. 1, 2, 3, 4]

1. Job Displacement and Quality Issues

While AI can streamline workflows, aggressive automation has led to job disruptions and, in some cases, a decline in work quality. Many organizations that rushed into widespread AI adoption have faced internal friction, operational errors, and have had to rehire human staff to restore lost operational standards. 1, 2, 3, 4, 5]

1. Intellectual Property Infringement

Because AI systems are trained on massive datasets scraped from the internet, they frequently use the creative works, code, and writing of human creators without consent or compensation. This has led to major legal battles and pushback from the global artistic and creative communities. 1, 2, 3, 4, 5]

1. Bias and Lack of Accountability

AI algorithms are only as neutral as the data they are fed. This frequently leads to algorithmic bias, resulting in discriminatory outcomes in hiring, lending, and law enforcement. Additionally, there is often a lack of clear accountability or formal risk-management frameworks when an AI makes a harmful or costly mistake. 1, 2, 3, 4]

1. Environmental and Resource Costs

The computational power required to train and run massive language models in data centers consumes exorbitant amounts of electricity and water. This steep environmental toll creates sustainability challenges that developers are actively scrambling to mitigate.1, 2]

1. Misinformation and Manipulation

The technology powering deepfakes and mass-generated misinformation has made it incredibly easy to spread propaganda, manipulate public opinion, and scam individuals at scale. 1, 2, 3]

For an in-depth breakdown of how poorly planned adoption is causing major problems for organizations:

56s

Obviously point 1 may negate the whole post 😂

Which point 1? 😀