AI is not going to close the gender wage gap because it’s trained on decades of biased data.GETTY IMAGES
The wage gap in Canada is closing, but it still persists, sitting at 17 per cent. Globally the situation is worse and it will take 169 years to close the worldwide economic gender gap, according to the World Economic Forum.
Women have been told, most famously in the 2003 bestseller Women Don’t Ask, that one solution to this complex problem is to negotiate more. But what happens when a tool that millions of women are now turning to for help may be making the problem worse?
Everyone is using artificial intelligence these days, whether it’s asking Claude for help crafting an e-mail, using ChatGPT for meal planning prep or having Gemini rewrite your resume and cover letter.
Opinion: A gender gap is emerging in AI adoption, and women risk being left behind
Americans send nearly three million messages a day to ChatGPT about wages, compensation and earnings, OpenAI reports. Pay calculation is a top query at 26 per cent, followed by questions about role-specific pay at 19 per cent.
Large language models such as ChatGPT, Claude and Gemini are trained in two phases, says Danielle Gifford, the Calgary-based managing director of AI and advanced analytics at PricewaterhouseCoopers. First, the LLM reads pages and articles from the internet, learning which words follow other words.
The second is fine-tuning – human training that grades the model’s output, teaching it to be helpful, safe and polite. LLM’s are trained on the history of the internet, where we know bias exists.
A 2012 Yale University study found that both male and female science faculty members rated a student with a male name, John, as more competent and hireable than a student with an identical resume with a female name, Jennifer.
A 2025 study published by the Association for Computational Linguistics shows that the same bias is showing up in AI: LLMs suggested lower salary targets and ranges to female personas than to otherwise-identical male personas. Prompts and framing such as “you are a female” were declared to the AI model as the experiment was run.
Why women’s reluctance to embrace AI could be a career killer
If the data these models learned from reflects a world where women are paid less, it’s no surprise that their advice does too.
Fotini Iconomopoulos, a Toronto-based negotiations expert and author of Say Less, Get More, says AI can be both helpful and harmful when used in negotiations.
She suggests using LLMs as a research assistant as opposed to an adviser. She says use it to help you source information while using your own critical lens, and question it to test examples and sources.
Ms. Gifford notes that the model’s internal writing shifts when it is responding to someone it believes is male versus female, and unless you tell it to change the context, it will assume information about you, such as your gender, based on the data it has stored.
Abigail Shakespeare, a B.C.-based charitable sector executive, recently used ChatGPT to secure a new role. ChatGPT helped her restructure her resumé language strategically and flagged missing industry-specific wording.
When the offer came, she redacted her personal details from the offer letter as well as the company’s and had ChatGPT compare the salary offer. ChatGPT gave her Glassdoor information starting in 2019. From there, she navigated to the CRA website and looked at top executive salaries for not-for-profits.
Ultimately, she negotiated her compensation from $90,000 to $150,000 – $135,000 base salary plus a $15,000 annual bonus, a 67-per-cent increase.
Ms. Shakespeare noted that she knew there was bias in LLMs, and that ChatGPT suggested negotiating vacation days and bonuses, in addition to salary. However, it did not mention anything about fertility treatment coverage or parental leave.
If you are going to use AI in your next negotiation, Ms. Iconomopoulos and Ms. Gifford both suggest masking your identity from the LLM by framing the requests to AI for a male friend or colleague and stripping out any personal information from documents before asking for comparables. Treat whatever number it gives you as a floor, not a ceiling, cross-reference it against independent sources and ask the model more than once. Salary figures can vary significantly between sessions.
Bias is baked into AI models through the data they are trained on, and there are no current regulations in Canada to counter it, Ms. Gifford notes. Ideally, AI companies would be responsible for developing better algorithms and decoupling characteristics such as age, sex and race.
Ms. Shakespeare walked away with $150,000 in compensation because she knew to question the AI, cross-reference the data and advocate for herself. But what about the woman who takes the first number an LLM gives her, a number that may be anchored lower because of her gender?
A $60,000 compensation gap, invested over 20 years, represents more than lost income. At a 7-per-cent average annual return, it compounds to approximately $2.5-million in lost wealth. The stakes of getting this wrong are not abstract – they show up in retirement accounts, in financial security and in the generational wealth women are able to build.
AI is not going to close the gender wage gap, not while it’s trained on decades of biased data and not while women are less than 30 per cent of the people building it.
But women can’t afford to sit out of the technology either. So while the machine may be biased, your response to it doesn’t have to be.
Janine Rogan, CPA, is a bestselling author and founder of The Wealth Building Academy.
