Female founders raised a record $73.6 billion in the US last year.

Which, at first blush, looks like much-needed progress chipping away at the funding gap (or rather, chasm) between the sexes. After all, that number corresponded to over 25% of total venture funding, a significant hike from the prior year’s 19%.

Yet if we subtract Anthropic and Scale AI (whose female co-founder actually departed in 2018), the trend line reverses. Without those two behemoths, which made up ~$30 billion of the total, female founders actually fared worse than the year before.

All-female teams had a particularly rough go, receiving only 2% of total funding, roughly what they got in 2019.

So what gives?

Female founders, in my experience, are no less competent than their male counterparts. And yet, they seem to find less success navigating the gamut of investment committees across our industry. Looking at the high-level data, it is tempting to come away with the conclusion that our industry is simply sexist, that venture’s love affair with a certain breed of Stanford alpha male is too deeply ingrained to be disrupted. But when we look closely at the available studies, the takeaways prove more nuanced than that.

In one study, Will Gornall and Ilya Strebulaev sent 80,000 fictitious pitch emails to thousands of VCs and angels, randomizing founder names by gender and race. If cold-stage, tacit sexism were at play, the female founders should have been screened out at a higher rate, receiving fewer replies. Yet the opposite happened: they received 9% more replies. Clearly, any discrimination that was occurring was not occurring in the inbox or pitch deck. It had to be somewhere else: in face-to-face (or more likely, webcam-to-webcam) interactions.

Two further studies give an indication of what happens in a live format. The first is Dana Kanze’s. Across seven years of TechCrunch Disrupt Q&A sessions, Dana observed that investors tended to ask male founders promotion-focused questions (about growth and early wins) while asking female founders prevention-focused questions (about risk, defensibility, and what could go wrong). Founders tended to answer in the register they were handed, and the promotion-focused group raised significantly more capital.

The second study, while less well-known, is even more salient to the topic at hand. Lakshmi Balachandra and her colleagues analyzed 185 pitch videos for gender-stereotyped behavior. Their headline finding was that while gender did not directly influence investor interest, any display of typically feminine-coded behaviors — such as warmth, emotional expressiveness, sensitivity, or hedging — reduced investor interest in men and women alike.

The gap, therefore, comes down to confidence rather than competence. Or, to be more exact, the performance of confidence, an area where men tend to literally outperform women, regardless of their actual traction or qualifications.

Christine Exley and Judd Kessler investigated this phenomenon from another angle in the Quarterly Journal of Economics. Across experiments involving more than 4,000 adults and 10,000 students, women asked to assess their own performance on math and science tasks rated themselves less favorably than men who had achieved identical scores. The gap held steady even when every strategic incentive to self-promote was eliminated, suggesting it was not merely a negotiating posture on the part of the men. Counting all their versions and replications, the researchers found that the gap appeared 64 times out of 64. It was even visible in sixth graders.

This delta between confidence and competence fascinates me, and it brings to mind an analogy with the realm of LLMs.

Last September, a team at OpenAI published a paper called Why Language Models Hallucinate. In it, they argued that models hallucinate because training and evaluation reward educated guessing over any admission of uncertainty. The illustration is stark. Imagine Model A, which signals its own uncertainty accurately and refuses to make things up on the spot. Now imagine Model B, identical except that it never says “I don’t know” and always guesses when unsure. Under the binary scoring rubric that underlies most benchmarks, B outscores A every day of the week.

The bluffing is not a defect in Model B. Rather, it is a designed output, functioning exactly as specified. The paper’s practical conclusion is that sprinkling in a few humility-aware evaluations on the side accomplishes nothing when set against hundreds of accuracy-based ones that penalize uncertainty. So long as the scoreboards continue rewarding lucky guesses, models continue hallucinating.

As investors, we often talk about conviction. We strive to reach conviction before making an investment. Yet we are too often swayed by a founder’s own conviction, penalizing founders who possess the requisite humility to acknowledge their own uncertainty while rewarding the swaggering Adam Neumann archetype. Investors who select for conviction are opting for Model B, even when it lies to them.

For better or for worse, men are more likely to act like Model B, while women operate like Model A — and pay for it.

I want to be precise about why this phenomenon should bother general partners, and not simply chief people officers.

Dana Kanze ran a follow-up experiment showing that founders who answer “prevention-focused” (downside-oriented) questions in “promotion-focused” (i.e. upside-oriented) terms raise more money. This suggests that, with a little coaching, female founders can dodge the Q&A trap, which is usually reported as an encouraging takeaway from her research.

I think it is the most damning part.

A performative signal that can be learned in the space of a workshop is a signal that carries no real information. If the facade of confidence is the main thing separating founders who get funded from ones who don’t, then what we are actually selecting on is who has been coached, who has been in the right room before, and who resembles founders we have already funded. In other words, we are reinforcing the existing advantages that accrue to people who have the right networks and pedigree, regardless of their aptitude in a given domain.

Now, there’s a natural objection here that we should flag: Isn’t projecting confidence a critical part of the job of CEO? Founders need confidence to to recruit effectively against better-funded competitors, close deals before their products have fully matured, and raise downstream capital. Faking it till you make it is often an effective strategy; we should be honest about how the world works.

Fair enough. But pitch contest formats, breezy by nature, have a way of turning any admission of uncertainty into an automatic red flag. We ought to recognize that while some forms of uncertainty are indeed disqualifying, others are admirable. From the CFO perspective, a degree of conservatism amid uncertainty is a desirable trait in founders, particularly when it comes to financial modeling, an area where many early-stage founders struggle mightily.

The founder I am excited to back is rigorous in diligence and compelling in a recruiting conversation. Those are two distinct capacities that pitch sessions too often conflate. There is a cheap empirical proxy for which trait we should prefer. Confident-wrong founders burn money faster. Female-founded companies ran a median burn of $350,000 a month in 2025, against $390,000 for the broader US market.

So what steps can we take to remediate this issue? Here are a few ideas:

• Audit your own Q&A. Analyze your last dozen or hundred IC-meeting transcripts for upside- versus downside-oriented questions, split by founder gender. Are the softballs and hardballs served out in equal measure?

• Don’t penalize uncertainty. Give founders explicit credit for a well-placed “I don’t know.” Score a confident error worse than an acknowledged gap.

• Move diligence procedures into writing, where the Gornall and Strebulaev result suggests bias is less rampant. This is especially critical for areas like unit economics and product architecture.

• Embrace tempered forecasting. While founders face understandable pressure to show audacious growth assumptions in their decks, we as investors should avoid overlooking founders who favor more realistic forecasting.

Do you have thoughts on ways to address the funding gap? I would love to hear from you in the comments.