They don't care. They are optimizing for maximum signups, presumably because signups = growth = VC money.

Yes, but conversion rate is only a part of that. If nobody shares your article, it doesn't matter how good your conversion rate is. Optimizing conversion rates while ignoring the viral coefficient is like the drunk looking for his keys under the streetlight because that's where the light is best.

Well, I guess the "viral coefficient" is something inherently unpredictable, so it's probably hard to include it in calculations.

Yes, but that's my point. Using only what's easy to include in calculations biases the outcome to what improves (easily measurable) conversion rate and not what improves the (difficult to measure) viral coefficient. This would be fine if the viral coefficient were independent of the variables they are adjusting to improve conversion, but I'm saying that they're not.