2 ms·

They probably also know that some types of queries are not ad-worth (like age of an actor, etc), while others would never show a box

derefr 2y ago
That's a really good point. Google has likely decided that some query categories are unprofitable-enough already that they can just drop the rest of the expected revenue from them, and instead use them as loss-leaders — aiming to give the customer a better (but zero-revenue) experience today, to increase retention, so they'll come back with positive-revenue-generating queries later. They've just done a rather braindead job of it, with the experience of these loss-leader queries feeling perceptually† worse than a regular SERP page would. It's like a grocery store running a sale on e.g. milk, selling it for below-cost — but then making you have to wait in a line for five minutes in front of the milk cooler to get it. It feels like so much friction that not only wouldn't lure people in; it would push people to avoid that grocer when its milk sales are happening, and shop at other stores instead! --- † I emphasize perceptually here. Even if the total time spent waiting for an AI answer is shorter than the time it would take to load a SERP page, read and select a link that looks like it'll have your answer, click it, load the resulting page, and read it — the AI experience could still feel longer, in large part because "the ball is in your court" for a lot of the self-navigated query, with you being given things to look at and click on; while "the ball is in their court" for the entirety of the AI response, with just a spinner and no indication of what's happening. As such, I wonder if this could be solved by just running a cute little animation under the AI prompt window: make it look like it's automating the loading of a [blurred to non-identifiability] SERP page, scrolling down, clicking a result, loading an [again blurred] page, highlighting a few random fragments of it, and then transposing them together to form the paragraphs of the answer — before the answer actually snaps into place on top of that animation. If you can predict how long before the AI answers, you can even use that to determine how long the animation should spend dicking around on the pretend SERP page and answer page.

derefr 2y ago

That's a really good point. Google has likely decided that some query categories are unprofitable-enough already that they can just drop the rest of the expected revenue from them, and instead use them as loss-leaders — aiming to give the customer a better (but zero-revenue) experience today, to increase retention, so they'll come back with positive-revenue-generating queries later. They've just done a rather braindead job of it, with the experience of these loss-leader queries feeling perceptually† worse than a regular SERP page would. It's like a grocery store running a sale on e.g. milk, selling it for below-cost — but then making you have to wait in a line for five minutes in front of the milk cooler to get it. It feels like so much friction that not only wouldn't lure people in; it would push people to avoid that grocer when its milk sales are happening, and shop at other stores instead! --- † I emphasize perceptually here. Even if the total time spent waiting for an AI answer is shorter than the time it would take to load a SERP page, read and select a link that looks like it'll have your answer, click it, load the resulting page, and read it — the AI experience could still feel longer, in large part because "the ball is in your court" for a lot of the self-navigated query, with you being given things to look at and click on; while "the ball is in their court" for the entirety of the AI response, with just a spinner and no indication of what's happening. As such, I wonder if this could be solved by just running a cute little animation under the AI prompt window: make it look like it's automating the loading of a [blurred to non-identifiability] SERP page, scrolling down, clicking a result, loading an [again blurred] page, highlighting a few random fragments of it, and then transposing them together to form the paragraphs of the answer — before the answer actually snaps into place on top of that animation. If you can predict how long before the AI answers, you can even use that to determine how long the animation should spend dicking around on the pretend SERP page and answer page.