A lot of SEOs seem to want to believe that Search Generative Experience (SGE) won’t roll out soon, or won’t ever roll out at all. One reason some people are citing is the cost and loss of ad revenue but they are missing some important points.
I personally don’t think that SGE will impact ad revenues, or rather it could do positively.
At the moment if someone puts in an upper funnel informational query that no one wants to bid on and goes to an article, then decides to buy something mentioned in that article, or just from the site, Google get nothing. If someone gets their question answered on Google via SGE and then asks a follow up question or just searches again for the next steps in their buying journey: sooner or later someone wants to advertise, so more zero click at the top of funnel could mean more ad revenue.
Then maybe for those transactional queries SGE simply doesn’t get shown, or, as with the examples they showed at I/O last year (see below), Google can include ads, especially shopping ads or similar (e.g. things to do ads) within the SGE and make them hyper-relevant.
Google can also simply not show SGE when there are people bidding on paid search ads or decide to try showing ads above or beside SGE, there’s nothing to stop them tweaking things, automatically even, to maximise ad revenue.
The other money related point that people are talking about is how expensive SGE is to generate but this point isn’t quite as cut and dry as people want to think.
SGE has been designed to be fast and efficient, some of those hallucination issues are due to a model that is based on a small enough LLM and has a low enough number of parameters that it can work fast and with less resource than you might think. BERT is already an example of a large language model created for speed and to not be impossibly resource hungry that is integrated into google’s search algorithm, for every search. Google have said that the technologies that SGE is based on, though 1,000 times more powerful than BERT are not so much more resource intensive.
Maybe more importantly Google also don’t have to generate a new answer every single time. They said they could generate a new answer for any unique search and even personalise your answers: but I don’t think they are, at least with SGE in labs. There are results I have seen for the same search terms, even with me supposedly in different countries, thanks to VPN, that seem to have the same SGE answer with the same sources, even days apart.
Generating SGE results and then storing them for a few days and not showing them for those unique one off searches would allow Google to launch this without it using up unmanageable amounts of computing power.
New search engines, AI search engines like Perplexity, Google have always known they need to innovate and push ahead to stay ahead, and can’t rest on their laurels, the sheer number of papers related to AI that have gone in to MUM show they have already invested a lot to get to where they are.
I have long, since before we knew it as SGE, been convinced Google will make Search Generative Experience a part of their main service in some form, and not a paid add on.
My best bet remains May 14th for the announcement of SGE rolling out (at Google i/o) and a soft launch a week or two after so let’s see who ends up with egg on their faces.