Google's AI search (still) looks like a bad idea
New research confirms the obvious downsides for digital publishers.
Google is testing an “AI Mode” for its search engine, a fully AI-powered experience that goes beyond the AI Overviews that now appear alongside some traditional Google search results. Using reasoning capabilities, Google claims it can answer what would otherwise be complicated, multi-step searches all at once.
An example Google offered asked the best time for an engagement photo shoot at a Boston park. The results pulled in real-time weather data, best practices for photography lighting, and historical data for when crowds are lightest.
Who asked for this?: Tech companies are building more AI-delivered knowledge into their products to justify the piles of cash being spent on it, but most users have remained largely indifferent — and when it comes to search engines, even a bit hostile, as they feel AI has made them less effective tools for finding the info they want.
AI Overviews are still plagued by very obvious inaccuracies, even after the glue-on-pizza fiasco shortly after launch.
Some recent errors pointed out online include claiming baby elephants can fit in the palm of your hand, suggest that sex toys be used to find behaviour changes in children, or believing the Underground Railroad was an actual railroad.
But also: If you’re a website that isn’t actively developing AI search, it’s probably bad for business, since getting an AI summary gives users one less reason to click through to your website.
Google — as well as other companies committed to AI search, like Perplexity and OpenAI — have promised that sources would be prominent on result pages, leading to more traffic from users looking to dive deeper.
What the research says: New data from content licensing platform TollBit seems to confirm publishers’ worst fears, finding that AI search engines sent 96% less traffic to a publisher’s website. Research by the Tow Center for Digital Journalism examined eight AI search tools, and found that they frequently fabricated links or cited copied versions of content. This didn’t seem to be much better for sites that struck licensing deals with AI companies, and premium subscriptions only seemed to make bots more confident about incorrect answers.