Google announces major update to combat AI-generated spam in search results
wired.com
wired.com
I've been using Kagi for a while and it makes it feel like Google Search is a dead man walking. Even Bing has better results for certain things than Google Search.
It's hard to see Google improving search to the point where it's good again, we needed a new disruptor in search to show the mega corps how to make good products.
https://searchengineland.com/google-issues-search-ranking-pe...
https://developers.google.com/search/help/report-quality-iss...
They are doing this now because AI chats and other search engine are becoming competition. They ignored spam because it was making them money.
Not a production-ready idea of course, and the ideal LLM for that might have yet to be invented.
Still, I have plenty of faith in Google still to apply techniques like this in a clever way.
They were very serious about their "Knowledge Graph" already 15 years ago, albeit with totally different data processing and acquisition techniques.
If all else fails, Google has historically put much of the burden on website authors, e.g. all the schema.org and other structured data stuff.
Their game has become much harder, but I still believe that Google can and will remain competitive in web search at least for quite a while to come.
Despite all the frustrating spam and a perceivable drop in results quality over the last 5 years, I still don't know about any fully competitive rival for Google.
Google never solved how to win the game against SEO, and in recent years it seems they've either effectively lost or given up on it entirely. Generative tools make that much worse.
Comparing content based on LLM summaries wouldn't really do the trick. For one thing, the LLM itself is a black box so we don't know how it works and could never measure or predict the accuracy or reliability. For another, when two pieces of content have a matching summary the only tie breaker would be publication date (or first seen date in the case of Google). That really doesn't definitively answer the question of which was LLM generated, and won't help at all for content that is originally generated by an LLM and won't match any known content.
There's simply no way for an algorithm to accurately recognize computer generated content when the generation tools are good enough to mimic human content.
Otherwise, Google being Google would ignore this issue just as they have ignored all the SEO and bot spams in the past. They lost their visions and has succumbed fully to corporatitis. I think anything short of a major surgery or blood transfusion would save Google now.
Google could still detect what content gets the most engagement and try to guess at why, but they'll never be able to detect and filter out content simply because it was AI-generated.
Article sounds like it was written by AI. So people take death information and make it into a youtube video. How does that relate to spam in google's search? Domain squatting isn't what she is describing. It accounts for so little of the spam problem. If google has been working on this update since last year and say they need more time meanwhile the llm scene exploded and keep changing daily how much faith do we have in google. We expect their own AI products to detect others?
I have a friend who's entire job is to generate ChatGPT spam articles about random shit to corporate blogs to drive traffic to their app.
Tells people she "works in IT" lol I suspect they pay like shit as employees have to use their own computer even.
Time to reign this shit in.
I read Plato but do not remember. I need to read more, thank you random SF developer.
When searching for Robert Jenkins Park, I am not looking for Bob Jenkins the person. Google needs to stop aliasing, and stop dropping search terms.
Google values giving gibberish over giving nothing, and nothing they do will help with irrelevant results, until the revert back to 2008 search behaviour.
I'd hope adding quotes around Robert Jenkins Park would avoid aliasing. If so, then they optimized for aliasing being the default with a simple way to avoid it for the (presumably) uncommon case. If aliasing was off by default, what would be a similar syntax to indicate that you actually want it on for a search?
A global flag seems reasonable if enough people would want it, though that is dependent on users being authenticated with preferences stored on the server.