After years of Google trying to force-feed AI to me in web search, I have to admit that I’ve developed a taste for it — on some plates and in limited servings.
Based on how thoroughly “AI Overview” results have taken over Google, the company clearly sees AI as central to virtually every aspect of making sense of the web. You, however, are under no obligation to follow Google’s lead — or to use it at all, since such search competitors as Microsoft’s Bing and the privacy-focused DuckDuckGo still focus on traditional web links.
But if you, like most people, mostly depend on Google for search, you should still know what its implementation of AI is — and is not — good for.
Short answer: Google’s AI search features — and the Gemini chatbot you can use in your browser — do better at things that old-school Google has often done worse.
The answer you want isn’t in one place or just a few places on the web
The worst kind of traditional web search is the one that never ends — you click through to a series of pages and posts and find that they either contradict themselves or fail to answer your question.
Large language models, however, have gotten quite good at summarizing large bodies of knowledge. They can synthesize the collective observations of people to yield concise answers to things like how-to or best-place questions.
One recent example from my own history: Google provided a usable cheat sheet about how to get the woeful companion app for a Panasonic camera to geotag photos reliably, something the manual explained poorly.
But you don’t have to go to Google for this kind of AI assistance, either. Reddit and Yelp already do this well, using AI to sum up the wisdom of their respective hive minds. In Yelp’s case, that option isn’t the default output; you have to tap the “Assistant” button in its mobile app.
Your question requires a personalized answer that any one page may not deliver
This use case is a variation of the previous one: The help you’re seeking is not only scattered around the web but requires some extra interpretation. Consider, for example, a travel search matching this pattern: I’m in [place] for [time] and am interested in [thing].
To cite an actual query I put to Google’s Gemini chatbot a few weeks ago: “What’s a good itinerary for this afternoon in central Dublin for somebody who’s seen the usual sights like Trinity College and Dublin Castle?” Google suggested two museums I hadn’t visited before, including one that was barely two blocks from my hotel.
In this case, you do want the AI to get in the way of raw web results by filtering it. But you should still check the sites cited in its responses — a crucial reference that Google says it has been working to elevate in its AI Overview results.
Your question is convoluted enough to be conversation material
Google’s AI Overview, its more conversational AI Mode and its Gemini chatbot allow for a back-and-forth exchange that isn’t possible in a traditional interface unless you are well versed in search-syntax modifiers.
You can ask a complicated query, then follow up with questions to get more information. One example from a test of AI Mode this spring: I asked “What is a good native groundcover to plant in shady ground in Northern Virginia?” and then replied to AI Mode’s list of possible plants by asking where I could buy them around D.C.
(My failure to plant those ground covers is no reflection on the usefulness of these results.)
My gardening search ultimately took me outside Google’s walled garden, which is how web search is supposed to work — and what Google did strikingly well at its debut when competitors like Yahoo were trying to turn their search sites into full-service portals.
When to avoid AI
Google has been making itself a portal for years, and the way AI can at least seem to answer questions by itself has led to existential dread among web publishers that have seen their search traffic sink.
Seeing AI chatbots effectively summarize stories, some — for instance, the New York Times and my freelance client PCMag’s publisher Ziff Davis — have also sued AI firms, alleging copyright infringement. The Washington Post, meanwhile, has partnerships with OpenAI and Perplexity.
But relying on AI Overview summaries instead of going to other people’s sites can serve you poorly as well. When you’re looking for a specific page or data point, Google putting more text between you and that resource doesn’t help things.
Accepting an automatically generated version of a page or data point that’s only a click away raises multiple risks.
Sometimes, you’ll merely get left with an incomplete picture: asking “what hours do the Dulles people movers run from the A gates to the D gates?” got an AI Overview response Monday that didn’t list two daily gaps in that schedule highlighted in the first link shown below, a Reddit thread. At worst, AI results can present fraudsters vying to game Google’s results.
Google does, however, still let you move past its AI results to the traditional kind. So while that company may have decided that AI is its hammer, and all web searches are nails, you don’t have to use its tools that way. You also get to use your own — specifically, the space bar, the Page Down key and your thumb, all of which work to skip the AI slop.
The post When you should use Google’s AI for search — and when you should skip it appeared first on Washington Post.
