Number Research
← Back to Explore
Communication Studies Media Studies Algorithmic Recommendation Systems

Ask Jeeves Died the Month Google Finally Became Him

Pew data show AI summaries cut click-through rates by nearly half — even as Google's search volume hits all-time highs

S.J. Nam 9 min read
Ask Jeeves Died the Month Google Finally Became Him

On May 1, 2026, IAC quietly closed Ask.com — the site the world once knew as Ask Jeeves, its butler mascot fielding plain-English questions since a full public launch
on June 1, 1997
. IAC thanked "the millions of users who turned to us for answers," and a TechRadar retrospective could not resist the obvious irony:
it's interesting that as Google and other AI-led companies try to make web search like a natural conversation again, the site that first pioneered the approach is closing
. Ask Jeeves died in the exact month that its founding premise — type a real question, get a real answer, skip the links — became the default behavior of the world's largest search engine. That's not a coincidence worth a chuckle and a footnote. It's the whole story of search's user-facing evolution, compressed into one obituary.

The conventional telling goes: keyword matching gave way to semantic understanding, which gave way to conversational AI. True, but beside the point. The user-facing history of search isn't a story of engines getting smarter. It's a story of who does the interpretive work of turning a question into an answer — and, more consequentially, of what happens to the economics of the web once that work moves from the user's browser tab into a black box the user never leaves.

The Butler Behind the Curtain

Ask Jeeves's 1997 pitch was radical for the era: type "What's the weather in New York?" instead of composing a query in the stripped-down pidgin — quotes, plus signs, Boolean operators — that engines like AltaVista demanded. AltaVista, which had launched in
December 1995
with a then-staggering
database of 16 million pages
, rewarded users who mastered its syntax:
Boolean operators — AND, OR, NOT — let you build complex queries, and you could search for exact phrases using quotation marks
. That was a real literacy, one an entire generation of early web users had to learn, the way you'd learn to file a card in a library catalog.

Jeeves promised to make that literacy obsolete. But the mechanism behind the butler wasn't language understanding at all. As a contemporaneous account of the era's search engines put it bluntly,
the secret to Ask Jeeves wasn't really the ability to understand language — instead, Ask Jeeves had over 100 editors monitoring what people searched for, then hand-selecting sites that seemed to best answer those queries
. A separate retrospective adds texture: co-founder David Warthen built the system around
a knowledge base of over five million templated questions
, with unmatched queries falling back to a metasearch across other engines. It's worth pausing on how closely that two-step design anticipates the architecture underneath today's AI answers, where
template matching followed by metasearch foreshadowed modern hybrid-search architectures and retrieval-augmented generation (RAG) systems
. Jeeves wasn't wrong about what users wanted. He just didn't have the machinery to deliver it at scale, so the whole apparatus curdled into a punchline — a chatty mascot standing in front of human editors and a lookup table, twenty-five years before anyone would call that "the wizard of Oz problem" in AI product design.

Google's Genius Was Modesty

Google's founding trade, in 1998, was almost the opposite of Jeeves's.
When Google launched in 1998, its clean, minimalist interface with simple blue links stood in stark contrast to the cluttered portals of early search engines like Yahoo and AltaVista.
Rather than promising an answer, Google promised a ranked set of doors and left the interpretation to you.
For users, it provided a consistent, scannable interface where they could quickly evaluate options; for search engines, it established a clear value proposition — we'll find the best pages, but you'll need to click through to get the information.
That modesty was the innovation, not the PageRank math underneath it. By refusing to claim authority over the answer itself, Google made itself trustworthy, and by forcing the click, it manufactured the very unit — the pageview — that would fund two decades of ad-supported publishing. The reader did the last mile of judgment: which of these ten links actually answers my question, and whom do I believe? That labor was tedious, but it was also the mechanism by which the open web got paid.

The Bargain Breaks

The reversal back toward Jeeves's original ambition began quietly and then didn't. Google previewed a generative rewrite of its results page, the Search Generative Experience,
announced at Google I/O in May 2023
, then pushed it to everyone as "AI Overviews" the following year — a rollout so bumpy that SparkToro's Rand Fishkin described the industry's whiplash over
May's awkward US rollout on May 14 and rollback on May 30
. The feature came back and kept expanding: independent tracking found AI Overviews present on
12.7% of US Google results
by that June, climbing toward roughly a fifth of queries within two years and, in Google's newly launched "AI Mode," pushing zero-click rates as high as 93% by some measures.

The user-experience effect is not subtle. Pew Research Center, tracking
68,000 real search queries, found that users clicked on results 8% of the time when AI summaries appeared, compared to 15% without them — a 46.7% relative reduction
. Google Search Console data compiled by Digital Content Next tells the same story from a different angle:
the average click-through rate for the #1 result on AI Overview keywords fell from 7.3% in March 2024 to just 2.6% a year later, a 34.5% decrease
. Ahrefs' later audit of 300,000 keywords, comparing December 2023 to December 2025, found the effect compounding rather than fading. None of this is abstract for the people who built businesses on that traffic. Chegg told regulators it saw a
49% decline in non-subscriber traffic between January 2024 and January 2025, coinciding with AI Overviews answering the homework questions that used to send students its way
. Helen Havlak, publisher of The Verge, was more blunt with NPR:
"The extinction-level event is already here. And a bunch of small publishers have already gone out of business."
A media lawsuit filed by Penske put the structural shift in a single sentence, accusing Google of pivoting
"from being a search engine (that sends traffic to websites) to an answer engine that removes the incentive for users to click to visit a website."

More Searches, Less Web

Here is the detail that most coverage of this shift skips past, and it's the genuinely counterintuitive part: none of this looks, from Google's side of the ledger, like search declining. Even in the immediate aftermath of the AI Overviews rollout, SparkToro's data showed
Google's market share remaining robust and the number of searches per searcher at historic highs — countering the narrative that users are abandoning Google for alternatives like ChatGPT, Perplexity, or Microsoft's Bing
. People are not searching less. They are searching more, and getting less out of each search in terms of where it sends them. That is the real inversion, and it's easy to miss if you frame the story as "AI killed search" or "AI improved search." Neither framing fits. What's happened is a decoupling of query volume from web traffic — Google can process record numbers of questions while the open web, the thing search was originally built to help you find, receives a shrinking fraction of the attention those questions generate. Google disputes the severity of the publisher numbers, saying it still sends billions of clicks daily and that AI Overview clicks are higher-quality when they happen. Both things can be true: aggregate traffic can remain large in absolute terms while its distribution collapses toward a handful of already-dominant sources, leaving smaller, specialized sites — the ones a directory or a ranked list of blue links used to surface — starved out first.

What the Butler Knew

Ask Jeeves failed in 2001 not because plain-English search was a bad idea, but because the only way to deliver it at the time was to hide human editors and a template library behind a cartoon valet, and users eventually noticed the seams. Twenty-five years of language-model progress finally supplied the missing machinery, and it arrived at the same moment Google needed a new answer to a newer problem: how to keep users inside its own product long enough to keep selling ads against them. The seams haven't disappeared; they've just moved. AI Overviews still get their sentences from someone else's website, they just no longer send anyone there to check. The question worth sitting with isn't whether AI-generated answers are more accurate than ten blue links — often they are, often they aren't. It's whether an "engine" that no longer wants to send you anywhere still deserves the word "search," or whether we've quietly agreed to rename the front door of the internet as its terminus, and to call the new arrangement progress because the butler, at least, finally learned to talk.

References