An interview with Ask Jeeves co-creator Garrett Gruener, who says his original conception was similar to the chatbots Microsoft and Google are trying to build
Charlie Warzel / The Atlantic :
Context & Ripple Effects
Before keyword search won the web, Ask Jeeves bet that people would type plain-English questions into a box and get an answer — and co-creator Garrett Gruener tells The Atlantic that the original conception maps onto what Microsoft and Google are racing to build with today's chatbots. The interview lands mid-cycle: Google's own AI chief had described moving "more conservatively than a small startup" on chatbots over reputational risk, even as the company pushes the same question-answer paradigm.
The gap between Gruener's 1990s vision and its current feasibility is also visible in the coverage around this story: niche publishers like Macworld and PCWorld launched scoped chatbots trained only on their own archives, while hardware startups such as Groq pitch chips that make chatbot answers nearly instantaneous — the latency and cost constraints that defeated the original natural-language search now being engineered away.
First-order effects
- Gruener and the Jeeves legacy gain renewed relevance as a prior-art narrative: the interview hands journalists and analysts a historical frame for judging whether Microsoft's and Google's chatbots are genuinely new or a rediscovery.
Second-order effects
- Google's deliberately cautious posture on chatbot rollout sits in tension with the ambition Gruener describes, pressuring the company to reconcile reputational-risk messaging with competitive claims about conversational search.
- Publishers watching the big platforms build answer engines have already begun hedging with their own closed-corpus chatbots like Smart Answers, keeping reader relationships inside their own archives rather than ceding Q&A traffic entirely.
Third-order effects
- If conversational Q&A becomes the default interface, search economics shift from ranked links to direct answers — the outcome Ask Jeeves gestured at decades ago — with value accruing to whoever controls the answering layer and the inference hardware beneath it.
- The pattern suggests failed paradigms from the early web can revive once compute costs fall far enough, meaning today's rejected interfaces (agentic browsing, fully conversational search) may simply be waiting on cheaper infrastructure.
The trend: Natural-language question answering — first commercialized as Ask Jeeves in the dot-com era and shelved when keyword search won — is being rebuilt end-to-end on modern AI by platform giants, chip startups, and individual publishers alike.