Kagi Changelog 2/13: Faster and more accurate instant answers and Wikipedia page
Kagi, a paid, privacy-focused search engine, is drawing both praise and criticism as it rolls out faster instant answers, Wikipedia and Wolfram Alpha integrations, and bundled AI assistants. Many users report better technical and less spammy results than Google or Bing, valuing features like domain blocking, bangs, and honest “no results” responses, but others say quality has recently declined, especially for non‑English, local, maps, shopping, and mobile use. The thread also surfaces concerns around Kagi’s account-linked model and security posture, alongside admiration for its rapid development pace and responsive handling of user feedback.
Overall Search Quality
- Many report Kagi now outperforms Google for technical and general web search, with fewer irrelevant or spammy results and useful “no results” behavior when nothing matches.
- Others say quality has declined in the last 1–2 months, especially for developer queries and deeper research, sometimes mirroring Google or surfacing spam/hacked/warez sites.
- Some note a turning point after Brave results were added. Others have not noticed any degradation and use Kagi almost exclusively.
- There is recurring frustration with “smart” query interpretation (ignoring terms, loosening constraints), with users wanting a more 2000s-style, literal search.
Privacy, Security, and Accounts
- Critics argue Kagi “does not take security and privacy seriously,” citing:
- Reported HTML injection / XSS-adjacent bugs and a permissive CSP that allows injected forms to affect settings.
- Required login and paid accounts tying searches to identifiable users, despite claims that searches are not logged per account.
- Supporters counter that:
- It’s still better than ad-based engines that monetize behavioral data.
- Anonymous-ish payment is possible via cryptocurrency.
- Several posters emphasize that privacy is a spectrum; some are uncomfortable with mandatory accounts, others prefer paying over being the product.
Features and LLM Integration
- Users praise:
- Quick Answer/instant answers with inline citations, backed by LLMs and now Wolfram Alpha.
- The ability to inspect sources mitigates (but doesn’t eliminate) hallucination concerns.
- Domain block lists and per-site ranking as major value-adds.
- “Ultimate” plan bundling chat, code, and research assistants with multiple LLMs (GPT‑4, Claude, etc.).
- Some dislike the conversational “human-like” tone and prefer data-forward displays.
Localization, Maps, and Special Use Cases
- Strong complaints about local search: storefront locales, business hours, and maps are notably weaker than Google’s.
- Non-English and regional results are mixed:
- Some say local-language results recently regressed or English is over‑prioritized, even with region/language set.
- Others report big improvements and rely on language/region filters and bangs (e.g.,
!de,!fr).
- Scholarly search is generally viewed as good, especially when boosting domains like PubMed.
Performance and UX
- Speed is good for many, but some users in Europe (e.g., Cyprus, Poland) find it slower than Google.
- Mobile/web UX complaints include:
- Flaky iOS/Safari integration, reliance on extensions, and lack of a simple dedicated Kagi app.
- Occasional buggy behavior on mobile (cursor issues, freezing).
- Others highlight positive experiences with Kagi’s responsiveness to bug reports (e.g., legacy Firefox ESR JS issues).
Business Model and Trust
- Posters admire the rapid changelog cadence and small-team productivity.
- Some fear eventual acquisition (especially by a large platform company), though Kagi representatives reiterate no plans to sell and note they’re near sustainability.
- Debate continues over whether Kagi’s paid, logged-in model is the “least bad” tradeoff compared with ad-powered incumbents.