Stop Using Google Search Right Now...

Main argument

Google Search has degraded: more ads and sponsored links, AI-rewritten headlines, and AI-generated low-quality content dominating top results. The presenter recommends avoiding big search engines and switching to self-hosted/local solutions for cleaner results and better privacy.

Problems described

• Google rewriting publisher headlines and changing how publishers’ content appears in search.
• Ads and sponsored links frequently appear before relevant results; sometimes malicious sponsored links appear (example: an OBS Studio incident).
• Proliferation of AI-generated or AI-regurgitated pages causing hallucinations and low-quality search results. (The summary cites industry analysis — e.g., Imperva “bad bot”/traffic reports — suggesting a large fraction of web traffic is automated/AI-driven.)
• Centralized search engines incentivize clickbait and algorithmic behaviors that amplify low-quality or AI-produced “slop.”

Alternatives and projects discussed

• Kagi — paid search engine (no ads). Presenter: good, but not worth paying for given free self-hosted options.
• DGoog / DGoo — search engine aggregator with a plugin/extension system; runs in Docker. Demonstrated with a hands-on install.
• SearXNG — open-source, privacy-focused search aggregator; available as a Docker container. Supports configurable output formats (JSON/HTML) for AI integrations.
• Local LLMs — example: Qwen 3.5 run locally in LM Studio to avoid sending queries to big providers.
• Perplexica — project for running Perplexity-like searches locally.
• Privacy-first search services mentioned: Startpage, DuckDuckGo, Brave Search (presenter prefers self-hosted alternatives).

Tutorial / guide highlights (high level)

Requirements:

• Virtualization enabled in BIOS (if needed).
• Docker (or Docker Desktop on Windows/Mac).
• Docker Compose.

Linux example (high level):

• Install Docker & Docker Compose (package-manager specific).

  • Example (Arch/Manjaro): sh sudo pacman -S docker docker-compose

• Enable and start Docker: sh sudo systemctl enable --now docker

• Add your user to the docker group (then reboot or re-login): sh sudo usermod -aG docker $USER

Dgoo (DGoo) quick install (high level):

1. Create a data directory: sh mkdir -p ~/dgoo/data

2. Create a docker-compose.yml in ~/dgoo with the project’s compose content.

3. Ensure permissions: sh sudo chown 1000:1000 ~/dgoo/data

4. Start the stack: sh cd ~/dgoo docker compose up -d

5. Visit the UI at: http://localhost:4444

6. In the UI: disable the Google engine (to avoid violating Google TOS), and enable desired engines and plugins (math, TMDB, Internet Archive, etc.).

SearXNG quick install (high level):

• Pull the Docker image (follow commands on searxng.org).

• Create directories for config and data, for example: sh mkdir -p ~/searxng/config ~/searxng/data

• Edit settings.yml to add output formats like HTML and JSON (useful for AI integrations).

• Start the container and visit the interface at: http://localhost:8888

Integrating a local LLM with SearXNG (high level):

• Run a local model (e.g., Qwen 3.5 in LM Studio) on a local port.
• In SearXNG developer settings, point the “MCP JSON” / API JSON entry at the local model endpoint.
• JSON output allows automated tools/LLMs to parse results and combine local LLM outputs with search results.

Notes:

• These containers are lightweight enough to run on many laptops and edge devices.
• Keep in mind permissions and port settings when deploying locally.

Analysis and recommendations

• Self-hosted search aggregators (Dgoo, SearXNG) can remove sponsored content and restore useful top results by aggregating multiple sources and stripping ads.
• Running a local LLM alongside a local search aggregator provides private, AI-augmented search without sending data to major providers. For many typical tasks, this covers a large portion of needs (estimated 80–90% of common tasks).
• Caveats:
  • Self-hosting is not a perfect privacy solution (a single instance can still be fingerprinted).
  • Local LLMs can regurgitate AI-written web content and may hallucinate; verification remains necessary.
• Encouragement: resist centralized, AI-driven search ecosystems and use self-hosted tools to preserve a healthier web experience.

Mentioned examples and sources

• The Verge (examples of Google rewriting headlines)
• Kagi
• DGoog / DGoo
• SearXNG
• Qwen 3.5
• LM Studio
• Perplexica
• Startpage, DuckDuckGo, Brave Search / Brave browser
• Imperva “bad bot” report (traffic/AI bot statistics)
• OBS Studio incident (malicious sponsored link example)

Main speaker / source

• Mudahar (the presenter / YouTuber)