Now: Archive

ChatGPT’s search rebuild replaced the JSON that FanoutFox reads with a pipe language, so the fan-out card sat empty on fresh sessions. The patch speaks the new format and is live on the Chrome Web Store, as promised in the teardown.

This site charges AI agents a penny per page over HTTP 402, and Claude Code paid it, with the settlements on a public chain. The writeup is live and the demo stays up at paid.suganthan.com.

A library of real shipped screens, now the first stop for every design decision on this site. The September component pass (related post rows, tool cards, the enquiry form, the feedback widget) came out of benchmarking against it.

web.run swapped its fan-out JSON for a pipe language in mid-August, and I read the new format line by line. Earlier in the month, the shortlist post showed ChatGPT often knows who’s in the running before it searches. Both sets of facts live in the research tracker, re-verified as the engines change.

Built and shipped alongside the AI Mode tracking guide. Paste your Search Console queries and a model separates ordinary keywords from the conversational queries leaking out of AI Mode.

Built and shipped. Reads 12 months of Common Crawl history for any domain or single page, dates the month a CCBot block appeared, and tells you whether a CDN template wrote it. The launch post covers why that archive decides whether you exist in AI training data.

Google started wrapping organic result links in /goto encryption for logged out sessions. I took the format apart and found a way to recover every URL in one pass. The note has the structure. The method stays private.

The teardown method packaged as a free Chrome extension, live on the Chrome Web Store. It reads your own ChatGPT and Perplexity sessions and shows the fan-out queries, the sources fetched, who got cited and who got mentioned. The launch story.

Read the raw answer stream to see how Perplexity picks its sources. It behaves nothing like ChatGPT. The findings live in the AI SEO research tracker, where I re-verify them as the engines change.

Read the raw network traffic to see how ChatGPT actually picks its sources. Part 2 followed when it changed the pipeline within weeks of the first post.

Ran it while writing the agent-ready guide. Good first-pass protocol scanner: robots, sitemap, Link headers, well-known files, Markdown negotiation, MCP, Web Bot Auth. Worth keeping as a quick first check.

Shipped a free, open skill that joins Search Console and GA4 in BigQuery to estimate revenue per organic keyword, with a confidence score on every row. The honest answer to “which keyword actually made money” once the query is hidden in (not provided). On GitHub.

Works really well. Already implemented across a few internal Snippet Digital tools. Read the note.

Implemented and working on this site at /mcp/. Be honest: getting it set up was painful. Documentation gaps, the widget paste UX is rough, and you fight the host page CSS. Once it works, it works. Still waiting for the W3C Chrome spec to mature so the widget step disappears.

Tested it. Then Anthropic shipped Claude design and the gap closed. Using Claude design now for the same workflow. Might still write a comparison guide.

The piping experiment turned into a content cluster: BigQuery MCP server with 32 tools, 6 GA4 + GSC blending tools, and the multi-signal attribution skill with a confidence score on every row. All free, all open source.

44-day Phase 1 experiment shipped. 1,421 markdown requests, growing volume, Claude as the #2 consumer at 500 hits, three brand-new markdown-specific user agents appeared during the test. Phase 2 is citation tracking via Ahrefs Brand Radar, when the budget allows.

Tested. Useful. Recommended. Reverse engineers Google’s SynthID watermark on Gemini-generated images using spectral analysis, no encoder access required. Holds up well enough to be in the toolbox.

Mixed verdict. Bloom is genuinely nicer for local recording and the cost story is unbeatable (free vs Loom subscriptions). For sharing, transcripts, and the polish layer Loom still wins, so I haven’t replaced it. Bloom for personal walkthroughs, Loom for client-facing.

Genuinely impressive. Makes managing Workspace feel like it should have always worked this way. Still using it daily.

Interesting adaptive scraping library for Python. But Firecrawl and ScrapingBee already cover everything I need. No gap to fill.

Open source AI second brain. Promising concept, but OpenClaw handles the personal knowledge layer better for my workflow.