Episode 6 · Read the feedback: who reads you, and how AI understands you
Silan Hu8 minAfter you publish, feedback is what matters. Silan Viking counts visits on the server, so AI crawlers that never run scripts are recorded too. This episode covers three things: who came, what AI read, and what to change next.
1. The dashboard at a glance
The top of the dashboard summarises:
Metric | Meaning |
|---|---|
Views / Likes / Comments | page views, likes, comments |
Crawlers | all crawler requests |
AI crawlers | requests identified as AI crawlers, such as GPTBot, ClaudeBot, PerplexityBot |
Search bots | search engine crawlers, such as Googlebot and Bingbot |
AI chat | visits that arrive from AI chats such as ChatGPT or Perplexity |
Click any metric to see it broken down by page. Traffic detail next to it lists today's visits, top content, sources, and locations.
For silan.tech, as of 4 October 2026: 1,473 human interactions, 14,059 AI crawler requests, 19,686 search bot requests, and 31 visits from AI chats. Clicking AI crawlers shows the home page is what AI crawlers read most (13,930 requests), followed by individual articles.
2. By day and by crawler
The heat maps below are split into Release activity (content commits per day), Human traffic, Unique visitors, SEO traffic, and GEO traffic. Click a day to see who visited each page.
In GEO traffic, visits are grouped by AI crawler and labelled by purpose:
- Model training crawl: fetched to train a model;
- User-requested fetch: fetched on the spot because someone asked an AI about it;
- AI search indexing: fetched to build an AI search index.
On 4 October, for example, silan.tech had 148 AI visits: ClaudeBot made 110 training requests across robots.txt, sitemap.xml, and many pages; ByteDance's Bytespider made 27; OAI SearchBot made 5 for search indexing; ChatGPT User and Claude User fetched pages 3 times and once because of user questions; and GPTBot read sitemap.xml once for training.
3. A GEO check for each article
Open an article and click Run AI/GEO content check in the toolbar at the bottom right of the editor (hover to see each button's name). You get:
- a score from 0 to 100 and a grade: Draft, Needs structure, Ready with edits, or Strong;
- counts from the text: words, sections, questions, media, links;
- real feedback for that page: views, AI crawler requests, and AI chat referrals;
- prioritised suggestions, for example: P1 split the body with headings so answer engines can retrieve a focused passage; P2 add a question-shaped heading that mirrors what readers actually ask; P2 add attributable evidence links; and check after deploying that AI crawlers can find the page.
For example, a 79-word moment scores 34 (Draft), with suggestions to add sections, a question heading, and evidence links. Written up as the GEM-Bench article, it scores 88 (Strong) with one remaining suggestion: keep watching after it goes live.
4. From the CLI
silan stats sync /blog/gem-bench-ai-answers-with-ads
silan stats show /blog/gem-bench-ai-answers-with-ads
silan stats crawlers /blog/gem-bench-ai-answers-with-ads
silan stats sources /blog/gem-bench-ai-answers-with-adssync pulls the latest numbers from your live site; the other commands read the local cache. For the GEM-Bench article: 129 views and 4 likes; 101 human visits, 28 search crawler requests, and 4 AI crawler requests.
5. How to read it
- A crawler request shows that a page was fetched. It does not show that the page was indexed, understood, ranked, or cited.
- Crawlers are identified by their User-Agent, which can be faked.
- After reading the feedback, the next step might be a sharper opening, one more evidence link, another language, or nothing at all. Change it, publish, and look again: that is the loop.
Series wrap-up
You have now walked the whole loop: capture an idea, write it up in two languages, publish, deploy, and check, bring in an AI assistant, and improve from what people and AI actually read.
No comments yet