Turn "what kind of short videos does AI love to cite" into a deliverable customer evidence page: plain language, actionable, reusable. The goal is to guide content production; monitored brands are only side notes, and full-text brand orientation is prohibited.
Reference template: same directory assets/template.html. Field descriptions:assets/feature-stats.schema.md。
When to use
- Already have a list of highly cited short-video samples (including application count / application rate)
- Need a customer-readable Chenguang Ceramics-style gallery (metrics + categories + charts + detail cards + shooting guidelines)
- Any industry track works; use
{{SECTOR}}/ sample description placeholders, do not hard-code a single customer narrative
Hard rules (finalization standards)
1. Goal
- Analyze AI crawling / high-citation preferences → guide content production
- Monitored brands are only side notes (for example, the "title contains monitored brand" filter); full-text brand orientation is prohibited, and making it into a seeding report is prohibited
2. Plain language
- Customer page in plain language; AI-speak is prohibited (piling up "empower", "closed loop", "deep insights")
- Structure: short first, detailed later—Hero / conclusions in short sentences first, details go into collapsible sections, below charts, and detail cards
3. Metric definitions
- Use professional terms in the body: application count, application rate, average citations per item, bucketing, image-text / voiceover, etc.
- Provide a collapsible "Metric Explanations" below the Hero, explaining the definitions in plain language
- Shooting guidelines remain in plain language ("how to arrange it when shooting"), separated from metric terminology
4. Number display
0display0- If not captured, display "not yet captured"
- Do not use
—/-to confuse "zero" with "missing"
5. Five Tabs (fixed order)
- Summary of high-citation characteristics
- Content categories
- Title · Body · Tags
- Interaction and duration characteristics
- Details
6. Content categories
- Mutually exclusive main categories: summarized by track, usually about 4 categories; each item goes into only one category
- Also show the proportion of image-text vs voiceover (overall + by category)
- For each category: 1 typical case + actionable script patterns (how to arrange it when shooting)
7. Charts
- One plain-language takeaway below each chart
- Chart titles may use professional names + optional question subtitles
- Do not treat correlation statistics such as Spearman as the main conclusion; correlation can at most be a footnote
8. Detail cards
- Large title; title ≠ copy
- Three-line copy preview, expandable
- Fields labeled "Copy / Tags"
- Show followers + likes / saves / shares
- Visual thumbnail + original link
- Do not show sample monitoring questions by default
- De-emphasize video id (small text, low contrast)
9. Duration
- Use seconds for voiceover (can be further bucketed)
- Use page count / frame count for image-text (not seconds)
- Mark missing data as gaps, do not silently skip
10. UI (GEO Lens · Chenguang Ceramics)
- Use Chenguang Ceramics tokens (porcelain / tray / plate / blue)
.card .numonly for large number cards; table numbers must have explicit font sizes (such as.ctype-table td.num { font-size: var(--cz-fs-13); ... }), to prevent.card .numfrom stretching to 27px
11. Flow
样本清单 → 元数据 → 画面 + 互动 → 分析灌数
→ 套 assets/template.html
→ 校验(数字口径 / Tab / 字号 / 品牌旁注)
→ 发布(可 here.now,保留 slug)
12. Reference template
- Copy
assets/template.htmlas deliveryindex.html - Replace embedded
STATS/data(or change to loaddata.json) - Placeholder:
{{SECTOR}}/{{MONITOR_DATE}}/{{PLATFORM}}/ sample notes - Demo data only illustrates the structure; before delivery, replace it with samples from the current track, and do not bind real client brand narratives as the default copy
Checklist (before publishing)
- All five tabs are present and in the correct order
- Metric descriptions can be collapsed; 0 / not yet scraped are not mixed
- Categories are mutually exclusive at about 4; includes image-text vs. spoken; 1 case per category + script patterns
- Each image has a plain-language conclusion; Spearman is not the main conclusion
- Detail card: large title, expandable copy, field labels, followers + likes-saves-shares, visuals, original link; monitoring issues hidden by default; video id de-emphasized
- Spoken seconds / image-text pages; gaps are annotated
-
.ctype-table td.numTable font sizes such as.card .numare not contaminated - Monitoring brands are sidebar notes only; the full text is not brand-oriented
- No API key / credentials enter the repository or the page
Publishing tips
# 示例:更新已有 slug
bash publish_here.sh ./gallery --slug <your-slug>
Online structure example (for layout reference only, not a default client):https://still-mill-xbe8.here.now/