theomnikit.com 按搜尋意圖推出 700 個工具頁面
Shipping 700 tool pages with one page per search intent
theomnikit.com 為不同搜尋意圖建立各自的工具頁面,並以獨立內容、預設設定和結構化資料區分頁面;其 /llms.txt 列出 768 項工具及 1,035 個頁面。
I build theomnikit.com, free browser-based tools: file converters, developer utilities and money calculators. Its /llms.txt counts 768 tools and 1,035 pages written for one specific job. Here is how those job pages are decided, what we cut, and what Search Console showed.
I wrote this with AI help and checked every claim against the repo: the rules are quoted from our playbook and decision log, the code from the tool's meta.ts, and every Search Console figure from one research note dated 20 November 2026.
The problem: one tool, many searches
One piece of code often does several jobs, and people search for each job in their own words. Our Image to PDF tool also answers "jpg to pdf", "heic to pdf" and "photo to pdf". A single page titled "Image to PDF" matches none of those phrases exactly. A page per phrase, with the same body, is the near-duplicate pattern search engines decline to index.
The rule
The playbook (docs/playbooks/discoverability.md) treats SEO, answer boxes and AI assistants as one job. Four lines do most of the work:
One page per search intent, and the exact phrasing people type in the H1, the first sentence and the URL (
webp-to-png, notimage-converter?to=png).Under every question-shaped heading (H2/H3), the first paragraph is a complete answer of 40-60 words that stands alone without the question, the page or the tool.
Alias pages are real pages with their own words, never redirects or duplicates
/llms.txtmaps every live tool, its one-line job and the facts that decide a recommendation
The checklist adds structured data on every tool page. In code, an intent page is an entry in its tool's meta.ts; its presets open the tool set up for the searched job:
aliases: [
{
slug: 'heic-to-pdf',
name: 'HEIC to PDF',
tagline: 'Convert iPhone HEIC photos straight into a PDF',
keywords: ['heic to pdf', 'iphone photo to pdf', 'heif to pdf' /* ... */],
presets: { from: 'heic' },
intro: 'iPhones save photos as HEIC, which most forms and portals refuse. ...',
},
],
One thing we got wrong: intent pages repeated their parent's FAQ markup. A comment in ToolPageView.tsx records the 25 September measurement: 489 landing pages, each repeating the parent's FAQ, and on average 91 per cent of a landing page's body text shared with its parent. Now the FAQ and HowTo are declared once, on the tool's own page, and an intent page declares only questions it wrote for itself. The intent pages below still carry SoftwareApplication and BreadcrumbList markup.
The fold
The rule had made too many pages. The decision log entry for 27 September 2026 opens:
Run by
scripts/qa/fold-candidates.tsover the 151 alias pages Google listed as "Discovered - currently not indexed". 108 folded (301 to the parent), 43 kept.
Its first run proposed folding pages it should not have, so the rule gained amendments:
A named scheme, a comparison or a decision is a different subject, not a synonym.
The second kept ten pages on a written list, among them pdf-to-jpg, heic-to-pdf and home-loan-emi-calculator, because the rule had no search-volume input. And the tie-breaker:
Keeping is reversible and folding a head query is not, so every doubt went to "keep".
A fold does not erase the phrase: "A folded page's name is printed on its parent ("Also works as: ..."), its phrase still finds the parent in on-site search, and it leaves the sitemap, llms.txt and the assistant feed." More batches followed in October, each a generated file of redirects. Today the Image to PDF page reads "Also works as: PNG to PDF · Photo to PDF · Multiple Images to PDF · WebP to PDF", and each of those four old URLs answers with a 301 to it.
What Search Console showed
The research note (docs/data/tool-demand-research-2026-11-20.md) covers 14 September to 18 November 2026. Totals: 214 clicks, 92.4k impressions, CTR 0.2%, average position 45.1. Small numbers for a young domain, and one reading rather than a before and after, so I can't credit any change to the fold. What it does show about intent pages:
-
/files/extract-videos-from-zip, an intent page of the Folder & ZIP Sorter: 12 clicks from 162 impressions at position 9.0, the best click-through rate on the site. -
/files/word-count-checker, an intent page of Word Counter: 869 impressions, with the query "word count checker" at position 9.7, and zero clicks. - The note's reading of pages like that: "The pages already on page 1 with zero clicks are a title/snippet problem, not a ranking problem".
So an intent page can reach page 1 for its own phrase; getting the click is a separate problem.
Five pairs, live today
| Tool page | Intent page | What it adds |
|---|---|---|
| Image to PDF | HEIC to PDF | Opens set for HEIC; its own table of twelve 4032 x 3024 iPhone photos by layout on A4 and Letter |
| PDF to Image | PDF to JPG | JPG preset; kept on the fold rule's written list |
| Word Counter | Word Count Checker | Written for checking a draft, PDF or Word file against a word limit |
| Folder & ZIP Sorter | Extract Videos from ZIP | Keeps only the videos; lists a ZIP's contents without unpacking it |
| India Income Tax Calculator | Old vs New Tax Regime Calculator | A comparison, which the first amendment treats as a different subject; presets a 15 lakh salary with deductions |
What I would keep
- An intent page earns its URL with something the parent lacks: a preset, a table, its own questions. A new H1 alone is not enough.
- Declare shared FAQ and HowTo markup once, on the parent.
- Put the fold rule in a script, keep its exceptions on a written list, and record both in the decision log.
- Expect page 1 without clicks. The snippet is the next job.
來源:DEV Community · dev.to