Writing
Research and notes from building CueFox. Every claim carries a source you can check.
- We built cookie consent before we had a single cookie
CueFox sets no cookies and runs no analytics. We shipped a consent mechanism anyway, because the alternative is installing a tool first and asking permission afterwards.
- How we read 92 megabytes of competitor content without skimming
The method behind our corpus audit: what we fetched, how pages were grouped, the three measurements we ran, and the specific things this data cannot tell anyone.
- Four bulk drops: what the publish dates in 3,038 pages reveal
217 pages on one day. 102 on another. 57 on a third. We pulled the publication date off every page in the corpus and the release pattern is not a ramp, it is a series of dumps.
- Reading time, and other small lies a website tells you
The reading time on this post is computed from the text. The last-modified dates in our sitemap are absent rather than stamped with today. Small honesty is mostly a series of defaults you decline.
- The 552-page tell: what happens when a comparison page is arithmetic
One competitor runs 552 pages comparing screen recorders. It is not 552 articles. It is 187 tool pairs multiplied by three fixed angles, and the multiplication is visible in the URLs.
- The citation gap: 12.4 sources a page, or 0.15
Across 3,038 competitor pages, the single measurement that separated researched writing from manufactured writing was how often a page linked to something outside its own domain.
- Every comparison page we publish must name something a rival does better
It is a required field in our schema, at least sixty characters, and no comparison page builds without it. Three reasons, one of which is legal.
- We built a gate that refuses to publish our own pages
Our content pipeline currently reports zero publishable pages out of fifteen records. That is the system working. Here is what it checks and why it is a function rather than a checklist.
- The laptop-model problem: 99 pages about one keyboard shortcut
A competitor publishes how to screen record on an Acer Swift, an Acer TravelMate, an Acer Nitro and 96 more. The task does not change. Only the noun does, and that is the whole pattern.
- We kept a 225KB HTML file forever, and it has caught every regression since
The original hand-written landing page is still in the repository as the permanent reference. Every build compares against it. Deliberate changes are recorded one at a time, with reasons.
- The orphan page problem, and why we derive links instead of writing them
Half the pages in a programmatic site end up with no inbound links, because humans author links in one direction and then stop. The fix is to stop authoring them.
- Two-stage validation: let the schema accept drafts, let the gate decide URLs
Most content systems have one validation step, which forces a choice between blocking half-finished work and shipping it. Splitting it in two removes the choice entirely.
- We reversed our biggest content decision after reading the data
We chose task-level tutorials as our volume format partly because competitors had converged there. Then we read their pages and found the convergence pointed at the cheapest content in the market.
- What a 186-word tutorial page actually looks like
One competitor publishes 115 tutorial pages at a median of 186 words. They score lowest in the corpus on every measure we applied, and none of them is about that company's product.
- What we would have to see before publishing a thousand pages
Our plan has a ceiling of roughly 1,200 pages in year one, and a gate at week four that stops everything. These are the conditions we would need to meet to keep going.
- Loom is targeted by all ten. Colossyan is targeted by one.
We counted how many rival vendors publish a comparison page against each tool in this market. The distribution is extremely lopsided, and the crowded end is the worthless end.
- Why duplicate detection misses generated text, and what we use instead
We ran our own thin-content check across 3,038 competitor pages. It caught five clusters cleanly and completely missed the largest one. Here is the measurement, and the replacement.
- Why there is no pricing page, and why the comparison tables say so
We have not decided what CueFox costs. Every competitor comparison we publish renders an explicit 'not published' cell rather than a plausible number, and that is enforced in code.
- Word count is not quality: two 3,000-word clusters, opposite substance
The longest pages in our corpus include some of the most substantial and some of the most padded. We measured what is left after removing everything a page shares with its siblings.
- We read 3,038 pages of our competitors' content. Most of it is furniture.
Before writing a single marketing page, we measured every indexable page across ten demo-video companies. The largest content cluster in the market is 552 pages built from one template, and the usual tests for thin content do not catch it.