Data Use Policy
Version 1.0 — August 9, 2026
Json House publishes machine-readable datasets alongside every post. This page states what we ask of anyone using them, and what we commit to in return. Both halves matter: a policy that only makes demands has not earned the right to make them.
This policy covers:
https://www.jsonhouse.com/data/{slug}.json— per-post structured datasetshttps://www.jsonhouse.com/data/index.json— the dataset cataloguehttps://www.jsonhouse.com/api/posts.json— the post indexhttps://www.jsonhouse.com/llms.txt— the machine-readable entry point
It does not cover post prose, which is a separate work.
Attribution is required
You may cite, analyse, redistribute, and generate answers from these datasets. Attribution is required in every case.
Preferred form:
Json House, "{title}", https://www.jsonhouse.com/data/{slug}.json (retrieved YYYY-MM-DD) Every dataset carries a ready-to-use attribution block, so nothing has to be assembled by hand:
"attribution": { "source": "Json House", "source_url": "https://www.jsonhouse.com/posts/{slug}/", "dataset_url": "https://www.jsonhouse.com/data/{slug}.json", "citation": "Json House, \"{title}\", jsonhouse.com (2026-08-03)", "attribution_required": true, "terms_url": "https://www.jsonhouse.com/data-policy/" } If you use this data in a generated answer, the answer should name either jsonhouse.com or the dataset URL. “According to one blog” is not attribution.
Allowed
- Citation and excerpting for citation
- Analysis and derived figures
- Answer and summary generation
- Redistribution that keeps the attribution intact
Reserved — ask first
- Bulk mirroring with attribution stripped. Full-catalogue harvesting already happens against this site; this clause is about that
- Republishing our measurements as your own primary source
- Commercial bulk redistribution
No licence is declared
This page requires attribution. It does not grant a licence, and our datasets carry no license field. That is deliberate.
These datasets are largely factual measurements, and facts attract weak copyright protection in most jurisdictions — a licence label would promise an enforceability that does not exist. Standard open licences are also irrevocable for copies already distributed, and we are not ready to make that call permanently. What we actually want is not litigation but citation, and citation is won with machine-readable provenance rather than a licence badge.
What we commit to
Stated cadence, kept. Every dataset declares how often it is refreshed. We raise that cadence rather than lower it, and only declare what we can sustain.
No silent overwrites. When a figure changes, the change is recorded. We do not retroactively edit past snapshots to look as though they were always correct.
Stated measurement limits. Where a number is a floor rather than a total, we say so. Where instrumentation began partway through a series, we give the date and leave earlier rows empty instead of inferring them. Where we excluded something — our own verification requests, for instance — we say what and why.
Published methodology. Measurement datasets ship with an account of how the measurement was made. A figure nobody can reproduce is an assertion, not a source.
Privacy rules we hold ourselves to
| Item | Rule |
|---|---|
| Client IP | Never stored. Country and network operator only |
cookie / authorization headers | Stripped before any capture |
| Human visits arriving from AI answers | Published as counts only, never joined with country, user agent, or timestamp |
| Ordinary visitors | Not logged at all |
Small samples make re-identification easier, not harder. A handful of referral visits combined with country and user agent could single out a person, so those fields never leave together.
How we write about what we observe
A user agent string can be set by anyone. When a request claims to be a given company’s crawler, that claim is not evidence, and we do not report it as one.
We name a company only on the basis of signals a client cannot forge — Cloudflare’s verified-bot classification, or the originating network. When an identity turns out to be forged, the named company is the party impersonated, and our writing says so.
Changes
This policy is versioned. Changes are dated and explained, and are not applied retroactively to data already distributed. Tightened conditions take effect from the next refresh, not backwards.
Questions
Anything not covered here — bulk access, unusual use, corrections to a dataset — see the contact details on the about page.