How this site is made
What is written by staff, what is drafted by AI, and what checks it.
UnGovr and LexLint publish across 200+ countries. Some of the text on these pages is drafted or corrected with AI tools working from sources UnGovr staff selected. This page says, site by site, what that means: what a person wrote, what a machine drafted, and what checks the result before a reader sees it.
Who makes this
UnGovr staff choose, write the rules, and answer for it.
UnGovr staff choose what is covered, write the rules every pipeline follows, select the sources a page or a tool is allowed to draw on, and answer for what is published. No pipeline picks its own sources or its own rules; a person wrote both before any AI tool ran.
UnGovr holds editorial responsibility for every UnGovr and LexLint site. UnGovr is a nonprofit organization based at 19 E Canon Perdido St, Suite B, Santa Barbara, CA 93101. Write to hello@ungovr.org with a question about anything on this page. For a correction or a removal request, write to privacy@ungovr.org instead.
What AI tools do here
What they do, and what they do not.
Under instructions staff wrote, working only from sources staff selected, AI tools on these sites draft and correct text, transcribe recordings, classify records, and select which news stories to carry.
They do not choose what is published. They do not make any decision about a person. No AI system on any UnGovr or LexLint site talks to a visitor; the one exception, UnGovr Request's voice assistant, already says so when a caller reaches it.
Site by site
What each site generates, and what checks it.
"Machine" means an automated check: a citation is fetched and compared against its primary source, or a model judges whether the cited text actually supports the claim drawn from it. "Recorded editorial review" means a named UnGovr staff member has checked that exact text against its source and the page says so, with a date; a later AI rewrite of that text brings the note back until the text is reviewed again.
| Site | What is machine-produced | What is checked, and how | Where the page says so |
|---|---|---|---|
| www.ungovr.org Insights | Which category a records request falls into, drafted by a locally run classifier. | Machine: classification accuracy is measured against a hand-labeled sample before an edition is allowed to publish. | The method note on each Insights report. |
| law.ungovr.org | Jurisdiction and instrument summaries drafted from the cited law. | Machine, for the research corpus: each citation is fetched and compared with its primary source, and a model judges whether the source supports the claim. Older summaries were checked by sampled review. Recorded editorial review, once a named reviewer has checked a summary against its source, shown on the page with the review date. | The note above a summary while it is unreviewed, and the review byline once it is reviewed. |
| lexlint.org | Every law note, drafted from the cited instrument. | Machine, the same per-citation check as the law.ungovr.org research corpus: each citation is fetched and compared with its primary source, and a model judges whether the source supports the claim. Recorded editorial review, the same as law.ungovr.org. | The note above a law note while it is unreviewed, and the review byline once it is reviewed. |
| The five Monitors (sheriffoversight.org, caoversight.org, watchwatch.org, computecompute.org, agegating.org) | Independence and capability scores computed from published rules and public records (sheriffoversight.org); text extracted from scanned reports by OCR (caoversight.org); deployment and facility summaries (watchwatch.org, computecompute.org); and, on agegating.org and computecompute.org, the same European Union and European Economic Area law summaries as law.ungovr.org. | The scores are computed, not drafted by AI. OCR text is published as extracted and is not corrected by hand. The European law summaries carry the same research-corpus citation check and recorded editorial review as law.ungovr.org. | This page, and the note beside a European law summary on agegating.org and computecompute.org. |
| civilgrandjury.org | Report text extracted by OCR; findings and topics classified by a locally run model; a machine translation of a page into another language. | OCR text is published as extracted and is not corrected by hand. | An OCR footnote on report pages, a classifier note on the reports index, and a banner on a translated page. |
| UnGovr Words (meeting pages) | Meeting transcripts and speaker attribution from recordings, and a summary of what was said. | Each transcript and attribution carries its own status, Pending, AI, or Verified, so a reader can see whether a person has confirmed it yet. | The status shown beside each transcript and attribution, and a notice above every AI-generated summary. |
| news.ungovr.org | Which stories run, and which places and agencies each one is tied to. | None beyond the model itself. Headlines and article text are always the publisher's own words; UnGovr never rewrites them. | The "how this page is put together" note on the news home page. |
| UnGovr Request | On UnGovr Request, a short description of a resident's reported issue, drafted from the details they gave, when a resident chooses that option instead of typing their own words. | None beyond the drafting itself. | The map popup and the detail page carry a "Description drafted with AI" label under an AI-drafted description. |
The line on EU pages
The note above a law summary, and the byline that replaces it.
On a law summary written for a reader in the European Union or the European Economic Area, a short line sits directly above the generated text: it says the text was drafted with AI from the cited sources, under UnGovr's direction, and that UnGovr holds editorial responsibility for the page. Once a named UnGovr staff member has checked that exact text against its source, the note is replaced by a line naming the reviewer and the date, and it reappears only if the text is rewritten afterward.
Images
An AI-generated illustration says so in its own file.
Illustrations generated with AI, such as the ones on LexLint Insights, carry that claim in the image file's own metadata (the IPTC digital source type), where a machine can read it. None of them depicts a real person or a real event as though it were genuine, so none needs a caption on the page to say so.