How This Site Is Written, and What the Machine Is Never Allowed to Do
Short answer. Anthony Braswell writes and is accountable for every guide on this site, under Quorum Industries LLC. Research, drafting, formatting and checking are done with AI assistance. A person chooses the question, goes to the primary source, reads it, decides what it does and does not say, and signs it off. The machine is never permitted to invent a citation, decide what a source says, upgrade an evidence tier, or add a claim to a draft. Those four prohibitions are enforced by gates that stop publication, and this page names them so you can check.
Most sites will not tell you this. That is the reason to publish it.
Who writes this
Anthony Braswell, of Quorum Industries LLC, Ocoee, Florida. Not a newsroom, not a staff, not a "data desk". One person, working with tools, saying so.
Where a guide carries somebody else's words, it carries their name, never ours. That is a term of the contributor licence, not a courtesy: a contributor's article publishes under the contributor's byline with their verified credential, and we may cut but never add. If you are reading a byline on this site, that person is answerable for the words underneath it.
How automation is used
The tools are Anthropic's Claude models, together with the automated search, parsing and checking code in this repository. We name the family rather than a version number, because versions change faster than this page can honestly track and a stale version string would be the one false sentence on a page about honesty. The review date at the foot of this page tells you how current the statement is.
Specifically, rather than in the vague phrasing that has become normal:
- Finding candidates. Statutes, administrative rules, bulletins and agency pages are located and queued for reading by automated search.
- Drafting. Once a person has read a source and decided what it establishes, prose is drafted with AI assistance and then edited.
- Formatting and structure. Tables, anchors, schema markup and cross-links are generated.
- Checking. Hundreds of automated controls run before anything publishes. They compare what a page claims against the register it claims it from, and refuse the deploy when the two disagree. The statute-guide gate alone fires 903 of them across 209 cited provisions.
- Maintaining. Registers are re-checked on a schedule, and staleness is surfaced rather than hidden. A dataset left unread past its review window fails a gate.
Why automation is used
Because the alternative is not a more careful site. It is a much smaller one.
This site holds fifty-one jurisdictions' repair and claims law, read from the jurisdiction's own statute, rule or agency document. One person cannot read fifty-one bodies of law, re-check them as they change, and confirm that several hundred published sentences still match their registers, by hand, every night. Automation is what makes the checking affordable, and the checking is the part that makes the reading worth publishing.
The honest trade: automation buys coverage and repeatability. It buys nothing whatever in the way of judgement, and this site is built so that it is never asked to supply any.
What automation is never permitted to do here
These are not aspirations. Each is enforced by a gate that fails the build.
1. It may not invent or approximate a citation. Every provision cited in a statute guide is quoted back and checked against the register it came from. That gate currently fires 903 controls across 15 guides and 209 provisions. A citation that does not match its register stops the deploy.
2. It may not decide that a source says something. This site separates what it has read from what it has merely cited, and labels the difference on the page. Where a government host refused us, the refusal is published as a refusal and never quietly worked around. Two rows on the AI-bulletin register record that a person, not the pipeline, obtained the document, with the date and the reason: New Hampshire's insurance department returns 403 to an automated reader on every path tried, so the bulletin was read by hand and the register says so in those words. The Virginia letter is the same story. Both are transcribed here in full because the agencies publish them only as files a machine cannot read.
3. It may not upgrade an evidence tier. Rule, record and assessment are three different things, and our evidence standard fixes what each is allowed to claim. A rule is what a jurisdiction enacted. A record is a document that exists. An assessment is our reading, labelled as ours. Nothing moves up a tier without a person moving it, and a separate gate fires 28 controls keeping the four claims we have promised never to make out of the corpus.
4. It may not add to a draft. The editing boundary here is cut, never add. We shorten, we clarify, we decline to publish. We do not make a claim stronger than the evidence supports, and on contributed work we never make an accusation stronger than the contributor made it.
Where AI is not used at all
The list above says what automation does. This one says where it does not go, because a disclosure that only describes use is half a disclosure.
No AI system is an author here. It is not credited, it is not a co-author, and it is never cited as a source. A byline on this site names a person who is answerable, or it names nothing at all. That is why the 188 county pages generated from the register carry no byline: nobody wrote them, so nobody signs them.
No image on this site is AI-generated. Every image is a Quorum brand mark, an icon, or a social card. There is no image-generation tool anywhere in the build.
No confidential contributor material goes near an AI system without that contributor's written permission. Technical materials sent to us under the confidential contributor agreement are governed by section 6A of that agreement, which asks the contributor directly whether we may process their files with AI tools. A signature is refused if the question is left unanswered. Say no and we read the material by hand instead. Whatever is answered, four things can never happen and cannot be changed by giving notice: the material is never included in a dataset we sell, never sold or licensed on its own, never used to suggest an endorsement, and never used to train anybody's AI model.
No AI system decides what a source says. That is the second prohibition above, and it is the one this whole site rests on.
How you can check any of this
Not by trusting the paragraphs above.
- Every guide names its sources and, where the underlying register records one, the date the source was read.
- Corrections are logged publicly rather than edited away. Where a page has been corrected, the correction is on the page.
- Two primary documents on this site exist nowhere else in machine-readable form, because the issuing agency published them only as scanned or unindexed files and we transcribed them: New Hampshire Bulletin INS 24-011-AB and Virginia Administrative Letter 2024-01. Compare our text against the agency's own file.
- Where a number appears, it comes from a register you can fetch. Several publish as JSON under an open licence.
If you find something here that does not survive that check, it is a defect and we want it: contact@theautobodydirectory.com. A person reads them.
Sources
This page exists because a search engine asked the questions and we thought a reader deserved the answers, so the questions themselves are cited rather than paraphrased.
- Google Search Central, "Creating helpful, reliable, people-first content",
developers.google.com/search/docs/fundamentals/creating-helpful-content, read 2026-08-21. The self-assessment questions quoted in the summary above appear there verbatim: "Is it self-evident to your visitors who authored your content?", "Do pages carry a byline, where one might be expected?", "Do bylines lead to further information about the author or authors involved?", "Is the use of automation, including AI-generation, self-evident to visitors?", "Are you providing background about how automation or AI-generation was used to create content?" and "Are you explaining why automation or AI was seen as useful to produce content?" - Our contributor licence, /contributor-license, for the byline and cut-never-add terms described above.
- Our evidence standard, /guides/evidence-standard-rule-record-assessment, for the rule, record and assessment tiers.
- The AI-bulletin register, which records for New Hampshire and Virginia that a person, not the pipeline, obtained the document, with the date and the reason. The transcripts are published in full at /guides/nh-insurance-bulletin-ins-24-011-ab-full-text and /guides/virginia-administrative-letter-2024-01-full-text.
- Our confidential contributor agreement, section 6A, for the AI-processing election and the four permanent prohibitions described above.
- Checked for completeness against two published frameworks: the CDC's Considerations for Disclosing Generative AI Use in Scientific Work, whose disclosure elements are content affected, action taken, AI tool, purpose and human oversight; and The Open Notebook's Creating a Public AI Policy for Your Newsroom, August 2026. Both read 2026-08-21. One item from the second is deliberately absent: commentary on the energy and employment consequences of AI generally. It is a real subject and not this site's, and every other sentence on this page is checkable against something.
When this page was last reviewed
Last reviewed 2026-08-21. This is a live document, not a launch statement. When the practice changes, this page changes and the date above moves with it. If the date is old and the practice has moved, that is a defect on the same terms as any other on this site, and the address above is where to report it.
What this page is not
It is not a claim that automation makes this site more accurate than a lawyer, a regulator, or a shop owner who has been doing the work for thirty years. It makes no such claim. It is a claim about provenance: that you can see where every sentence came from, who is answerable for it, and what was and was not read by a person.
That is a smaller claim than most sites make, and a far easier one to check.
General consumer information. Not legal, insurance, or financial advice.
General consumer information: not legal, insurance, or financial advice. Requirements, coverage, and practices vary by state, policy, and manufacturer.