Public Records and Governance Surfaces: The Most Authoritative Data You Never Check

Companies House, the FCA, the SRA and the CQC describe your business with more authority than your own website, and almost nobody checks what they say. They are public, structured, independent and treated as reliable. That makes them the strongest corroborating sources available to a UK business, and the most damaging when they are wrong.

This page is part of the ecosystem validation cluster.

Why registers carry disproportionate weight

Three properties, and few sources have all three.

They are independent. You cannot write your own entry in the way you write your About page. A model treats an assertion you cannot control differently from one you can.

They are structured. Fields, not prose. Company number, incorporation date, registered office, SIC code, officers, accounts filing history. Machine-readable without interpretation.

They are authoritative by definition. Companies House is not a third party’s opinion about whether your company exists. It is the record that determines it.

The result is that a discrepancy between a register and your website is not treated as two views. It is treated as your website being wrong. That asymmetry is the reason this page exists.

There is a second effect that catches people out. Company register data is mirrored widely, across dozens of company data aggregators and credit reference sites. An error at source does not sit in one place. It propagates, and it keeps coming back after you have corrected it everywhere else, because those sites re-sync from the register.

The UK register stack

Two things to note about the voluntary ones. They are less authoritative individually but they are often the most descriptive, because a trade body listing usually says what you actually do in a sentence rather than in a SIC code. And they lapse silently, which turns an asset into a contradiction the day the membership expires.

The five failures we find most often

1. The registered office nobody updated

The single most common finding. The business moved, the website and Google profile were updated within a week, and the registered office was left. Now two authoritative sources disagree about where the company is, and the more authoritative one is wrong.

Fixing it is an online filing that takes about ten minutes and costs nothing.

2. A SIC code from incorporation

Chosen by an accountant on day one, frequently the catch-all 82990 “other business support activities not elsewhere classified”, and never revisited. The register’s own statement of what your company does now says approximately nothing, while your website claims a specialism.

You can change SIC codes on the next confirmation statement or file early. Also free.

3. Directors who left

Officers who resigned years ago still showing as current. This matters more than it looks, because leadership is one of the anchors a model uses to resolve an entity, and it is one of the facts most often asked about a company. It also interacts badly with LinkedIn, where the same person has probably updated their own profile, producing a direct contradiction between two strong sources.

4. Dormant and dissolved companies with similar names

Very common in businesses that restructured, and genuinely difficult.

A dissolved “Smith Consulting Ltd” and a live “Smith Consulting Group Ltd” give a model two candidate entities with similar names and overlapping directors. Confused, blended or plainly wrong AI answers about a company very often trace back to this.

You cannot delete a dissolved record. What you can do is make the live entity unambiguous everywhere else: consistent full legal name including the suffix, the company number published on your site, and an entity map that states which presences belong to which company. This is one of the few ecosystem problems with no clean fix, only a mitigation, and it is worth saying so plainly.

5. A lapsed accreditation still displayed

The certification expired, the badge is still in the footer. This is now a claim contradicted by the issuing body’s own register, which is materially worse than never having had it, because it is checkable and it is wrong.

Regulated sectors: the highest return work available

Models are measurably more conservative on health, legal and financial topics, because a wrong answer causes real harm. Fewer sources get cited, and the ones that do make verification easy.

A registration number that resolves to a live entry on an independent regulator’s site is about as verifiable as a claim gets. Publishing it, and making sure the register entry matches your site exactly, does more for a regulated firm’s citation prospects than most content work.

The practical checklist for a regulated business:

• The registration number published visibly on the site, not buried in the footer terms
• The registered name on the site matching the register entry character for character
• Trading names declared on the register where the regulator requires it, which the FCA does and which firms frequently miss
• The registered address matching
• Current status, permissions or scope described accurately rather than aspirationally
• Any rating or judgement stated with its date

The sector detail sits in GEO for regulated industries, SRA transparency rules and AI visibility, FCA Consumer Duty as a content opportunity and YMYL and AI search.

An annual check, and the four triggers

The check itself is short. Pull up every register you appear on, read the entry as if you were a stranger, and note anything that differs from your website.

The triggers matter more than the calendar, because they are what creates the problems:

1. A move. Registered office, regulator address, ICO address, trade body listing
2. A rename or rebrand. Every register, plus the question of whether the old name should remain visible as a former name, which it usually should
3. A director or manager change. Companies House, and the regulator where a registered individual is named
4. A change in what the business actually does. SIC codes and regulator permissions

Add these four to whatever process already handles the practical side of each event. The failure is never that the task is difficult. It is that it belongs to nobody.

For the wider version covering directories, reviews, social and media, use how to audit your off-site AI ecosystem.

Public Records and Governance Surfaces: Common Questions

Do AI models use Companies House data?

Yes, directly and indirectly. The register is public, structured, free to query and treated as authoritative, which makes it close to an ideal source for facts about a UK company. It is also mirrored across dozens of company data sites, so an error there propagates into many places at once and keeps returning after you correct it elsewhere.

Which public records matter for AI visibility?

Companies House for every UK company, then your sector regulator: the FCA register, the SRA, the CQC, Ofsted, the Gambling Commission or the equivalent. Then the ICO register if you handle personal data, trade association registers, and any accreditation body. Together they form the most authoritative description of your business in existence.

How often are public records wrong?

Far more often than businesses expect. The common failures are a registered office that was never updated after a move, a SIC code chosen in year one that no longer describes the business, dissolved dormant companies with similar names still on the register, and directors who left years ago. None of these break anything, so nobody notices.

Does my SIC code affect AI search?

It contributes. The SIC code is the register’s own statement of what your company does, so if it says other business support activities and your website says specialist financial technology consultancy, you have given a model two different classifications from two authoritative sources. Updating it is free and takes minutes.

Can I get a public record corrected?

Usually yes, through the body’s own process, and it is more straightforward than people assume. Companies House address and SIC changes are a same-day online filing. Regulator records generally need a form and a short wait. The delay is rarely the process; it is that nobody in the business owns the task.

Do dissolved companies affect my entity?

They can, and it is a common cause of confused AI answers. A dormant or dissolved company with a similar name sitting on the register gives a model two candidate entities and no obvious way to choose between them. You cannot remove a dissolved record, so the fix is to strengthen every other signal that points at the live company.

Do regulator registers help with AI trust signals?

Substantially, particularly in regulated sectors where models are more conservative. A registration number that can be checked against an independent authority is verifiable, which is exactly what a system trying to avoid a wrong answer wants. Publishing the number on your site and making sure the register entry matches is one of the highest-return things a regulated firm can do.

How often should I check public records?

Annually as a standing task, and immediately after any move, rename, director change or change in what the business actually does. Those four events cause almost every discrepancy, and the gap between the change and somebody remembering the register is where the problem is created.

Where to Go Next

The rest of the cluster: ecosystem validation hub · review platforms and trust signals · social platform consistency · media distribution and entity reinforcement

If your entity is being confused with another: AI giving wrong information about my company · entity signals checklist · what is entity authority

Have it delivered: MarGen is a UK GEO agency with published pricing, a published method and documented results.