TL;DR (60 seconds):
There is a playbook going round that says registering your business on government directories gives you a profile page on a .gov domain that AI search reads as a trust signal. We tested it on six real registered companies across three countries and logged all 131 citations. SAM.gov was cited zero times. UK Companies House was the most cited source in the entire test. The difference is not authority, it is whether the register serves a readable page or a JavaScript app.
There is a playbook going round. Register your business on SAM.gov, tick the small business box, sign up to your state vendor portal, and you get a profile page on a .gov domain carrying your company name and your website. A free trust signal that AI search reads.
It is a good story. Government registers really are free, really are authoritative, and really do list ordinary businesses. The forms are the only barrier, and forms are exactly the sort of work that is now cheap to get through.
So we tested it, because the claim is checkable and nobody seems to have checked it.
The playbook does not work. The way it fails is more useful than the trick it promises.
What we measured
Six real companies, every one verified as currently registered before we asked anything.
Two US firms confirmed active in SAM.gov, each with a website recorded in the register. Two UK limited companies on the Companies House register. Two Australian companies with active ABNs on ABN Lookup. Small and mid-sized operating businesses, not household names, because a household name gets answered from general knowledge and tells you nothing about what the machine went and looked up.
Each company was put to four AI search lanes with the same question: is this a real, legally registered company, and what sources did you use. Four lanes, six companies, 24 answers, 131 citations. Every citation was logged and classified by domain.
What AI actually cited
| Register | Citations | Share of answers citing it |
|---|---|---|
| UK Companies House | 16 | every UK answer, 8 of 8 |
| Australian ABN Lookup | 8 | 4 of 8, plus ASIC and the Modern Slavery Register |
| US state corporate registers | 3 | 2 of 8 |
| SAM.gov | 0 | none |
Zero. Not one citation in 131, for two companies that are registered in SAM.gov right now with their website sitting in the record.
Companies House was not merely present. It was the single most cited domain in the whole test, ahead of Wikipedia and ahead of the companies' own websites.
Why SAM.gov never appears
Request a SAM.gov entity page the way a search crawler does and you get about 47KB of JavaScript and a page title that reads, in full, "SAM.gov". No company name. No address. No website. No content of any kind.
The record is real. The page is assembled inside your browser after it loads, and a crawler does not run that. Nothing that reads pages has ever seen your SAM profile.
The SBA small business search has the same shape. The old Dynamic Small Business Search now redirects to a single-page application whose entire crawlable content is one generic sentence describing the site.
Now compare Companies House. Ask for a company page as a crawler and the company name is right there in the markup, on a gov.uk domain, fully readable, along with the number, the status, the registered address and the filing history. ABN Lookup does the same thing.
That is the entire difference. Not domain authority, not government trust, not any of the things the playbook talks about. One set of registers publishes pages. The other publishes an app.
This is a general lesson worth carrying beyond registers. When a system looks like it is working but produces nothing downstream, the fault is usually mechanical and boring rather than strategic. The same pattern shows up in most automation projects that stall: the plan was fine, one link in the chain simply never carried the data.
These are mentions, not backlinks
Here is the part that breaks the framing completely.
Companies House does not collect your website. Neither does ABN Lookup. The pages that AI actually cites carry your legal name, your company number, your status and your registered address, and no link to you at all.
So the asset is not a backlink. It is corroboration.
An independent, authoritative record confirming that this company exists, under this name, at this address, and is currently active. When a model is asked whether you are real, that is the record it reaches for. It does not need to link to you in order to answer.
That distinction matters commercially, because it changes what you optimise. Chasing directory links is a volume game that rewards whoever fills in the most forms. Being verifiable is a consistency game that rewards getting one set of facts right in the handful of places that count.
Consistency is also where most businesses quietly fail, and for the same reason their internal reporting fails. The name on the register does not match the name on the invoice, which does not match the name on the website. That is the messy data problem showing up in public rather than in a spreadsheet.
SAM data does get cited, just never from SAM
Look closely at the US answers and SAM data is in there. It arrives like this:
opengovus.com/sam-entity/C43DLCSKXGM8besagov.com/companies/C43DLCSKXGM8contractology.com/contractors/global-connections-to-employment-inchighergov.com/awardee/global-connections-to-employment-inc-10000465
Those are SAM entity records, keyed to the company's own SAM unique entity ID, published on commercial mirrors.
SAM releases a free monthly bulk file covering all 895,000 registered entities, and a small industry rebuilds that file into pages a crawler can actually read. Four separate mirrors turned up in our citations.
So registration does eventually reach AI answers. The route is register, wait for the monthly file, wait for a mirror to rebuild it, hope that mirror is the one a model reaches. You control the first step and nothing after it, and the .gov page was never in the chain at any point.
This is worth knowing before you rely on it. Anything that depends on a third party republishing your data is a dependency you do not own, with the same failure profile as any vendor you have no contract with.
The US register that works is the one the playbook skips
Two US answers cited a state corporate register: Florida's Sunbiz entity search and Maryland's Department of Assessments and Taxation.
Those are Secretary of State style business registers, and they are the direct US equivalent of Companies House. They serve readable pages, they carry your legal name and status, and they get cited.
Note what they are not. They are not the state vendor or procurement portal. Procurement portals exist so that government buyers can find suppliers. Corporate registers exist so anyone can confirm a company exists. For getting cited you want the second one, and in most states you are already in it automatically from the day you incorporated.
The playbook sends you to the procurement portal. The corroboration is in the register you never had to sign up for.
What to do instead
1. Check your corporate register entry is correct and current. Companies House in the UK, ABN Lookup and ASIC in Australia, your Secretary of State entity search in the US. This is the record that gets cited. If your registered name, address or status is wrong, lapsed or filed under an old trading name, fix that before anything else. It is free and it takes minutes.
2. Make the register and your website agree exactly. Same legal name, same suffix, same address format. A register corroborates you only if a machine can tell the two records describe one company. This is the single highest-value step and almost nobody does it.
3. Put your entity details on your own site in plain text. Legal name, company or ABN number, registered address, in the footer or on an about page. More than half the answers in our test cited the company's own website alongside the register. Give both sides of the match something to match against.
4. Register on SAM.gov only if you want federal contracts. It is free, it is legitimate, and anyone charging you to do it is a middleman. Just do not do it for search visibility, because there is no readable page and there is no link.
5. Read your own citation profile. Ask three AI search products whether your company is real and read the sources each one cites back. That list is what the market currently believes about you, and it takes about five minutes to pull.
The honest limits of this test
Six companies, one run, on 22 August 2026. The lanes were Perplexity's own search, Exa's answer endpoint, and two models running over a third-party web search layer, so this measures retrieval layers rather than any single consumer product's exact behaviour. Registers change and search products change.
Rerun it on your own company name before you rely on it. That is a five minute job and the result is specific to you.
What will not have changed is the mechanism. A register that serves a readable page gets cited. A register that serves an application does not. Everything else is downstream of that.
Next Steps
The question worth answering is not how many directories you are listed in. It is whether the public record agrees with itself about who you are.
Start by pulling your entry from your national corporate register and putting it side by side with your website footer, your invoices and your Google Business Profile. Compare the legal name character by character, including the suffix. Compare the address line by line. If any two of those four disagree, that is the work, and it costs nothing but attention.
Then ask three AI search products whether your company is real, and read the sources they cite. If they cite your register entry, your own site and nothing embarrassing, you are in good shape and you can stop. If they cite a stale directory listing, a dissolved entity with a similar name, or nothing at all, you now know exactly what to fix and in what order.
Most businesses we look at have between two and five public records that contradict each other, none of which anyone has read in years. Fixing that is not a project. It is an afternoon.
We help established small and mid-sized businesses work out where AI and automation genuinely pay, and where the honest answer is that the fix is simpler and cheaper than the software. Our free 20-minute diagnosis will tell you which one you are looking at.
About AutoSpark
AutoSpark helps established small and mid-sized businesses find the one place AI or automation is genuinely worth applying, then builds and deploys it. The method is plain: interview the people doing the work, find where work repeatedly gets stuck, rank the problems by what they cost, and only build when the maths shows a clear payback.
AutoSpark is led by Patrick Nesbitt, a CA(SA), CFA and former private-equity investor, so AI is treated as an investment rather than a trend. Not an AI audit. Not a transformation programme. A short, evidence led diagnosis of where the money is leaking and what fixing it returns.
Start here: autospark.ai