It is easy to type a real company name into the company field of your test data. Nothing breaks, so it does not feel like a problem. The meaning changes the moment that screen leaves the building as a screenshot or a manual figure. This article covers why real trade names do not belong in test data, and how to build fictional ones instead.
1. What goes wrong with a real trade name
A company name on its own is harmless. The problem is the values sitting next to it. Admin-screen test data usually carries states like these:
- Contract status (cancelled, suspended)
- Credit or screening result (rejected, needs review)
- Billing status (unpaid, in collections)
When a screen carrying a real company name in that context ships as a manual figure or a sales demo, you have published untrue information attached to that business. Your intent does not travel with the picture.
There is a practical reason too. Testing with well-known company names weakens your search and layout coverage. Famous names tend to be short, free of symbols and similar in length, so you never see what happens to a long trade name or one containing an ampersand or a parenthesised abbreviation.
2. There is no reserved company name
Email addresses and IP addresses have ranges reserved for testing. example.com (RFC 2606) and 192.0.2.0/24 (RFC 5737) are guaranteed to belong to nobody, so a mistaken send does no harm.
Company names have no such mechanism. Anyone can register a company called Sample Inc. or Test Trading. So "it sounds made up" is not a test you can rely on, and you need your own construction rule and your own check.
3. Build from three parts
The practical approach is to split the name into three pieces.
| Part | Role | Examples |
|---|---|---|
| Stem | The distinctive part; sets readability | Mirai / Aoba / Next / Komorebi |
| Industry word | Hints at what the company does | Shoji / Systems / Seika / Kensetsu |
| Legal entity type | Prefix, suffix, LLC, or none | 株式会社◯◯ / ◯◯株式会社 / 合同会社◯◯ |
The three-way split buys you one thing in particular: you can mix where the entity type sits. Japanese trade names put 株式会社 either before the name (maekabu) or after it (atokabu), which changes both length and word order. Test data that uses only one form hides wrapping and truncation bugs until production.
The implementation is just one random pick per part.
const STEMS = ["みらい", "あおば", "ネクスト", "こもれび"];
const TAILS = ["商事", "システムズ", "製菓", "建設"];
const KAKU = [
(n) => `株式会社${n}`,
(n) => `${n}株式会社`,
(n) => `合同会社${n}`,
(n) => n,
];
const pick = (a) => a[Math.floor(Math.random() * a.length)];
const company = () => pick(KAKU)(pick(STEMS) + pick(TAILS));
console.log(company()); // e.g. 合同会社あおば建設Run it for as many rows as you need and you get data with no industry bias and no entity-type bias. If you need romanized names, give each stem and industry word a romaji form and join them in the same order.
4. Check before you ship
A randomly assembled name can still coincide with a real trade name. For internal test data this rarely matters, but it is worth checking when the data:
- appears in manual or help-page figures
- appears in sales material or a trade-show demo
- ships as sample data in a public repository
Searching the trade name in Japan's National Tax Agency corporate number registry is the quickest check. If it matches, swap the stem or the industry word. You do not need to check every row, only the handful that actually appear on screen.
5. Summary
Company names are one of the few fields with no safe reserved value. The workable answer is to understand why real trade names are a problem, build names from a stem, an industry word and an entity type, and check the few that go public.
Our Dummy Company Name Generator implements exactly this. Pick an industry and entity type, generate in bulk and export as CSV. For people and addresses see the Japanese Name Generator and the Japanese Address Generator; for the reserved ranges that do exist, see Do Not Use Real Values in Test Data.
Frequently asked questions
Why should I avoid real company names in dummy data?
Screenshots, operating manuals and demo environments leave the building. When a real company name sits next to a status like cancelled, credit check failed or payment overdue, you have published untrue information attached to that business. The value was only there for convenience, but nobody looking at the picture can tell.
How do I build a fictional company name?
Split the name into three parts: a stem such as Mirai or Next, an industry word such as Shoji or Systems, and a legal entity type such as Co., Ltd. or LLC. Pick each part at random and join them. Keeping the parts separate lets you fix the industry, or deliberately mix the entity types, later on.
What if a generated name turns out to be a real company?
Before using a name in published material, search the trade name in Japan's National Tax Agency corporate number registry. If it matches, swap the stem or the industry word. For internal test data this rarely matters, but it is worth checking the handful of names that actually appear on screen.