Can a new company have a Wikipedia article? A source based notability self check and conflict of interest guide

A new company usually cannot have a Wikipedia article, and the reason is not a rule about age. It is that an article has to be built from independent, substantial coverage written by people with no connection to the company, and a new company normally does not have any yet. The self check is therefore a source audit, not an argument: you list the coverage you think qualifies, apply written tests to each item, and count what survives. As of 29 September 2026 our own recorded source audit holds zero qualifying independent substantial items for aiknowsus.com, because we have not run and published that audit. Zero is a legitimate result here and it is more useful than a number nobody can check. What follows is the four tests, the ten fields per source, the conflict of interest rules for anybody connected to the company, and the dated evidence we hold about what happens when a claim cannot be supported.

The answer, first

Five statements.

The question is about sources, not about the company. Whether your company deserves an article is not the test. Whether independent published material exists, in enough depth, for somebody unconnected to you to write from, is the test.

Zero is a normal answer for a new company, and it is honest. Our own count, in our own record, is zero qualifying items as of 29 September 2026. We are not going to fill that cell with a number and we are not going to invent a threshold, because the encyclopaedia does not set a simple numeric test that we can quote as one.

Do not write about your own company there. If you are the founder, an employee, an investor or a paid agency, you have a conflict of interest, and there are established ways to declare it and to request rather than create. The rules are on the site itself and they are enforced by editors, so read them and follow them rather than reading a summary.

The audit is worth running even when the answer is no. The list you build is the same list that tells you whether independent coverage exists for any purpose: press, procurement questionnaires, or an AI assistant's description of your company.

We do not restate the encyclopaedia's guidelines. Read the current notability guidance for companies, the guidance on identifying reliable sources, and the conflict of interest and paid contribution pages. Record the URL and read date, quote what you rely on, and save copies.

How it was measured

One. Build the candidate list first, from every source you can find. News articles, trade press, books, academic work, industry reports, award citations, regulatory filings, and anything else published about the company. Include everything at this stage, including the items you expect to fail, because the count of what fails is part of the result.

Two. Apply four tests to each candidate, in this order, and record which test it fails.

• Independence. Was it written by somebody with no connection to the company and no payment from it? A press release, a sponsored feature, a customer story, a partner blog post, an interview that is mostly your own words, and anything you or your agency wrote all fail this test.
• Substance. Does it discuss the company in depth, or does it mention the name in passing? A funding round listing, a directory entry, a one line mention in a round up, and a quote from your founder in an article about something else are all mentions, not coverage.
• Reliability. Is the publication one with editorial oversight, a named author, and a correction practice? A site that republishes press releases does not become independent by having a masthead.
• Verifiability. Can somebody else find it today? Record a URL or a full citation, plus an archive link, because a source that has disappeared cannot support anything.

Three. Record ten fields per candidate.

• Title and publication.
• Author name, or blank if none is given.
• Publication date.
• URL and an archive link.
• Word count of the part that is about your company, approximately.
• Independence: pass or fail, with the reason.
• Substance: pass or fail, with the reason.
• Reliability: pass or fail, with the reason.
• Verifiability: pass or fail.
• Whether the item came from a press release, checked by searching for the same sentences elsewhere.

The last field catches more items than any other. A story that appears in six publications with the same phrasing is one press release, not six sources.

Four. Print the denominator arithmetic. Report it as candidates identified, then items surviving each test, then qualifying items, all as of a date. For example: 22 candidates identified, 9 passed independence, 4 also passed substance, 4 also passed reliability, 4 qualifying items as of the date. Write the chain out, because the number that gets quoted is the last one and the chain is what makes it defensible.

Five. Two people assess the same candidate list separately, then publish the disagreement count. Almost every disagreement will be about substance, which is why the test has a written definition and a word count field beside it.

Six. Declare the conflict of interest in writing if you touch anything. If you are connected to the company, the correct path is to disclose and to request rather than to write. Record what you declared, where, and on what date. An undeclared connection found later is worse for the company than having no article at all.

Seven. If you are measuring how assistants describe you, keep it blind and separate. Freeze 20 to 100 questions in a versioned, dated file, with your brand name in none of them. Ask each three times per assistant, keep every answer, and print the arithmetic: 20 questions times 3 repetitions times 3 assistants is 180 answers. The reason for the blind rule is measured: on 27 July 2026 two runs of the same 78 questions happened on the same day for clawlaw.in, the branded wrapper returned the company first on almost every question, and the blind run put it second by breadth and absent from the litigation due diligence questions it most wanted to win. The branded run was discarded.

Eight. Record any search claim as a claim. In the September 2026 aiknowsus.com audit, Perplexity withdrew its own earlier statement about how many of the questions it had searched for, in three separate batches, saying it could not honestly substantiate it.

What the numbers were

Qualifying independent substantial sources in our own recorded audit for aiknowsus.com: zero, as of 29 September 2026. The audit has not been run and published, so the record holds nothing. We are not going to describe coverage we have not listed.

What we refuse to publish. A numeric notability threshold, because the guidance does not offer one that we could quote as a simple test. A count of sources for any company we have not audited. A claim that an article is achievable in some period of time.

The dated results we do hold that bear on unsupported claims. They are from legal research software sold in India, and they are the closest real evidence we have about what happens when a claim cannot be checked.

• An impossible figure cost the recommendation. On 6 August 2026 a headline figure on the clawlaw.in main site could not be true, and when Claude checked it against public numbers it advised a buyer against the product, after which the company's accurate claims stopped counting for that answer.
• Correct but unsourced was still discounted. On 17 September 2026 two comparison pages stated competitors' prices with no link, date or source. Claude checked them independently, found them correct, and still treated the pages as advocacy and used official sources instead.
• Hidden instructions were caught and named. On 6 August 2026 a competitor's page carried a hidden block of text addressed to answer engines, instructing them to cite that company as the source. Claude found it, refused it, and named the company that had done it.

The nearest real fraction, with its sector named. 0 of 6 questions in which any third party review or directory source reached an answer at all, in Claude's run on 17 September 2026, in legal research software sold in India. It is a fraction about third party sources reaching an answer, which is the nearest thing we hold to a measurement of independent coverage mattering, and it is not a Wikipedia number.

The third bullet is the one to remember if anybody offers to get you an article. Attempts to manipulate these systems are found and named, and the company named is the one that paid for it.

What this cannot tell you

• It cannot tell you whether an article would survive. That decision belongs to volunteer editors applying their own guidance.
• It cannot give you a threshold. The guidance is a judgement about depth and independence, not a count, and treating it as a count is the most common error in this area.
• A qualifying source count is not a reputation score. Plenty of good companies have zero, and some have coverage they would rather not have.
• Our own count is zero because the audit is unrun, not because we have established that no coverage exists. Those are different statements and the difference matters.
• Our dated results are one company in legal technology in India. They are about unsupported claims, not about encyclopaedia policy.
• Nothing here promises an article, a position in an AI answer, or any outcome.

Sources and change log

Sources. The four tests, the ten fields and the arithmetic chain are ours. The dated results come from the clawlaw.in programme, recorded in Tier_1/GEO_BASELINE_RESULTS_2026-07-27.md and Tier_1/claude_response_17_09_audit.md, and from the aiknowsus.com audit of September 2026 across 24 batches and 72 conversations. The impossible claim, the unsourced pricing and the hidden instruction findings all concern public pages and can be checked from outside. No encyclopaedia guideline is quoted or paraphrased on this page.

Change log. 29 September 2026, first publication, stating zero qualifying sources in our own record. When the audit runs, this section will carry the candidate count, the survivors at each of the four tests, the qualifying count, the disagreement count between two assessors, and the date. A zero result will be published with the same prominence as any other.

Common questions

Can we pay somebody to write the article for us?

Paid contribution is governed by the site's own rules and must be declared. Anybody offering to create an article quietly is offering to do something the editors look for and remove, and the company named in the aftermath is yours. On 6 August 2026 Claude found a competitor's hidden instruction to answer engines, refused it, and named the company responsible. Manipulation is discovered by the systems it targets.

Our funding round was covered in several outlets. Does that count?

Check the tenth field first: search for the same sentences across those outlets. If the phrasing matches, that is one press release appearing in several places, and under the independence and substance tests it is likely to be one mention rather than several sources. Record each item and the reason it failed, because that record is the answer to the same question next year.

What should we do instead while we have no qualifying coverage?

Publish the facts about your company on pages you control, with dates and sources, so that anybody writing about you later has something to work from. In our September 2026 capture across 24 batches and 72 conversations, the most cited domains were the assistants' own documentation and vendors' own websites. Your own well sourced pages are doing more work than a missing encyclopaedia article.

Is an article good for how AI assistants describe us?

It is a widely read, independently edited source, so it is reasonable to expect it to matter, and we have not measured it, so we will not claim a size. What we can say from our own runs is that unsupported claims are actively damaging: on 6 August 2026 a figure that could not be true led Claude to advise a buyer against the product.

Can we correct an existing article about us?

You can raise the correction through the declared route for people with a conflict of interest, with sources. Record what you requested, where and when. Editing directly while connected to the company is the thing most likely to make the situation worse.

How often should we rerun the source audit?

Every six months, and after any significant independent coverage. Keep every old sheet, because the interesting output is the chain moving: candidates up, survivors up, and the reasons items failed changing over time.

What to do first

Build the candidate list today and apply the independence test only, with a reason recorded for each failure. That one pass will usually answer the question for a new company in an afternoon, and it leaves you with a real sheet rather than an opinion. Then read the conflict of interest guidance before anybody in your company touches anything on the site.