Score every source before you write a word
Open a spreadsheet and give every article ever written about the company its own row. Score each one against the four criteria in Wikipedia's notability guideline for organizations and companies: significant coverage addressing the company directly and in depth, completely independent of the company, reliable, and secondary. A source has to pass all four on its own. Fail one and it contributes nothing, however good the outlet is, because the test doesn't care about the masthead.
Companies get this rule wrong. Sources are evaluated separately and independently of each other, so coverage doesn't add up, and three articles that each half qualify are worth nothing together. The guideline also wants more than one survivor, since a single significant independent source is almost never sufficient, and multiple is not a set number. It says outright that a tech start-up in a major US metro area is held to a higher standard than an organization from a place and a decade nobody wrote much about.
Then one filter on whatever survived. At least one source has to reach past your own city. The guideline requires at least one regional, statewide, provincial, national or international source. Attention solely from local media, its example being the weekly newspaper for a small town, is not an indication of notability. If nothing clears all of this, stop. Nothing later in this procedure fixes a source list.
Cross off the coverage that counts for zero
The guideline lists what it treats as trivial, and that's most of a modern press file. Routine announcements don't count: raised capital, expansions, acquisitions and mergers, a sale or a closure, hiring and promotions and departures, a product launch or discontinuation, a branch opening, quarterly results, appearances at trade fairs. A funding round written up at length in a national title is still a routine announcement. Length doesn't change it.
Only unpaid sources count, in the guideline's own bold type. Any paid or sponsored article or post is out, and so is any material substantially based on a press release, even when an independent outlet published it. The guideline names non-staff contributor pieces too, at publications that accept public contributions without meaningful editorial oversight, with Forbes, Huffington Post, Entrepreneur.com, Inc.com and TechCrunch as the examples. Your executives quoted as story sources are a passing mention, not coverage of the company.
Also out: best-of, top 100 and fastest-growing lists, non-notable awards, listings of your offices, employees, products or patents, and patents themselves whether pending or granted. Watch the near miss, too. A profile of your founder is coverage of your founder, not of the company, unless it devotes significant attention to the company itself.
Disclose before you make a single edit
Register an account in a person's name, not the company's. Wikipedia's conflict of interest guideline says not to create a shared organizational account or to use an organization's name as the account name. The account is yours, not your employer's. A shared login is one of the fastest ways to get blocked before you've written anything.
If anyone is being paid, disclosure is mandatory, and it comes from the Wikimedia Foundation's terms of use, not from etiquette. Put {{paid|employer=name of employer|client=name of client}} on your user page and {{connected contributor (paid)}} on the draft's talk page, as the paid-contribution disclosure policy sets out. Editors who fail to disclose paid contributions are prohibited from editing. There's no confidentiality for the employer, the client or the affiliation, and a non-disclosure agreement doesn't invalidate the requirement.
The definition is wider than a contract with an agency. Anyone compensated for any publicity effort connected to the subject counts as a paid editor whether or not they were paid specifically to edit Wikipedia, and interns count as employees. If you own the company, disclose anyway. The policy deems anyone compensated for any publicity effort connected to the subject a paid editor, it carves out no exception for owners, and over-disclosing costs you nothing. You do have a conflict of interest, and the Articles for creation process states it must be used by editors with one.
Write it as a draft, never straight into the encyclopedia
On the Articles for creation page the button reads Click here to start a new article, and it opens the Wikipedia Article Wizard. The first screen is headed Welcome and offers Practice in the community sandbox or Practice in your personal sandbox before Next. The final screen is headed You're all ready to go! and its button reads Create new article draft. Your title carries Draft: in front of it until it's accepted, and drafts aren't indexed by Google or most other search engines, so nothing surfaces on your name while you work.
Write it from the sources that survived the test and from nothing else. AfC's own instructions say articles should reflect only what reliable sources have said about the topic. Don't add categories to a draft, because a reviewer adds those on acceptance. Don't copy or closely paraphrase your sources. And articles that are generated entirely by LLMs will be rejected. Reviewers screen for it.
Do not create the page directly in the encyclopedia to save the wait. Conflicted and paid editors are strongly discouraged from editing affected articles directly and are told to put new articles through Articles for creation instead. Bypass the review by moving the page or pasting it into mainspace and it can get moved back to draft, speedily deleted or listed for deletion, while AfC also warns that repeated attempts may lead to consequences for the account.
Submit it, then watch the queue instead of your inbox
Click Submit for review on the draft. If that button isn't there, switch to the source editor and add {{subst:submit}} to the top of the page. Then confirm it landed. A yellow box reading Review waiting, please be patient means you're in the queue. No yellow box means nothing was submitted. The place to ask is the Articles for creation help desk.
There's no service level, and nobody writes to you while you wait. Wikipedia says only that getting a review can take a while but your draft will be reviewed eventually. The wizard's last screen says only that reviews can take a long time and your draft will be reviewed in due course. The pending banner on your own draft carries a live estimate in the form 'This may take [n] weeks or more, since drafts are reviewed in no specific order'. The figure moves with the backlog, so read it off your own banner and never off a number quoted anywhere else. The queue is public. The pending submissions category shows a live count and one bucket per day of waiting, so read that instead of anyone's promise about timing.
Put two dates in the calendar. A draft nobody has edited for six months gets deleted under speedy deletion criterion G13, so touch it before then, and know that a G13 deletion can be undone on request. Second, ignore anyone who contacts you offering to get your draft published. Wikipedia carries a standing scam warning on that final wizard screen: Real Wikipedia editors will never contact or solicit you for payment or compensation of any kind.
Watching doesn't stop once the draft is accepted. A company article attracts edits from competitors, former staff and people with a grievance, and the ones that do damage rarely look like vandalism. The edits that cause real problems are small, plausible and sourced. That's why they survive.
Read which word the reviewer used
Declined and rejected are different outcomes, and only one of them is over. A declined draft keeps its submit button, and Wikipedia encourages you to take the reviewer's feedback and resubmit. A rejected draft has no submit button, because reviewers reject only when it's unlikely the draft will ever meet the guidelines, for example when the topic isn't relevant to an encyclopedia or the draft is spam. Don't resubmit a rejection.
On a decline, fix the source list, not the prose. Rewrite the same article around the same citations and you get the same answer, because the decision was made source by source, so replace the references with ones that each pass all four tests alone, then resubmit with the button or {{subst:submit}}. Resubmitting without materially changing the sourcing burns volunteer time and eventually gets the account treated as a problem.
If the reasoning doesn't make sense to you, ask at the help desk or the Teahouse, where volunteers answer newcomer questions. Those two pages are the whole appeal. If an article gets created and is later deleted, the venue is deletion review. The argument there has to be that the process was misapplied, not that your company deserves a page.
If the sources aren't there yet
Wikidata is a separate project with a far lower bar, and it's the one worth knowing about. Its notability policy accepts an item that refers to a clearly identifiable entity which can be described using serious and publicly available references. That's the structured record search engines and AI assistants read when they work out who you are. It is not a Wikipedia article, and nobody should sell it to you as one.
A sourced mention inside an article that already exists is a much lower bar than a page of your own. An industry article, a parent company, an acquisition. You still don't edit it yourself. Put the proposed sentence and its source on that article's talk page using the {{edit COI}} template and let an uninvolved editor act on it, which is the same route for a correction to an article about you that's already live.
Then the part nobody selling this says out loud. An article isn't reputation control, and it can cut against you. The conflict of interest guideline puts it this way: if there's anything publicly available on a topic that you would not want included in an article, it will probably find its way there eventually. Maybe your company has a lawsuit, a layoff round or a bad quarter in the public record. In that case a successful article is a permanent, editable page collecting all of it, and you can't take it down. Think about that first.