Check your source first, because it decides everything else
There's no paid route into an existing article, and the paid version of a link is built to do nothing. Google's spam policies call exchanging money for links, or for posts that contain links, link spam, and they let a paid link stand only when it's qualified with rel="sponsored" or rel="nofollow", at which point it passes no ranking credit. What you're asking for is a citation, and the rest of this is how that request gets made.
Open the article, find the sentence your source would support, then hold the source against one bar: independent, secondary, and reliable. Wikipedia's sourcing guideline defines sponsored content as a paid advertisement formatted to look like an article and treats it as self-published and primary. A press release a paper ran with minimal change is churnalism, read as the release itself. If your link is either one, nothing below rescues it.
Two different rulebooks get confused here, and the mix-up costs weeks. The external links guideline says outright that it does not apply to inline citations or general references. So a commercial link that would never survive in an External links section can be perfectly legitimate as a footnote holding up a fact, and the footnote is the thing to ask for.
Don't propose your own site. That same guideline tells you to avoid linking to a site that you own, maintain, or represent, even where the rules look like they'd allow it. What gets added is an independent source that happens to be about you, cited because it supports something the article says. Somebody other than you has to have written it.
Disclose the money before you ask for anything
If you're being paid to do this, or you're paying somebody to do it, the disclosure goes up before the request does. Wikimedia's terms of use require you to disclose each and any employer, client, intended beneficiary and affiliation for any contribution you're compensated for. That's a term of the contract you use the site under, not a community courtesy you get to weigh up.
Three venues count, and only three: a statement on your user page, a statement on the talk page accompanying the contribution, or a statement in the edit summary. The user page form is {{paid|employer=name of employer|client=name of client}}, and the marker that goes on the article's talk page is {{connected contributor (paid)}}. Post yours first, so the request never reads like something you tried to slip past.
The paid-contribution policy is blunt about the consequence: paid editors who cannot disclose their employer, client and affiliations are prohibited from editing, and non-disclosure agreements do not invalidate this requirement. If your client can't be named in public, there's no compliant version of this job, not by you and not by anyone you hire to do it quietly.
Post the request on the article's talk page
Do not edit the article. The conflict of interest guideline says you're strongly discouraged from editing affected articles directly, with one narrow opening: you may add independent reliable sources when another editor has asked for them, and even then it's better to supply them on the talk page for somebody else to add.
Go to the article's talk page and click Add topic. On a protected article there's a second way in: the article's View source tab carries a Submit an edit request link at the bottom right that preloads the template onto that same talk page. Don't take it to the COI noticeboard, which exists to decide whether an editor has a conflict of interest and will buy you scrutiny of your account instead of action on the article.
Paste the template unindented, with no leading colon and no asterisk, or the request never enters the queue: {{edit COI|summary=Add Short summary of requested change here}}. Under it go four things and nothing else, which the template's own page lists in order: the exact text to add or remove, the reason for it, at least one URL supporting it, and your signature, which is four tildes, ~~~~.
Write the ask as a diff, and the citation as a citation
The edit request guidance hands you the constructions to use: Change X to Y. Insert X after/before Y. Remove X. Quote the existing sentence exactly, then give the exact sentence you want in its place. A paragraph describing the outcome you'd like gets declined as non-specific, and the reviewer will not supply the judgment you left out.
Hand over a citation nobody has to rebuild. The citing guideline wants enough to identify the source: URL, author, article title, website name, publisher and publication date. Then match the format the article already uses, because each article should use one citation method or style throughout, and a footnote in a foreign style is extra work for the volunteer doing you a favor.
Two mechanical details reviewers do enforce: the marker goes after adjacent punctuation such as periods and commas, with no space before it, and it never goes inside a section heading. If you want to hand them finished markup, Wikipedia's referencing help page walks through the pieces it is built from, a <ref> pair around a {{cite web}} or {{cite news}} template.
Wait, then read the letter the reviewer leaves
Nobody can give you a date. The guidance says it can take anywhere from a day to several months for an edit request to be acted upon, and on a semi-protected article it sets a floor: wait at least two weeks before asking follow-up questions. Put a date in your calendar, then leave it alone until that date arrives.
The answer comes back as a letter on the template rather than a message to you. A means accepted and done. G means you're cleared to make the edit yourself. U means under review. P means partially done. D means declined. Open the article's history and read the actual change before you believe any of them, because partially done is the one people write down as done.
Don't file it twice. Requests sit in a public queue of conflict of interest edit requests that volunteers work through, and a duplicate doesn't move you up it. What moves you up it is a request a stranger can execute in a minute without having to decide anything.
When it comes back declined
Fix the defect it names, and change nothing else. ADV means the wording reads promotional, so strip the adjectives and re-propose it as plain fact. V means the source doesn't support the claim, and that needs a different source, not a better argument. S means it wasn't specific, so restate it strictly as Change X to Y.
The rest need a different move entirely. R means you asked to remove sourced content, which takes consensus rather than a request. C means no consensus, so it becomes a talk page discussion. LLM means it read as machine-generated, so write the next one yourself. And never answer a decline by editing the article, which is the move that turns a declined request into a conduct case.
Sometimes the honest answer is that there's nothing to cite. If no independent reliable source exists yet, no phrasing rescues the request, and the work moves upstream to earning real coverage first. Volume is its own offense too: the spam guideline defines citation spamming as the repeated insertion of a particular citation in multiple articles by a single contributor, and every one of them can look defensible while the pattern gets reverted.
If the article is on a news site, not Wikipedia
No newsroom has an add a link door. The one sanctioned channel into a published article is the corrections desk, and it acts on factual error, not on link inventory. ProPublica's corrections page is the shape to expect: a page headed Corrections, an email address, no form and no button. Each outlet sets its own, and the fix renders on the article itself as a dated line starting Correction, followed by the date and what changed.
So send an error report, not a request. Identify the article, quote the incorrect sentence, state the correct fact, and attach the primary evidence for it. The citation or the link you wanted is a byproduct of the outlet sourcing its own fix, never the ask itself. Ask for the link instead and you get no reply at all.
Read that outlet's own corrections page before you send anything, because the address and the process are set per publication and nobody else's page tells you theirs. Outlets don't publish a turnaround time, so don't promise anyone a date. If they decline there's no appeal route in the US, the outlet's decision is final, and the lever left is a documented factual dispute to the standards editor.