Wikipedia and Your Brand: Editing Rules and Edit Requests

Wikipedia treats anyone with a close connection to a subject as having a conflict of interest. That covers employees, founders, investors, and agencies or freelancers you pay. The norm is simple: disclose the connection, and don't edit the article about your own company yourself.

Paid editing carries a stricter rule. The Wikimedia Terms of Use require editors who are paid for their contributions to disclose their employer and client. Hiring an agency that edits quietly on your behalf does not remove the conflict. It moves it, and it adds the risk of an undisclosed-paid-editing finding attached to your brand name.

What this means in practice:

• You can point out errors and supply sources.
• Independent volunteer editors decide what goes into the article.
• Wording that reads like marketing will be removed, whoever adds it.

Wikipedia's policies are maintained by its community and change over time. Read the current conflict-of-interest guideline and the edit request help page before you post, rather than relying on any summary, including this one.

The honest position is narrow. No study in our evidence base measures how much a Wikipedia article changes what ChatGPT, Gemini, Perplexity or Claude say about a company. Do not let anyone sell you a number for it.

What the evidence does support is an association. AI answers lean heavily on a small set of prominent sources: a study of AI search citations (arXiv 2507.05301, data from March to May 2025) found that the top 20 news sources accounted for 67.3% of citations for OpenAI models. When engines cite other sites instead of yours, being described correctly on those sites is the route into the answer. That is an association, not a promised effect. A broader critical survey of 45 GEO studies (arXiv 2607.14035) found no technique with a stable, cross-platform causal effect on discoverability.

Timing also matters. It helps to separate two stages:

• Parametric answers come from training data frozen at a cutoff. A correction you get accepted today cannot reach those answers until a later model is trained.
• Retrieval answers fetch pages at question time. A corrected article can show up there once an engine fetches it, if it fetches it at all.

If you are deciding whether you need an article in the first place, read whether you need a Wikipedia page to show up in AI answers. This page assumes one already exists and has something wrong in it.

Follow these steps for each error. One request per distinct problem keeps reviews fast.

1. Confirm the error. Write down the exact sentence in the article and what is wrong with it: a wrong founding year, an outdated CEO, a product you discontinued.
2. Find an independent source. Wikipedia prefers reliable, published, independent sources. A news article or official filing carries more weight than your own press release. Your own site is weakest for anything that sounds like a claim, and acceptable at best for plain facts like headquarters location.
3. Create or use a disclosed account. State your connection on your user page: who you work for, and whether you are paid to edit.
4. Post on the article's Talk page. Use the conflict-of-interest edit request format described on Wikipedia's edit request help page, so the request enters the review queue.
5. Write the change precisely. Quote the current text, give the replacement text, and link the source. Neutral wording only.
6. Wait and respond. Volunteers review requests on their own time. Answer questions politely and don't edit the article while you wait.
7. Record the outcome. Note the date, the request, and whether it was accepted, declined, or changed.

If the same wrong fact appears on directories, review sites or news pages, fix those too. Our playbook on asking third-party sites to fix wrong facts covers that route.

Run through this list before every request:

• The error is a fact, not a tone you dislike.
• You can quote the exact sentence that is wrong.
• You have at least one independent, published source for the correct fact.
• Your own About page states the same fact, in plain HTML, so the sources agree.
• Your account discloses your employer and any payment.
• The proposed text is neutral, short, and free of adjectives like "leading" or "innovative".
• You are not asking to remove sourced criticism just because it is unflattering.
• You have checked for a related Wikidata item. Wikidata is a free, collaborative knowledge base that stores structured facts such as founding dates, and it has its own policies to read first.
• You have logged the request date so you can follow up.

Good results

• An independent editor accepts the request and updates the sentence with your source.
• An editor rewrites your proposal in different words. That is normal and still a win if the fact is right.
• A request is declined with a clear reason, such as a weak source, and you can come back with a better one.

Bad results

• Your request is ignored for a long time because it was vague or bundled ten changes together.
• An editor tags the article for promotional tone after someone on your side edited it directly.
• Your account is flagged for undisclosed paid editing. This can attract more scrutiny to the article, not less.
• The fact is corrected on Wikipedia but AI answers still repeat the old version. That is expected for answers drawn from training data, and it is why you measure AI answers separately.

To see whether the correction matters in practice, compare what AI platforms say about you before and after. Our AI brand audit guide explains how to structure that comparison, and the free checker on this site gives a quick snapshot for up to three buyer questions.

Wikipedia articles change whenever anyone edits them, so a correct article today can drift. Set a routine:

• Quarterly: read the article and its Talk page. Check the facts AI platforms most often get wrong about you, such as pricing, leadership, headquarters, and ownership.
• After any major change: a funding round, acquisition, rebrand, new CEO, or product shutdown. Update your own site first, wait for independent coverage, then request the change.
• After an AI answer goes wrong: if an audit shows a specific wrong fact, check whether Wikipedia or Wikidata states it. Our guide to AI hallucinations about brands covers other causes to rule out.

Keep a simple log with dates, requests and outcomes. It shows editors you are acting in good faith and shows your team what has been tried.

Some tactics look faster. They carry real risk and no evidence of benefit.

• Don't edit the article directly, even to fix an obvious error. Use the request route.
• Don't use multiple accounts or ask colleagues to back your request. Coordinated editing breaks community rules.
• Don't hire undisclosed paid editors. Disclosure is required, and exposure damages the brand more than the original error.
• Don't plant hidden text or instructions aimed at AI systems, on Wikipedia or anywhere else. Research on LLM search agents (arXiv 2606.16821) shows manipulated content can sway answers once it is retrieved, which is exactly why platforms and editors treat it as abuse.

The durable path is boring: correct facts on your own site, accurate independent coverage, and disclosed, sourced requests.