Does Wikipedia or Wikidata Presence Affect AI Citation?

Yes at the domain level. No as a personal guarantee. Wikipedia is one of the pages ChatGPT, Perplexity, and Google AI Overviews already credit when they need an encyclopedia node. That is not the same as "get a Wikipedia article and you will be cited." A missing article is not a rank you can buy. A thin stub you wrote about yourself is a deletion waiting to happen.

I am William Spurlock, founder of Spurlock Studios, AI Systems Architect, and Fractional AI CTO. I have been SEO-certified since 2021; the work now sits under AEO, AIO, and GEO. I have shipped 600+ automations with 500+ live, logged 20,000+ hours inside agentic systems, and deleted 35,000+ hours of client busywork across that book of work. I do not invent Wikipedia-rank tables, fake Q-ids, or a 2026 "wiki boost" percentage nobody published.

This spoke owns one question: Does Wikipedia or Wikidata presence affect AI citation? July 4
entity SEO
owns the brand node, schema, and entity home. June 11
digital PR for AI visibility
owns how you earn mentions on sites AI already reads. This page stays on the encyclopedia node: when a wiki page helps, when it does not, and what a small brand should do instead of chasing an article English Wikipedia will not keep.

I am not rewriting the entity pillar. I am not handing you a digital-PR outreach list. I am not telling you to open a second account and write your own article. If that is the plan, stop. Wikipedia's conflict-of-interest guideline and paid-contribution disclosure exist because that move burns the page and the editor.

---

Does Wikipedia or Wikidata presence affect AI citation? #

Wikipedia presence affects AI citation the way a library card catalog affects research: the catalog is cited constantly; your unpublished pamphlet is not. The public studies measure wikipedia.org as a source domain. They do not measure "brands with a page get N% more citations." Treat those as two different claims.

I keep three layers separate so the sentence stays honest:

Layer	What it is	What the citation studies actually count
Encyclopedia domain	wikipedia.org as a URL the engine credits	Mention share and citation share for that domain
Your article	A standalone English Wikipedia page about you	Not measured as a brand-level lift in the studies below
Your Q-id	A Wikidata item (Q plus a number)	Almost never the URL on the citation chip

Ahrefs Brand Radar, June 11, 2025 looked at about 76.7 million Google AI Overviews, 957,000 ChatGPT prompts, and 953,500 Perplexity prompts for June 2025. Wikipedia led mention share on all three: 16.3% on ChatGPT, 12.5% on Perplexity, 8.4% on AI Overviews. That is domain gravity. It is not a score for your company page.

Semrush, November 2025 ran 230,000 prompts across ChatGPT search, Google AI Mode, and Perplexity from July 14, 2025 through October 12, 2025 and counted more than 100 million citations. Wikipedia stayed in ChatGPT's top two cited domains. Semrush also recorded a mid-September 2025 drop: Wikipedia appeared in roughly 55% of those ChatGPT responses before the shift and under 20% after. AI Mode held near 3%. Perplexity held near 0.8%. The domain still matters. The share is not a constant, and it is not your article.

Profound's citation-pattern write-up (dataset described as 680 million citations across ChatGPT, Google AI Overviews, and Perplexity, August 2024 through June 2025) put Wikipedia at 7.8% of ChatGPT's total citations and 47.9% of ChatGPT's top-10 source share. Google AI Overviews in that same write-up put Wikipedia much lower in the mix (0.6% of total citations in their AIO table). Different crawls. Different denominators. Same directional fact: ChatGPT leans encyclopedia harder than Google's snapshot does.

I will not mash those percentages into one "Wikipedia rank." The methods disagree on the unit. Ahrefs reports mention share. Semrush reports the share of responses that cited the domain. Profound reports share of citation events. If a vendor slide turns that into "get a wiki page, gain 16%," they invented the last step.

Wikidata is the quieter sibling. It is a structured knowledge base with a stable item ID. Foundation models and retrieval stacks use those IDs as join keys. The citation chip a buyer sees is still usually Wikipedia, a review site, a docs page, or your own URL — not wikidata.org/wiki/Q…. Presence here is an entity node, not a byline.

The Wikimedia Foundation is not shy about the training-side role. On April 30, 2025, Chris Albon and Leila Zia wrote that Wikipedia's success is "why Wikipedia is at the core of every AI training model." The Foundation had already made the human-curation case on July 12, 2023. As of May 13, 2026, Wikimedia Enterprise is still shipping structured English and French Wikipedia extracts with Wikidata Q identifiers attached, explicitly framed for model work and retrieval. Monthly Wikimedia dumps remain public. That is how an encyclopedia node gets into GPT-5.5, Claude Opus 4.8, Gemini 3.1 Pro, and Llama 4 long before your homepage does.

So the honest answer is a split:

1. Yes — Wikipedia as a corpus and as a cited domain affects how answer engines write definitional answers.
2. Only sometimes — a page about your brand affects whether you are the entity they resolve.
3. Almost never as a shortcut — a Q-id with no serious public references does not mint citations on buying queries.

If you only remember one sentence: engines cite the encyclopedia they already trust; they do not award you a page because you want to be in that pile.

Training dump versus live retrieval #

I split "Wikipedia affects citation" into two clocks, because founders mash them.

Clock	What happened	What it does for you
Pre-training / dumps	The model ate Wikimedia text, often upsampled	Famous entities and generic facts get defaulted. Your unbuilt article was never in that pile.
Live retrieval	ChatGPT with search, Perplexity, Gemini grounding, AI Overviews	The engine can fetch a URL that exists today. A page created last week can appear here. It will not rewrite last year's weights.

The April 2025 Foundation line is about the first clock. Ahrefs, Semrush, and Profound are about the second: which domains show up as sources in answers. A small brand lives or dies on the second clock. You cannot backfill a dump. You can publish a durable URL — and for most of you that URL will not be English Wikipedia.

Semrush's mid-September 2025 ChatGPT drop is the receipt that retrieval settings move the encyclopedia share without anyone "optimizing" a page. If Wikipedia's response-level share can fall from the mid-50s to under 20% on one engine in a few weeks, your personal article was never a stable ranking. It was a passenger on someone else's retrieval policy.

I also refuse the slide that says "Wikidata is in the training set, therefore a Q-id cites you." Dumps include items. Citation chips include URLs. Those are different artifacts. If your buying prompt cites a G2 profile and your homepage, the Q-id did not fail. It was never the thing being cited.

---

When does a Wikipedia page help ChatGPT, Perplexity, or AI Overviews — and when does it not? #

A Wikipedia page helps when the query is "what is this named thing," the article is stable, and independent sources already agree on the facts. It does not help when the query is "who should I hire," the article is a stub you cannot defend, or the engine is pulling reviews, videos, and forums instead of an encyclopedia.

That is the whole fork. Everything else is examples.

When the page actually does work #

I have watched this play out on definitional and disambiguation prompts, not on "best X near me."

• The buyer asks what you are. "What is [Brand]?" / "Who founded [Brand]?" / "Is [Brand] a real company?" A live article with a sourced lead and an infobox gives ChatGPT and Perplexity a page they can quote without inventing a founding year. Google's Knowledge Graph Search API examples return detailedDescription blocks that point at Wikipedia URLs for famous entities. That is Google showing its homework on known nodes. It is not a promise that your future stub becomes a Knowledge Panel.
• The name collides. Two agencies, one surname, one city. A Wikipedia article — or even a sourced section on a parent article — is a disambiguation edge. Wikidata's instance of and official website statements do the same job in a form a model can join. Without that, GPT-5.5 and Claude Sonnet 5 will happily merge you with the larger namesake.
• The facts already match your entity home. Same legal name, same founding year, same official URL as your Organization schema and About page. The encyclopedia node confirms the site node. That is entity reconciliation, not a new marketing channel.
• The article survived time. Months old. Independent citations. No {{promotional}} banner. No active Articles for Deletion fight. A page that is about to die is a worse signal than no page.

When the page does not help — or hurts #

This is the list I wish more founders heard before they hired a "Wikipedia specialist."

Situation	What the engine sees	What you should assume
Buying and shortlist queries	Reviews, comparison pages, Reddit, YouTube, your service URL	Wikipedia is optional. Ahrefs' June 2025 mix already shows YouTube and Reddit competing with Wikipedia on Perplexity and AI Overviews.
Local service queries	Google Business Profile, maps, NAP, LocalBusiness facts	An encyclopedia article about a 12-person shop is rare and usually fails WP:NCORP.
No independent coverage	A page that cannot meet significant coverage	The article gets declined or deleted. You spent a quarter on a ghost.
COI / paid / sockpuppet edit	A promotional voice, a new account, undisclosed pay	The page is tagged, shortened, or removed. You taught the web you manufacture sources.
Stub with one press release	A paragraph that restates your homepage	Retrieval may skip it. If it is cited, it cites a thin claim you could have put on your own site.
Deleted or draft-only page	A red link, a log, an AfC rejection	Some models still remember an old dump. Live search (Perplexity, ChatGPT with search, Gemini 3.5 Flash grounding) will not find a durable URL.

WP:NCORP is blunt. An organization is generally notable if it has been the subject of significant coverage in multiple reliable secondary sources independent of the subject. Trivial mentions, routine announcements, interviews you sat for, and material you published about yourself do not get you there. Public listing, revenue, and headcount are not substitutes. I am not going to invent a "you need 12 press hits" rule. The guideline asks for depth, independence, and more than one source. Most small brands do not have that, and pretending they do is how articles die.

The same page says no company is inherently notable and notability is not inherited. A famous founder does not mint an article for the studio. A notable parent company does not mint an article for the regional shop. "We exist" is not a source. NCORP also warns that articles about very small local companies are typically unacceptable because Wikipedia is not a soapbox.

Each source has to clear four gates at once: significant, independent, reliable, and secondary. NCORP's own worked example ("Acme Inc.") fails all four of its citations — a one-line Times mention, a Forbes.com contributor profile, a tech-blog review, and a court filing — and ends at zero qualifying sources. That is the bar agencies skip when they say "we got you a mention, so you are notable."

Trivial coverage, copied from the guideline so I do not invent a list:

Counts as trivia (does not establish notability)	Why founders still try to use it
Funding, hire, launch, merger, or earnings notices	They look like "press" in a media-monitoring tool
Branch openings, trade-show booths, panel quotes	Easy to schedule. Shallow on the company itself
A sentence inside a piece about someone else	The subject of the source is not you
Awards nobody outside the industry can name	The award page is often a paid listing
Directory rows, hours, SKUs, patents, lawsuit dockets	Listings and primary records, not secondary analysis

Quantity does not become quality. Twenty launch notices are still twenty launch notices. Employee count and revenue do not upgrade them. A 400-word independent feature can outrank a one-sentence mention in a larger newspaper. Depth beats logo size.

Wikidata notability (WD:N) is a lower bar, and that is the trap. An item is acceptable if it has a valid Wikimedia sitelink, or it refers to a clearly identifiable entity that can be described with serious and publicly available references, or it fills a structural need for other items. Criterion 2 is how a company without a Wikipedia article still gets a Q-id. Criterion 2 is also how a vanity item with a homepage and a LinkedIn URL gets challenged. As of late 2025 the community opened an RfC on reforming that policy. Do not build a 2026 plan on "Wikidata is free Wikipedia."

Engine split, in plain language #

I run the same brand through three prompts and I expect three different encyclopedia behaviors.

1. ChatGPT (GPT-5.5, with search). Most likely to quote Wikipedia on "what is X" if the article exists. Least likely to invent a vendor from a Q-id alone when the buyer asked for a shortlist. Profound and Ahrefs both show the encyclopedia lean. Semrush shows that lean can shrink when retrieval settings change.
2. Perplexity. Live web first. Wikipedia can appear. So can YouTube, trade press, and your own docs. Ahrefs' June 2025 Perplexity mix already had YouTube ahead of Wikipedia on mention share. A wiki page is one candidate, not the ticket.
3. Google AI Overviews (Gemini 3.1 Pro class snapshots). More of a search-derived citation list. Wikipedia shows up for entities Google already treats as encyclopedia topics. For a contractor or a specialist studio, the Overview will cite the pages that already rank and the sites Google trusts in that vertical. Your missing Wikipedia article is usually not the reason you lost the chip.

Claude Opus 4.8 and Claude Sonnet 5, when they browse, behave closer to "read the open web and the encyclopedia if it exists" than to "must have a wiki page." I still do not treat a Q-id as a Claude citation.

A mention is not an article #

Founders hear "we got you on Wikipedia" and picture a blue link with their logo. What they often have is a sentence on someone else's page, a category list, or a footnote.

What you actually have	Entity-node value	Citation value
Standalone article	High, if stable and sourced	Can be quoted on "what is X"
Section on a parent article	Medium — you are an attribute of a larger node	Sometimes fetched; you do not own the URL
One sourced sentence	Low — a mention	Closer to digital PR than to an encyclopedia home
Category or "see also" only	Near zero	Almost never the chip
Deleted / redirected away	Negative or none	Live retrieval loses the URL; old dumps may still lie

I will take a sourced sentence on a page that will still be there next year over a founder-written stub that lasts six weeks. The sentence is evidence. The stub is a risk. Neither one is a hiring recommendation.

If a reporter already wrote the independent piece, you do not need to "get it onto Wikipedia" this quarter for ChatGPT to retrieve the reporter's URL. Perplexity and AI Overviews can cite the original. Wikipedia would have cited it too. You skip the volunteer gate and keep the source.

My opinion, held loosely: if your category is "who should I hire / which tool should I buy," a Wikipedia article is a trophy. The citation you wanted was always going to come from a comparison page, a review, or a service URL that answers the question. Build those. The July entity pillar and the August
ChatGPT / Perplexity recommendation playbook
already own that work.

---

Is a Wikidata Q-id enough for a small brand? #

No. A Q-id is an identifier. It is not a citation, a Knowledge Panel, or a substitute for sources. It helps when other systems can join on it. It does nothing when the item is a self-issued stub with no references a volunteer would keep.

Think of Wikidata as a row in a shared spreadsheet the web is allowed to reuse. The columns that matter for a brand are boring on purpose:

Property (examples)	What it records	Why an answer engine cares
P31 instance of	Organization, business, person	Type. Stops the model from treating you as a product or a place.
P856 official website	Canonical URL	Join key to your entity home.
P112 founder / P169 CEO	People nodes	Leadership edges. Useless if the person item is also empty.
P571 inception	Founding date	A fact models love to hallucinate. Only keep it if a source exists.
External IDs	Registries, VIAF, GND, social handles	Crosswalks. Not vanity badges.

Help:Sources is the part founders skip. A statement without a reference is a rumor with a PID. Volunteers delete rumors. Models that ingest dumps will still see the rumor for one training cycle, then the next dump will not. That is not a visibility strategy. That is a time bomb with a Q number on it.

What a Q-id is good for #

• sameAs on your own site. schema.org/sameAs can point at https://www.wikidata.org/wiki/Q… the same way it points at LinkedIn. That is identity reconciliation. The mechanics live in
  how structured data helps AI understand and cite your business
  and the July entity pillar. I am not pasting another JSON-LD block here.
• Join keys in dumps. The May 2026 structured Wikipedia dataset attaches Wikidata IDs to articles. If you ever do have an encyclopedia page, the Q-id is how language editions and infoboxes stay one entity.
• Disambiguation when the name is common. A sourced official-website claim is sometimes enough for a retrieval system to stop merging you with a larger namesake.

What a Q-id is not #

• Not a Google Knowledge Panel. Google's Knowledge Graph Search API lets you look up entities Google already has. A Wikipedia detailedDescription in the Taylor Swift example is a famous-entity pattern. I have not seen Google publish "create a Wikidata item, receive a panel." If a panel appears after you create a Q-id, something else moved: coverage, a Knowledge Graph match, a Google Business Profile, or time. Do not credit the Q-id by default.
• Not a ChatGPT recommendation. Recommendation is a shortlist behavior. The August playbook covers that. A Q-id does not put you next to two competitors on "who should I hire."
• Not a way around NCORP. Wikidata criterion 2 still wants serious public references. If you do not have those, you do not have a durable item. You have a row someone can nominate for deletion.

I have seen agencies sell "we will get you on Wikidata this week" as if it were a citation program. It is data entry. If the references are your homepage, your Crunchbase, and a press release you wrote, you built a mirror of assets you already control. Answer engines already had those.

The test I use: if you deleted the Wikidata item tomorrow, would any independent URL that names you disappear? If the answer is no, the Q-id was never the asset. The independent URLs were.

---

What should a small brand do instead of chasing a Wikipedia article? #

Do the work Wikipedia would have demanded as evidence, then stop at the encyclopedia door unless a volunteer would keep the page without you in the room. For most operators that means a clean entity home, matching structured data, and a short list of independent pages that already name you. It does not mean a secret article.

I will be blunt. If you cannot pass WP:NCORP today, you will not pass it because you want AI citations. The coverage gap is the job. The Wikipedia article is a lagging certificate.

The order I actually run #

This is not a six-month PR program. That lives on the digital PR spoke if you need the mention-earning motion. Here I only sequence the encyclopedia decision.

1. Prove you are one entity on property you control. Legal name, founder, official URL, and service category stated in plain sentences on the About page. Same strings in Organization or LocalBusiness JSON-LD. Same strings on Google Business Profile if you are local. If those disagree, a wiki page would only freeze the conflict.
2. Collect the independent URLs you already have. Trade press. Association directories. Registry filings. Review platforms with a real profile. University or conference pages. If you cannot list three URLs that a stranger wrote, you are not in a Wikipedia conversation. You are in an entity-home conversation.
3. Decide, in writing, whether those URLs are significant coverage. Significant means a reader could write more than a stub from them. A logo in a "clients" row is not coverage. A sentence in a roundup is usually not coverage. WP:GNG and NCORP both care about depth.
4. If the pile is thin, spend the next quarter on the pile. Guest expert quotes, registry completeness, and one durable industry page beat a draft article. I will not script the outreach. I will tell you the encyclopedia is downstream of it.
5. If the pile is real, let a volunteer find it. You may disclose a conflict of interest and use Articles for Creation. You may hire disclosed paid editing and still lose. You may do nothing and still get an article years later because a journalist already did the work. All three beat sockpuppetry and autobiography.
6. Only then consider Wikidata. Create or request an item when criterion 2 is true with references you did not write. Add sameAs. Stop. Do not "complete" forty properties with guesses.

What I refuse to recommend #

• Writing the article from a founder account. WP:AUTO exists. The page will read like a brochure because it is one.
• A second account "so it looks independent." That is sockpuppeting. It is also how you get the topic salted.
• Paying someone who will not disclose. The Wikimedia Terms of Use require paid-contribution disclosure. Hidden pay is a policy violation, not a growth hack.
• Buying a "guaranteed Wikipedia page" package. If they can guarantee it, they are not describing the volunteer process I just linked.
• Treating a mention on a celebrity's page as your article. A sentence on someone else's article can help disambiguation. It is not your node. Do not vandalize a larger article to insert a company.

Substitutes that actually show up in answer engines #

I want you to put the hours where GPT-5.5 and Perplexity already look when the query is commercial.

Substitute	Why it substitutes for a wiki page	Where it lives on this site
Entity home + schema	Machine-readable identity without a volunteer gate	Entity SEO pillar, structured data spoke
Independent mentions	The same raw material NCORP would have required	Digital PR for AI visibility
Answer-ready service pages	What buying prompts retrieve when Wikipedia is silent	Get ChatGPT and Perplexity to recommend you
Review and registry profiles	Third-party nodes with less editorial theater	Same two posts; do not clone them here

A 12-person company that spends six months on those four rows will have more citation chances than the same company spending six months arguing on a talk page. I have never seen the inverse for a brand that failed NCORP on day one.

If you already have a stable article you did not write, protect it by not touching it. Update your own site so the facts match. File a talk-page request for a factual error with a source. That is maintenance. That is not a campaign.

---

How do you check whether encyclopedia presence is doing anything? #

You check three surfaces you do not control, then you run two prompts you do control. Wikipedia search, Wikidata search, and a Knowledge Panel lookup tell you if a node exists. "What is [Brand]?" versus "Who should I hire for [category] in [place]?" tell you if the node is doing citation work or sitting in a dump.

I do this in one sitting. Thirty minutes. No vendor seat required.

Surface 1 — English Wikipedia #

Search the exact legal name, then the public brand name, then the founder.

Result	Read it as
Standalone article, stable, sourced	You have an encyclopedia node. Do not "optimize" it.
Section on a parent article	Partial node. Useful for disambiguation. Not your page.
Redirect to a larger topic	The project decided you are not a standalone subject.
Red link / no hit	No English article. Stop planning around one.
AfD, promotional tag, or draft	Negative or pending. Treat as no article until it is quiet.

Open the page history if an article exists. A burst of edits from new accounts in one week is a smell. A slow history with many usernames is the healthy pattern.

Surface 2 — Wikidata #

Use Special:Search for the same strings. Open the item.

Check, in this order:

1. Does P856 match your real official URL? If it points at a parked domain or a dead Webflow, the join key is wrong.
2. Do the statements have references? Hover the references. Homepage-only references are weak. Registry or independent news is stronger.
3. Is there a Wikipedia sitelink? Criterion 1. If yes, you already knew from Surface 1.
4. Is the description a slogan? Slogans get rewritten. Facts stay.

If you find two items for the same company, that is a split node. Merging is a volunteer process. Do not create a third item to "win."

Surface 3 — Google entity lookup #

Search the brand name in an incognito window. A Knowledge Panel on the right is Google saying it has a node. No panel is not a failure of Wikidata. Plenty of real businesses have no panel.

If you want the machine interface, Google documents the Knowledge Graph Search API. A returned entity with a Wikipedia detailedDescription means Google is willing to show encyclopedia text for that node. A miss means Google has not attached that text. I do not treat a miss as "create Wikipedia now." I treat it as "Google does not have an encyclopedia-backed card for you yet."

Prompt pair — the only citation test that matters #

Use the same two prompts in ChatGPT (GPT-5.5 with search on), Perplexity, and Google's AI Overview on a matching query. Freeze the wording.

1. Entity prompt: "What is [legal name], and what does the company do?"
2. Buying prompt: "Who should I hire for [service] in [market]?" or "What are the options for [category] if I need [constraint]?"

Score each engine with four labels. No composite index. No invented weights.

Label	Meaning
Named + wiki cited	Your name appears and wikipedia.org is a source
Named + other cited	Your name appears; the sources are not Wikipedia
Named, no source	A mention without a URL you can open
Absent	You are not in the answer

I care about the buying prompt more than the entity prompt. A wiki-backed definition that never appears on the hiring question is a trivia win. Trivia does not pay invoices.

Re-run the pair in 30 days if you shipped entity-home or schema changes. Do not re-run it the morning after you create a Wikidata item. Dumps and retrieval caches are not your deploy hook.

If the entity prompt cites Wikipedia and the buying prompt still names two competitors, you have your answer: encyclopedia presence affected the definition, not the shortlist. Spend the next month on the shortlist surfaces, not on a second wiki draft.

A scorecard you can fill without a vendor #

I use one row per engine. I do not average them. Averaging hides a Google miss behind a chat definition.

Engine	Entity prompt	Buying prompt	Wiki URL present?	Decision
ChatGPT (GPT-5.5, search on)	Named + other cited	Absent	No	Entity home is working. Shortlist is not. Do not start a wiki draft.
Perplexity	Named, no source	Named + other cited	No	They can say the name. The cite is a review or your site. Keep those URLs clean.
Google AI Overview	Absent	Named + other cited	No	The snapshot is using search-derived pages. Encyclopedia was never the chip.

Those rows are a pattern, not a client. If your ChatGPT entity prompt is Named + wiki cited and the buying prompt is still Absent, you already have the rare case this post is about: the encyclopedia node resolved you, and it still did not recommend you. That is success on the wrong question.

Write the date on the sheet. Screenshot the answer. The next run is a comparison, not a vibe.

---

Frequently Asked Questions #

Does having a Wikipedia page help you get recommended by ChatGPT? #

It can help ChatGPT describe you. It rarely gets you onto a hiring shortlist by itself. Ahrefs' June 11, 2025 Brand Radar pass put Wikipedia at 16.3% mention share on ChatGPT — the highest of the three engines they published. That is the encyclopedia domain. Your article is useful when the prompt is "what is this company." Recommendation still needs third-party proof and answer-ready pages, which I covered in the ChatGPT and Perplexity playbook.

What is the difference between Wikipedia and Wikidata for AI citation? #

Wikipedia is a sourced article people and models can quote. Wikidata is a structured item with a Q-id that other datasets join on. Citation studies keep putting wikipedia.org in the top domain lists. They do not put wikidata.org in those same top-10 mention tables. Use Wikidata as an identity key and sameAs target. Use Wikipedia, if you ever have a stable page, as a quotable node. Do not treat the Q-id as a byline.

Can I write my own Wikipedia page for my company? #

You should not write it as if you were a volunteer with no stake, and you must not hide the relationship. WP:COI strongly discourages direct editing of affected articles. WP:PAID requires disclosure if anyone is compensated. WP:AUTO exists because founder-written pages read like marketing. Use Articles for Creation with sources you did not write, or wait. A deleted article is worse than no article.

Does a Wikidata item get you a Google Knowledge Panel? #

No published Google document promises a panel from a Q-id. The Knowledge Graph Search API searches entities Google already stores. Famous entities in Google's own examples often carry Wikipedia text. That is a display pattern for known nodes, not an application form. If you want a panel, the boring path is still consistent NAP, a complete Business Profile where it applies, and coverage Google can reconcile — not a weekend of Wikidata fields.

Should I hire someone to create a Wikipedia article? #

Only if they will disclose paid editing, use Articles for Creation, and tell you they might lose. Anyone who guarantees a live article is selling a process Wikipedia does not run. Hidden paid editing violates the Wikimedia Terms of Use disclosure rules. I would rather you hire help to earn the independent coverage NCORP requires than hire help to dress up a stub.

What if my Wikipedia page is nominated for deletion? #

Do not start a second account. Do not stuff the page with press releases. Add independent sources on the talk page or accept the deletion. An Articles for Deletion discussion is a volunteer review of notability, not a customer-support ticket. If the sources were never significant and independent, the page should go. Keep the facts correct on your own entity home so the next model dump does not freeze a fight.

Does a mention on someone else's Wikipedia article count? #

It can help disambiguation. It is not your encyclopedia node. A sourced sentence on a larger article is a mention, closer to digital PR than to a standalone page. Do not insert your company into a celebrity or city article to manufacture a sitelink. That edit will be reverted, and it teaches reviewers you treat Wikipedia as an ad slot.

Do AI models still train on Wikipedia dumps in 2026? #

Yes — Wikimedia is still publishing the dumps, and the Foundation still describes Wikipedia as core training material. Monthly XML and content exports remain on dumps.wikimedia.org. The May 13, 2026 structured Wikipedia dataset is framed for pre-training, fine-tuning, and retrieval. The Foundation's April 30, 2025 AI strategy said Wikipedia sits at the core of AI training models. Training on the corpus is not the same as citing your page.

---

Get an AI-visibility-ready site, not a fake encyclopedia page #

A Wikipedia article you will not get cannot be the plan. The work that survives is an entity-clear site: one official name, one About page that states facts, Organization or LocalBusiness schema with an honest sameAs list, and service URLs that answer the questions buyers already type into ChatGPT and Perplexity.

I build Premium AIO/AEO websites for operators who want that layer designed in. I will not create a sockpuppet article. I will not sell a Q-id as a citation program. I will tell you if your independent coverage is too thin for NCORP, and I will put the hours on the pages and schema you control.

If you want an AI-visibility-ready site built so answer engines can resolve your brand without a volunteer gate, use /contact. Bring the two prompt results if you already ran them. If you have a Wikipedia article you did not write, bring that URL too. If you do not, we will not pretend one is the next ticket.

Bring the Wikidata URL only if the statements have references you did not write. An empty Q-id is not homework. It is a row.

SEO-certified since 2021. Founder, AI Systems Architect, Fractional AI CTO. The encyclopedia is a node. It is not a ranking. Build the entity the models can join — then let Wikipedia stay optional.