Does Wikipedia or Wikidata help AI recommend my brand

Wikipedia can help an answer engine describe your brand when an independent encyclopedia article already exists, because that prose is widely reused as training and retrieval text. It is not a paid switch you flip to get recommended. Wikidata helps when the item is a real, referenced entity: a Q-id plus matching official-website and sameAs facts reduce name collisions. Neither object buys a citation rate, and I will not invent a percentage of ChatGPT answers that “come from Wikipedia.”

This spoke sits under the Answer Engine Optimization playbook. I have been SEO certified since 2021. That credential does not get you a Wikipedia article, and it does not make a Q-id appear in a chat product’s source panel.

The short answer

• Treat Wikipedia as a notability test with independent sources, not as a media buy. WP:NCORP is the company rule.
• Treat Wikidata as an identity record. Wikidata:Notability is easier to meet than English Wikipedia, and a sitelink is not required.
• A real item helps entity consistency. A promotional stub, a duplicate Q-id, or a Wikipedia draft that dies at deletion does the opposite.
• Google’s Knowledge Graph help says facts come from a variety of sources. A Wikipedia URL is not a required field, and a knowledge panel is not an AI recommendation.
• Measure cited / mentioned / absent / hallucinated on a frozen prompt panel. Do not report “we got Wikipedia” as the KPI.

What is Wikipedia for a brand — and what is it not?

Wikipedia is a volunteer encyclopedia. English Wikipedia’s company guideline is explicit: an organization is presumed notable if it has been the subject of significant coverage in multiple reliable secondary sources that are independent of the subject. The same page says only a small percentage of the world’s organizations meet that bar, and that no company is inherently notable because it exists, raised a round, or hired a notable person.

It is not a directory, a press room, or a product you can purchase from an “AI authority” vendor. What Wikipedia is not includes advertising and promotion. A page that reads like your About copy will be tagged, drafted, or deleted even if the facts are true.

Object	What it is	What a brand team wants it to be	What actually happens
English Wikipedia article	Volunteer encyclopedia entry	A ranking / citation switch	Independent editors decide keep, merge, redirect, or delete
Wikipedia draft / AfC	Review queue	A staging site you control	Reviewers apply NCORP; most company drafts fail on sources
Your About page	First-party claim	A substitute for independent coverage	Primary. Does not establish Wikipedia notability
Press release / wire	Owned distribution	“We were in the news”	Routine / primary. NCORP lists this class as trivial
Paid “Wikipedia SEO” package	Undisclosed or disclosed paid editing	A guaranteed article	Disclosure is required; direct article editing is strongly discouraged
AI recommendation	A model’s named suggestion	The thing you bought the page for	Separate job. Encyclopedia text is one input among many

Checklist before anyone opens a Wikipedia account:

• You can name two or more independent, in-depth, secondary sources about the company, not a product launch blurb
• Those sources are not your site, not a reprint of your release, not a founder interview that only quotes you
• You accept that editors — not your agency — will write the tone
• You can live with criticism, lawsuits, and funding history if independent sources already published them
• You are prepared to disclose paid involvement under WP:PAID

If the first box is empty, you do not have a Wikipedia problem. You have a coverage problem.

Articles for Creation is the honest on-ramp when sources exist and someone with a COI still wants a page. You submit a draft. A volunteer reviewer applies NCORP. Decline is the default for company drafts that are built from releases. Accept does not mean the article is “yours.” It means the community will now argue about it in public.

Procedure if you are going to try AfC anyway:

1. Finish the four-column source table in the NCORP section below. Stop if it fails.
2. Write the draft from those sources only. Every sentence should be something a volunteer could footnote.
3. Disclose paid involvement on the user page and the draft talk page.
4. Submit once. Do not shop reviewers with a second account.
5. If declined, read the decline reason. Most of the time it is “these sources are not independent.” That is the end of the Wikipedia workstream until the coverage changes.

What is Wikidata, and why is the bar different?

Wikidata is a structured knowledge base: items with Q-ids, properties, and referenced statements. It is not a second Wikipedia article. There is no lead paragraph for a model to quote. There is a graph: instance-of, official website, inception, headquarters, founded-by.

Wikidata:Notability accepts an item if it meets at least one of three tests: a valid sitelink to a Wikimedia project; a clearly identifiable conceptual or material entity described with serious, publicly available references; or a structural need (the item is required to make other statements useful). Criterion 2 is how a real company can exist on Wikidata with no English Wikipedia page.

That is not a free-for-all. Unreferenced promotional items still get proposed for deletion. A LinkedIn URL alone is a weak reference. A government registry, a court record, a serious news article, or the official site used as a source for a specific statement is the usual pattern. Do not treat “we incorporated” as a Wikipedia-grade notability win. It is a Wikidata-shaped fact, if you can cite it.

Test	Wikipedia (NCORP)	Wikidata	Operator implication
Independent in-depth coverage	Required, multiple sources	Not required for criterion 2	Most SMBs fail Wikipedia and can still be a real Wikidata entity
Sitelink to Wikipedia	The article is the object	Optional	Q-id can exist first, or never
Tone / prose	Neutral encyclopedia prose	Labels, descriptions, statements	Models quote Wikipedia prose more readily than a property list
Paid editing	Disclosure + COI rules	Conflict-of-interest norms still apply	Do not sockpuppet either project
Typical failure	Articles for deletion	Unreferenced / duplicate item	Both can leave a mess in AI answers

Procedure to see whether you already have an item:

1. Search Wikidata for the legal name, the trading name, and the founder.
2. Open each candidate. Check official website (P856) against your canonical URL.
3. Check instance of (P31). A person item is not the company. A product item is not the company.
4. If two items describe the same organization, that is a merge job, not two “authority pages.”
5. If nothing matches, you do not invent a Wikipedia URL. You decide whether criterion 2 is honestly met.

A Q-id is an identifier. It is not a medal.

Sitelinks are a separate object. If an English Wikipedia article exists, the Wikidata item should sitelink to it — one page, one item. If the sitelink points at a redirect, Wikidata:Notability says that redirect link does not count toward criterion 1. Do not manufacture a sitelink to a section on someone else’s article and call it your Wikipedia page.

Signal on the item	Healthy	Sick
English label	Matches the homepage name	Three spellings, no aliases for the real dba
Description	Disambiguates industry + country	“Leading provider of solutions”
P856	Live canonical URL	Parked domain, HTTP redirect farm, competitor
References	At least one serious URL per contested fact	Empty, or only your blog
Duplicate items	One item	Two Q-ids, two official websites

How do answer engines actually use Wikipedia vs Wikidata?

Two different jobs get flattened into “AI authority.”

Prose reuse. Encyclopedia articles are long, stable, widely mirrored, and licensed for reuse. The Wikimedia Foundation’s April 2025 crawler note said automated requests collecting training data for large language models have put real load on Wikimedia infrastructure. That is evidence that Wikipedia is a heavily scraped corpus. It is not a statistic about what share of ChatGPT answers cite Wikipedia, and I will not invent one. A later March 2026 follow-up described throttling crawlers. Access patterns change. The corpus still matters.

Identity join. Structured identifiers help a system decide which “Acme” you are. schema.org sameAs is defined as a URL that unambiguously indicates the item’s identity, with Wikipedia and Wikidata given as examples. Google’s Organization structured data documents sameAs as a recommended property to help disambiguate your organization in Search. That is Search identity work. It is adjacent to, not identical with, a chat product recommending you as the vendor to hire.

Input	What a system can do with it	What it cannot do
Wikipedia article	Lift a definition, founding year, category, controversy	Guarantee you appear in a “best X” answer
Wikidata Q-id	Join names, aliases, official URL, identifiers	Supply a 200-word recommendation paragraph
Organization JSON-LD sameAs	Point Google (and anyone who reads the HTML) at those URLs	Create the Wikipedia page the URL invents
Knowledge panel	Snapshot of Knowledge Graph facts in Search	Proof that ChatGPT or Perplexity will name you
Your service page	The citeable offer sentence and table	Inherit notability from a Q-id

Google’s Knowledge Graph overview says facts come from a variety of sources, including licensed data and claimed-panel suggestions. It does not say “add Wikipedia or you will not exist.” Do not write a deck that pretends otherwise.

Wikimedia also sells high-volume access through Wikimedia Enterprise. That is infrastructure for reusers who should not scrape the live site. It is not a brand-recommendation product, and it is not a listing you buy so ChatGPT names you. Treat Enterprise as a reminder that Wikipedia text is a commercial input for some companies — still not a lever you purchase for your brand.

Dumps and mirrors lag the live article. A model trained or retrieved against a snapshot can keep a deleted stub or a wrong founding year after volunteers already fixed it. That lag is why “we corrected Wikipedia yesterday” is a weak SLA for this week’s sales call.

Path	What you control	Lag risk
Live wikipedia.org article	Talk-page requests, not the prose	Days to forever, volunteer-paced
Wikidata item	Statements you can reference	Faster than AfC; still public and reversible
Your HTML + schema	Full	Next crawl / next fetch
Chat product memory / index	None	Snapshot-dependent; you will not get a ticket number

Operator rule: if the prompt is “who should I hire for Y,” the engine needs a recommendable offer, third-party mentions, and consistent identity. An encyclopedia page can corroborate identity. It rarely is the recommendation.

Does a Wikipedia article make AI recommend my brand?

Sometimes it helps the engine describe you. That is not the same as recommending you.

A recommendation is a choice among vendors. Encyclopedia articles recap what independent sources already said, including the unflattering parts. If the only independent coverage of your company is a lawsuit and a funding recap, that is what a compliant article will emphasize. Models that condition on that page will not magically convert it into a five-star pitch.

A Google knowledge panel is a third object. Google’s Knowledge Graph help says panels are generated automatically from a variety of sources when systems decide a snapshot is useful. You can claim a panel and suggest edits if you are the official representative. That is Search identity. It does not mean Perplexity will put you in a “best [category]” list, and it does not require a Wikipedia article as a documented prerequisite.

Prompt class	Wikipedia’s usual job	Wikidata’s usual job	What still has to exist on your site
“What is [Brand]?”	Strong, if the article exists and is current	Labels + official website	About facts that match the item
“Who founded [Brand]?”	Can help if the article covers it	founded by if referenced	Founder bio that matches
“Best [category] for [job]”	Weak unless independent sources already frame you that way	Almost none	Comparison table, criteria, proof
“Is [Brand] legit?”	Mixed. Controversies travel	Identifiers, not testimonials	Receipts you can defend
Navigational “Brand pricing”	Often absent or stale	Not a price list	A live pricing URL

Procedure for an honest “does Wikipedia help recommendations” test:

1. Freeze 20–40 prompts a buyer would type, including the bare brand name and three “best / vs / near me” jobs.
2. Run each in Google AI Overviews or AI Mode, ChatGPT with search on, and Perplexity. Screenshot sources.
3. Label the brand: cited, mentioned without a link, absent, or hallucinated (wrong city, wrong offer, wrong legal name).
4. Note whether a Wikipedia URL or a Wikidata-derived fact (founding year, HQ) showed up in the answer text.
5. Repeat after any Wikipedia or Wikidata change — not the week you submitted a draft.

If you are absent on “best [job]” and present on “what is [Brand],” the encyclopedia did identity work. It did not win the recommendation. Fix the content cluster that answers the job, and the digital PR that independent writers can cite. Do not blame the missing Wikipedia page for a missing offer sentence.

Can Wikidata help if I will never pass Wikipedia notability?

Yes, when the item is real. No, when you are stuffing a graph with marketing copy.

A Q-id that matches your legal name, canonical URL, and founding date gives retrieval systems a join key. Chat products and Search both see collisions: two LLCs with similar names, a founder whose personal Wikipedia page is not the studio, a product line that shares a trademark. Consistent structured facts reduce the odds that the model fuses you with the other one.

It will not mint notability you do not have. Wikidata will not make a local HVAC company look like a public company. It will not replace the independent articles NCORP wants. If your only third-party mention is a directory listing, a Q-id is still optional hygiene, not a visibility program.

Situation	Wikidata worth doing?	Wikipedia worth attempting?	First move
Registered company, unique name, official site, registry you can cite	Usually yes, if no item exists	No, unless NCORP sources exist	Create or clean the item; wire sameAs
Same, but a competitor already occupies a similar label	Yes — disambiguate labels and official website	Only if you have independent coverage	Fix the collision before you publish more aliases
NCORP sources already exist; no article	Yes	Maybe, via Articles for Creation, disclosed if paid	Source table first, then AfC
Famous founder, obscure company	Founder item ≠ company item	Company still needs its own sources	Do not “inherit” the founder’s page
Purely local, no independent coverage	Optional, low stakes	No	Entity sheet on your own site
Agency promises a Wikipedia page in 30 days	Ignore the upsell	No	Fire the vendor

Checklist for “Wikidata only” work:

• Legal name, trading name, and canonical URL written on one internal fact sheet
• At least one serious public reference per statement you add (registry, news, official site used narrowly)
• P856 official website equals the URL in your Organization JSON-LD
• Labels in English match how you write the name on the homepage
• No invented Wikipedia sameAs

If you cannot point at a real-world entity a stranger could verify, skip the item. Wikidata is not a sandbox for brands that want to look indexed.

How do I know if my company even qualifies for Wikipedia?

Run NCORP as a source audit, not a vibe. WP:NCORP requires each source to be significant, independent, reliable, and secondary. Multiple sources that each fail one test still add up to zero. The guideline’s own example table walks a Times mention, a Forbes.com contributor profile, a tech blog, and a court filing — and scores the company at zero qualifying sources.

Independence has two layers. NCORP’s ORGIND test is not only “did we pay the publisher.” Intellectual independence means the piece contains original analysis, investigation, or fact-checking attributable to someone unaffiliated with you. A glossy feature that is your press kit with a byline still fails — even on a masthead you recognize. That is the test company drafts die on.

Routine coverage does not count toward significant coverage. NCORP’s list includes funding announcements, hiring, product launches, branch openings, earnings, and trade-show participation. A pile of Business Wire hits is not a Wikipedia article waiting to happen.

Source you want to count	Significant?	Independent?	Reliable?	Secondary?	Usually counts?
Your website / About	Maybe depth	No	First-party	Primary	No
Press release / wire reprint	Often no	No	Varies	Primary	No
Founder interview that only quotes you	Sometimes	Weak (ORGIND)	Depends	Often primary-flavored	Usually no
Staff “contributor” post on a big masthead	Yes	Often no	Often no	Secondary-shaped	Usually no
Independent reported feature about the company	Yes	Yes if unaffiliated	If the outlet is reliable	Yes	Yes
One-sentence mention in a roundup	No	Maybe	Maybe	Maybe	No
Award listicle you paid to enter	No	No	Weak	Weak	No
Court opinion discussing the company in depth	Yes	Mixed	The court record is a record	Often primary	Not for notability
Book or peer-reviewed paper with a real section on the firm	Yes	If independent	Often	Yes	Yes

Procedure:

1. Dump every URL you think “proves we are notable” into a sheet with four columns: significant, independent, reliable, secondary.
2. Score each as yes/no. A maybe is a no for the purpose of this decision.
3. Count the rows that are yes on all four. You need multiple.
4. If the count is below two, stop. Do not file an AfC draft to “get the process started.”
5. If the count is two or more, read those pieces again. Could a volunteer write a neutral article without your brochure? If not, you still fail.

I will not pretend Spurlock Studios, or a typical client studio, clears NCORP because we shipped hundreds of sites or 500+ automations. Those are receipts I will say on a sales call. They are not independent secondary coverage of the company as a subject.

Why is Wikipedia not a paid lever?

Because the project’s rules say so, and because undisclosed paid editing is how articles get deleted and accounts get blocked.

WP:PAID is policy: if you receive or expect compensation for a contribution, you must disclose employer, client, and affiliation on the user page, the talk page, or the edit summary. The Wikimedia Terms of Use require that disclosure. WP:COI strongly discourages people with a conflict of interest — including paid editors — from editing the affected article directly. The advised path is talk-page proposals and Articles for Creation.

A vendor who says “we will just publish it” is selling you a policy violation. A vendor who discloses and still writes promotional tone is selling you a draft that reviewers will decline. Either way, you do not own the page after it exists. Independent editors can rewrite it tomorrow.

Move	Allowed?	Typical outcome	AI-side cost if it blows up
Undisclosed paid article	No	Deletion, sockpuppet investigation	You become “the company that faked Wikipedia”
Disclosed paid direct edit of mainspace	Discouraged even when disclosed	Reverts, COI tags	Stale or hostile revision in the corpus
Talk-page request to fix a wrong founding year	Yes, disclosed	Slow, sometimes ignored	Correcting a hallucination source
AfC draft with real NCORP sources	The honest path	Review may still decline	Neutral article if it survives
Buying links to the Wikipedia page	Irrelevant to notability	Can look like spam	Does not create the article
Editing a competitor’s article down	COI / advocacy	Sanctions	You taught the model a fight, not a fact

Checklist if counsel or a founder still wants “our Wikipedia”:

• Legal reads WP:PAID and WP:COI, not the vendor one-pager
• You will disclose if anyone is paid to touch the topic
• Nobody from the company will edit the live article “just to fix tone”
• The draft is built from independent sources, not the pitch deck
• You have a plan for negative facts that independent sources already published

If any box is a shrug, you are not ready. Wikipedia will still be there when the sources are.

What should I do instead of hiring a Wikipedia vendor?

Earn the sources NCORP would actually count, and make the entity facts on your domain boringly consistent. That is the work that helps recommendations whether or not an encyclopedia page ever ships.

Independent coverage is digital PR for citations, not a Wikipedia retainer. On-site, a cluster that answers the buyer’s job is still the citeable object. A Q-id without a page that can be quoted is identity only.

Week	If you fail NCORP (most teams)	If you pass NCORP
1	Entity fact sheet + Organization JSON-LD + frozen prompts	Same, plus a source-assessment table
2	Wikidata item create/clean if criterion 2 is real	AfC draft from independent sources only
3–4	One independent story worth pitching; one cluster URL with a table	Disclose if paid; submit AfC; do not edit mainspace
Ongoing	Prompt log + fact-match	Watch the article for drift; request talk-page fixes

Procedure for the entity sheet (do this even if you never touch Wikimedia):

1. Write one line each: legal name, trading name, founding date, HQ city, canonical URL, one-sentence offer, categories you will not claim.
2. Diff that sheet against the homepage, About, schema, Google Business Profile, LinkedIn, and any existing Wikidata item.
3. Pick the canonical string for the name. Kill the three spellings.
4. Put the offer sentence in the first 60 words of the money URLs.
5. Only then add sameAs URLs that exist.

I will cite receipts I can defend: SEO certified since 2021, 500+ automations built, 20,000+ hours on agentic systems, 35,000+ hours saved for clients, hundreds of production sites. I will not cite a fabricated “Wikipedia boosts ChatGPT share by X%.” If a vendor slide has that number and no primary method, throw the slide out.

What belongs on a real Wikidata item?

Minimum viable identity, referenced. Not a press kit.

Useful properties for a company item, when you can cite them:

Property	ID	What to put	Typical reference
instance of	P31	company / business / whatever is accurate	Registry or independent source
official website	P856	Canonical https URL, no tracking junk	The site itself
inception	P571	Founding date you also print on About	Registry, independent profile
headquarters location	P159	City item, not a poetic region	About, registry
country	P17	Country item	Registry
founded by	P112	Person item if it exists and is the same human	Independent source preferred
legal form	P1454	LLC / etc. if you know the item	Registry
official name	P1448	Legal name if different from the label	Registry

Do not add awards you invented, revenue you will not put on a sales call, or a Wikipedia sitelink you wish you had. Do not create a person item for a living founder just to decorate the company. Biographies of living people have their own risk; skip unless the person already has independent coverage.

Procedure:

1. Search for duplicates again. Merge is better than a second Q-id.
2. Create the item only if criterion 2 is honest.
3. Add statements one at a time with references.
4. Write a short English description that disambiguates (“American web studio”, not “the leading AI-powered…”) — and do not use banned brochure adjectives there either.
5. Wait. Then point your site’s sameAs at https://www.wikidata.org/wiki/Q….

If an existing item has the wrong website, that is a high-priority fix. A Q-id that points at a parked domain or a competitor is worse than no item.

How do I wire sameAs without inventing a Wikipedia URL?

Only list URLs that unambiguously are you. schema.org sameAs is an identity join, not a wishlist. Google’s Organization documentation treats sameAs as recommended, with examples that look like profile pages on other sites. Put the JSON-LD on the homepage or About, not on every blog URL.

sameAs candidate	Include when	Skip when
Wikidata item	P856 matches your canonical URL	The item is a different org, a person, or a product
Wikipedia article	The article is about this org and is not a redirect to something else	The page is a draft, a deleted title, or a founder bio
LinkedIn company page	It is the official company page	A personal profile, a showcase page for a dead product
Crunchbase / registry / OpenCorporates	The record is this legal entity	A similarly named firm in another state
YouTube / X / GitHub org	Official, matching handle	A fan account or an old agency login
Wikipedia wish URL	Never	Always

Checklist:

• Organization.name matches the homepage visible name
• url is the canonical site
• logo meets Google’s crawlable-image rules if you set it
• sameAs array contains zero 404s and zero Wikipedia titles you do not have
• Wikidata P856 and schema url are the same host
• Rich Results Test / URL Inspection sees the block in fetched HTML
• You did not paste a competitor’s Q-id because the label search was messy

Google’s Organization docs describe sameAs as a URL on another site with more information about the organization — a social or review profile is the example they give. schema.org is more specific: Wikipedia page, Wikidata entry, or official website as identity URLs. Use both documents as a join-key spec, not as a promise that Search or chat will recommend you.

Failure here is a self-inflicted hallucination. The model is not “bad at brands.” You gave it two official websites.

What failure modes actually show up in AI answers?

The expensive ones are identity collisions and promotional stubs that get deleted after models have already ingested them.

Promotional stub. An agency publishes a brochure article. It fails NCORP or gets draftified. For a window, crawlers and dumps still contain the puff version. Then the live page is a deletion log or a redirect. Chat products oscillate between the old puff and “no page found.” You paid for volatility.

Wrong Q-id in sameAs. Your schema points at another company’s item. Google’s disambiguation job and any graph-using system now have a documented lie. Answers pick the other HQ, the other founding year, or the other product line.

Founder page ≠ company. A notable person has a Wikipedia article. The studio does not. Models treat the person’s employers as interchangeable, or they invent that the studio is the person. Your offer sentence never appears because the retrieved entity is a biography.

Stale encyclopedia facts. Independent sources freeze a headcount, a city, or a product name. You moved. Nobody filed a talk-page request. The model is faithfully wrong.

Dump lag after a good edit. Volunteers fix P856. A dump or a cached retrieval still has the old URL. Your prompt log looks “broken” for two weeks. Annotate the date of the Wikimedia change and retest on a schedule, not in the Slack thread the same afternoon.

Inherited notability theater. NCORP is explicit: a company does not inherit notability from a notable founder, a notable client, or a notable subsidiary. Models still blur those edges. If your schema and About copy lean on “we built sites in the orbit of $uicideboy$ / The Wrecks / Arizona Zervas” as if that were a Wikipedia article, you taught a chat product a portfolio list, not an entity. Portfolio names are fine on a work page. They are not NCORP sources for the studio.

Failure	What it costs	What you do instead
Paid stub deleted	Cash, COI stain, unusable corpus residue	Never start without a four-column source table
Duplicate Wikidata items	Two official websites in the graph	Merge; keep P856 unique
sameAs to a 404 Wikipedia title	You asserted an identity that does not exist	Delete the URL
Relying on Wikipedia for “best of” prompts	Zero movement on recommendation queries	Cluster + PR + extractable tables
Editing the article to remove a lawsuit	Policy violation; the lawsuit is still in sources	Live with independent facts or dispute them off-wiki with evidence
Knowledge panel treated as ChatGPT KPI	Wrong scoreboard	Separate Search identity from answer logs

The pattern I see on client calls is not “we lack Wikipedia.” It is “we have three legal names, two cities, and a schema block that still lists the agency that built the last site.” Fix that before you petition a volunteer project.

How do I measure whether Wikipedia or Wikidata is helping?

You measure answers, not page-creation tickets.

Freeze the prompt panel. Log the engine, the date, whether search/retrieval ran, the named brands, the URLs in the source list, and whether facts matched the entity sheet. A Wikipedia URL in the source list is a useful annotation. It is not a win by itself if the recommendation still names a competitor, or if the description uses the wrong city.

KPI	Source	Pass	Fail
Brand-name prompt fact-match	Manual log	Name, URL, offer, city match the sheet	Hallucinated package or HQ
“Best [job]” inclusion	Same log	You are named for a reason you can defend	Absent, or named with a fake differentiator
Wikipedia URL as source	Screenshot	Present on identity prompts if the article exists	You expected it to close sales prompts
Wikidata / schema consistency	Item + Rich Results	P856 = site url	Two websites, two founding years
Knowledge panel	Google Search	Facts you would sign	Panel exists with the wrong logo / category
Pipeline	CRM	Protectable conversions	“We got a Q-id” as the slide

Cadence:

Cadence	Owner	Artifact
Weekly	SEO / visibility lead	10-prompt subset, identity + one sales job
After any Wikimedia edit	Same	Diff of the item or article + 5-prompt retest two weeks later
Monthly	Marketing lead	Fact-match errors vs entity sheet
Quarterly	Founder	Keep / stop Wikimedia effort; PR vs on-site cluster investment

Example log row (copy the columns; do not invent fill):

Date	Engine	Prompt	Brand status	Wikipedia URL in sources?	Fact-match	Note
2026-08-24	Google AI Mode	“What is [Brand]?”	Mentioned	Yes / no	City + URL match	Identity prompt
2026-08-24	ChatGPT search on	“Best [job] in [region]”	Absent	No	n/a	Recommendation prompt
2026-08-24	Perplexity	“[Brand] vs [competitor]”	Hallucinated offer	No	Fail	Schema still lists old package

Run the same three rows after a Wikidata edit. If identity prompts improve and sales prompts do not, you learned the split. Keep investing in clusters and PR. Stop buying encyclopedia theater.

There is no SLA. Wikimedia reviewers do not run your funnel. Chat products do not publish a “percent of answers from Wikipedia” dashboard you can quote. If someone puts that percentage in a proposal, ask for the method. When they cannot produce one, you already have the answer.

What is worth doing in a week — and when should I skip this?

A week is enough to stop lying to machines. It is not enough to become encyclopedia-notable.

Seven-day order if you insist on touching Wikimedia at all:

1. Day 1 — Write the entity sheet. Diff homepage, About, schema, LinkedIn, Business Profile.
2. Day 2 — Search Wikidata and English Wikipedia for legal name, dba, and founder. Screenshot collisions.
3. Day 3 — Fix P856 and sameAs only where the URL is real. Delete fantasy Wikipedia links.
4. Day 4 — Four-column NCORP table. If it fails, Wikipedia work stops.
5. Day 5 — Frozen prompt log, three engines, identity plus one sales job.
6. Day 6 — One talk-page fact correction if an article already exists and a fact is demonstrably wrong. Disclose if paid.
7. Day 7 — Decide: Wikidata hygiene only, AfC with real sources, or neither. Put it in writing.

If you only have a week	Do	Skip
Entity sheet	Legal name, offer, URL, city, founding date	A Wikipedia campaign
Collision check	Wikidata + Google + LinkedIn name search	Creating a second Q-id “just in case”
Schema	Honest sameAs, no fantasy Wikipedia	12 fake profile URLs
Prompt log	20 prompts, three engines	An “AI authority score”
Talk-page fix	One demonstrably wrong fact, disclosed if paid	Rewriting the article’s lead
PR / cluster	Outline one independent story; ship one table on a money URL	40-page “ultimate guide to Wikipedia SEO”

Skip Wikimedia work for now when:

• Money URLs are noindex or blocked
• The company cannot agree on the legal name or the offer sentence
• You have zero independent sources and a vendor invoice for a Wikipedia page
• The founder wants the article to omit facts that are already in reliable sources
• Nobody will keep the prompt log

A visibility audit is the right hammer when entity collisions, schema, and answer logs disagree and the argument has gone political. DIY the source table and the Q-id check if you have someone who will read NCORP. Lane map: /visibility. The system map is still the AEO playbook.

Good, 90 days later, looks like this — not “we are on Wikipedia”:

• Entity sheet matches fetched HTML, Wikidata, and the answers on brand-name prompts
• Zero invented Wikipedia sameAs URLs
• Recommendation prompts either name you for a reason you can show, or you know you are absent and you are working PR plus clusters instead of encyclopedia theater
• If an article exists, it was earned, disclosed if paid, and you are not secretly editing it

90-day artifact	Pass	Fail
Entity sheet	Matches HTML, Wikidata, brand-name answers	Three names still in the wild
sameAs	Only real URLs	A Wikipedia title you do not have
Prompt log	12 weeks, same prompts	Ad-hoc screenshots after a vendor call
NCORP table	Honest zero or a real two-plus	“We were in a roundup” counted as significant
Recommendation prompts	Named with a reason, or an explicit absent + PR plan	“Wait for Wikipedia” as the strategy

If none of that is true, you optimized a volunteer project you do not control.

FAQ

Does Wikipedia or Wikidata help AI recommend my brand?

Sometimes Wikipedia helps an engine describe you, because encyclopedia prose is widely reused. Wikidata helps when a real Q-id keeps name, URL, and founding facts consistent. Neither is a paid lever that makes models recommend you as the vendor to hire, and there is no honest “X% of ChatGPT answers come from Wikipedia” figure to quote.

How do I measure whether Wikipedia or Wikidata is helping AI recommend my brand?

Freeze 20–40 buyer prompts and log cited, mentioned, absent, and hallucinated, plus fact-match against your entity sheet. Annotate whether a Wikipedia URL appeared as a source. Retest weeks after a Wikimedia change, not the day you submitted a draft. Do not use “article created” as the KPI.

What usually fails first when teams try this?

A paid or promotional Wikipedia draft that cannot pass NCORP, or a sameAs URL that points at a Wikipedia title or Wikidata item that is not the company. The next failure is treating a knowledge panel or a Q-id as a recommendation win while “best [job]” prompts still name competitors.

How long does this take to show results?

Identity hygiene (schema, P856, killing collisions) can show up in the next crawl and the next search-enabled chat run — days to a few weeks. A Wikipedia article, if it is even possible, is a volunteer review process with no SLA. Recommendation prompts move when independent sources and citeable pages exist, which is a PR-and-cluster timeline, not a page-creation ticket.

What should I skip if I only have a week?

Skip Wikipedia vendors, AfC drafts without a four-column source table, and any sameAs Wikipedia URL you do not have. Spend the week on the entity sheet, a Wikidata collision check, honest Organization markup, a 20-prompt log, and one extractable table on a money URL.

When is this not worth doing yet?

When you cannot agree on the legal name or offer, when money URLs are not indexable, or when the plan is “buy a Wikipedia page” with no independent sources. Fix facts and eligibility first. Wikimedia will not rescue a site that still contradicts itself.

CTA

If a vendor sold you Wikipedia as an AI recommendation switch, stop — run NCORP, clean the Q-id you actually have, and log whether engines name you.

Lane: /visibility · Book a visibility audit.