Every industry has vendors who inflate their portfolios, and most industries give you no way to check. Design case studies are screenshots you can't trace; SEO "results" are charts with the baseline cropped; references are hand-picked. Verification is expensive, so buyers skip it and hope.
Wikipedia is the one exception, and almost nobody uses it. Every edit ever made is public, timestamped, attributed to an account, and kept forever. When an agency says "we created that page," the claim is checkable down to the minute — free, no permission needed.
This guide turns that fact into a procedure: the Portfolio Autopsy — five checks, three minutes each, fifteen minutes per vendor. It assumes you already have a shortlist. If you're earlier in the process, start with the nine warning signs of a fake agency and the vendor scorecard guide, then come back here.
Key takeaways
- Wikipedia is the only vendor market where the entire work history is public: every edit, deletion, block, and disclosure — timestamped and free to audit.
- The Portfolio Autopsy is five checks, three minutes each: existence and deletion log, authorship percentages, the editing account, the disclosure trail, and the sockpuppet and COI archives.
- XTools shows which account wrote what percentage of a page's live text. "We created that page" is a number, not a story.
- Disclosure is binary: if no account connected to the agency is disclosed on the pages it claims, either the work isn't theirs or it broke the Terms of Use. Both end the evaluation.
- Enforcement archives outlive rebrands: the Wiki-PR case put 250+ blocked accounts on the permanent record in 2013, and the archive is still the first thing a diligent search finds.
The asymmetry nobody exploits
Wikipedia's revision history is a public ledger. Each article carries a complete page history listing every change, who made it, and when. Deleted articles leave a log naming the administrator and the policy reason. Every account's contribution record, creation date, and block history is one click away. Because paid editors must disclose employer and client on-wiki, even the commercial relationships are part of the record.
A Wikipedia portfolio is, as a result, the most verifiable claim in all of marketing. It is also among the least verified. Buyers who would never accept an unaudited financial statement accept "we've published 500+ pages" on a sales call, because nobody told them the audit is free and takes a quarter of an hour. Agencies know this; the dishonest ones price it in. The five checks below close that gap.
The Portfolio Autopsy: five checks, three minutes each
Ask the agency for its three best examples — pages it says it created, with the article titles. (How they respond is itself data.) Then run each claimed page, and the agency name, through five checks:
- Existence and survival. Does the page exist, and what does its deletion log say?
- Authorship. Which account actually wrote the live text, and in what proportion?
- The account. Is the claimed account real, established, active, and unblocked?
- Disclosure. Was paid editing disclosed where the rules require it?
- The rap sheet. Does the agency's name appear in the sockpuppet or conflict-of-interest archives?
Each check produces a hard signal, not an impression. Together they answer the question that matters at final shortlist: is the track record real, and was it built in a way that won't detonate under your brand later.
Check 1 — existence and survival: read the deletion log
Sixty seconds per claimed page. Type the exact article title into Wikipedia's search box.
If the article exists, open "View history," sort to the oldest revision, and note the creation date and creating account. A page "created" two years before the agency existed is a borrowed trophy.
If the article does not exist, Wikipedia tells you why. Visiting a deleted page's URL automatically surfaces its deletion log: date, administrator, and policy reason — most often G11, unambiguous advertising, and sometimes G5, created by a banned or blocked user, meaning the page was written by an account that had no right to edit at all.
Live page, plausible creation date: continue. Deleted as advertising: the portfolio piece failed Wikipedia's most basic test. Deleted as G5 or in a sockpuppet cleanup: stop reading the proposal.
Check 2 — authorship: who actually wrote the text
The most common portfolio inflation is not inventing pages — it is claiming pages someone else wrote. Authorship attribution is the antidote.
XTools, the standard analysis suite for Wikimedia projects (xtools.wmcloud.org), has two relevant reports. Page History summarizes an article's life: creator, creation date, top contributors by edits and by text added. Authorship goes further — using WikiWho token-based attribution, it calculates what percentage of the article's current live text each account wrote. Not who edited most often: whose words survive on the page today.
Paste in the article title and read the table. If the agency wrote the page, its account should sit at or near the top of the authorship list — community expansion dilutes the share over time, and that is fine. What you should not see is the account missing entirely, or at a fraction of a percent for fixing a comma in 2023.
The first run takes three minutes; every page after, thirty seconds. The check converts the industry's squishiest claim — "we wrote it" — into a percentage.
Check 3 — the account: professional record or burner
By now you have account names — from page histories, authorship tables, or because the agency named them. Each gets the inspection an employer would give a résumé: open the user page, then the contributions. Three things matter:
- Age and volume. When was the account registered, and how many edits does it have? An agency claiming a decade of experience whose accounts are eight months old has a gap to explain — Wikipedia experience is account-bound.
- The shape of the work. Scroll the contributions. A credible professional account shows the texture of real editorial life: drafts through review, talk-page discussions, edits across many topics. A burner shows a thin strip of promotional-adjacent pages and nothing else.
- The block log. Open the account's logs from the contributions page. A block entry — particularly for undisclosed paid editing or sockpuppetry — is the loudest signal in the entire procedure. Active, unblocked, multi-year accounts pass; blocked, abandoned, or freshly minted replacements fail.
One honest caveat: an editor's personal account may legitimately predate their employment — a fine answer, when the agency names the account and lets you check. "Our editors prefer to stay anonymous" is not a privacy stance; it is a refusal to be checked.
Check 4 — the disclosure trail
Wikipedia's Terms of Use require paid editors to disclose employer, client, and affiliation, and the WP:PAID framework specifies where: the editor's user page, the affected article's talk page, or the edit summary. Not etiquette — the line between legitimate practice and a policy violation.
So look in those three places: the account's user page for a paid-editing statement naming clients, the article's talk page for a connected-contributor notice, the page history's edit summaries.
Then apply the binary. Disclosure present: an agency that works in the open, the way this work survives. No connected account disclosed anywhere on a page they claim: either the page is not their work, or it was produced in breach of the Terms of Use. One is portfolio fraud, the other a compliance time bomb attached to your brand. You do not need to determine which.
Absence of a disclosure trail on "their" pages tells you everything, which is why this is the check to run if you only run one. Our own approach to disclosed, policy-compliant work is described on the Wikipedia page creation service page.
Check 5 — the rap sheet: SPI and COI archives
Wikipedia's enforcement history is public and searchable, and it is where agency reputations go to be preserved against their will.
Two archives matter. Sockpuppet investigations (SPI) document coordinated multi-account abuse: accounts involved, evidence, blocks issued. The conflict-of-interest noticeboard (COIN) archives years of community discussions on specific firms and undisclosed-paid-editing operations. Search both: type the agency name into Wikipedia's search box together with prefix:Wikipedia:Sockpuppet investigations, then again with prefix:Wikipedia:Conflict of interest/Noticeboard. Add former brand names — archives outlive rebrands, and rebranding after a ban is a known pattern.
The precedent that proves the archives' reach: the Wiki-PR scandal, in which the community linked a single firm to a network of undisclosed paid-editing accounts. More than 250 accounts were blocked or banned in 2013, the Wikimedia Foundation sent a cease-and-desist, and the case file is still public twelve years later. Every client whose page went through that operation inherited the cleanup.
A hit in these archives is not automatically fatal — read the case, check the dates, see whether the firm acknowledged it and changed practice. A hit the agency failed to mention when asked directly, however, is.
Fraud patterns and the checks that bust them
The same portfolio claims appear in almost every inflated proposal. Here is how each is faked, and which three-minute check exposes it.
| Portfolio claim | How it's faked | The 3-minute check | Verdict signal |
|---|---|---|---|
| "We created [recognizable company]'s page" | Credit claimed for a page written by uninvolved volunteers, or one that predates the agency | Check 2: is their account a top author of the live text? Check 1: creation date vs. agency founding | Account absent from the authorship table = borrowed trophy |
| "500+ pages published" | Round number counting deleted pages, minor edits, or nothing at all; no list provided | Ask for ten URLs from the last twelve months; run Check 1 on three | A real list arrives in minutes; a lecture about confidentiality arrives when the number is decorative |
| Logo wall of famous clients | Logos imply authorship; the agency may have made one edit on the page, or none | Pick two logos and run Checks 2 and 4 on those articles | Logos are free to display; authorship percentages and disclosures are not |
| "Senior Wikipedia editors on staff" | Unnamed "editors" with no linked accounts, or accounts with purchased edit history | Check 3: account age, contribution shape, block log; Check 4: whose disclosures name the agency? | An editor who cannot be named is an editor who cannot be checked |
| "Our pages never get deleted" | Survivorship framing — the deleted work is invisible unless you go look | Checks 1 and 5: deletion logs on claimed pages, plus the SPI/COIN search | Everyone loses pages sometimes; honest vendors quote a survival rate and define how they count it |
Every row shares one dependency — a buyer who never looks. None of these fakes survive contact with the ledger.
What honest agencies show instead
Knowing what fraud looks like is half the brief; the other half is recognizing legitimate evidence. Honest vendors face a real constraint: some clients do not want to be named, and a serious firm will not burn a confidentiality agreement to win your deal. What a credible portfolio looks like in practice:
- Disclosed accounts, offered without friction. The agency names at least one editing account whose user page carries the paid-editing disclosure, and invites you to read the contribution history — the strongest single trust signal in this market.
- Process artifacts over logo walls. Notability assessments, source maps, draft excerpts, AfC correspondence — redacted where needed. Work product is hard to fake; logos are trivial.
- Survival-rate framing with definitions. Not "guaranteed" anything — "of the pages we published, this share is live today, counted this way." We publish ours (a 93% success rate, and an 80% refund if a page cannot be defended after three attempts within the 90-day monitoring window) because a number with a definition can be interrogated and a slogan cannot.
- References you can actually call, plus named client outcomes where permission exists — ours are on the clients page.
- A straight answer about deletions. Every long-running practice has lost pages. The honest version names a case and what changed afterward.
"We can't show client pages" is therefore half-legitimate — but only half, because confidentiality never prevents naming a disclosed account, showing process artifacts, or pointing at archives that come up clean. For how the major providers compare on this dimension, see our Wikipedia agency comparison.
The follow-up email that settles it
After running the Autopsy, send the shortlist one email with four questions. Each answer is verifiable on-wiki, so each also tests whether the vendor knows you can check.
- "Which disclosed account or accounts would handle our work? Link the user page with the paid-editing disclosure." Verifiable in Check 4. The only acceptable answer contains a username.
- "Link three articles from the last twelve months where that account is a top-three author of the live text." Verifiable in Check 2, via XTools, in ninety seconds per link.
- "Which of your pages were deleted in the last two years, and what did the deletion log say?" "None" from a high-volume practice is its own red flag — and you can pull the logs yourself, so this really tests honesty.
- "Has your firm, under any current or former name, appeared at SPI or the COI noticeboard — and if so, what changed afterward?" Verifiable in Check 5 — you may already know the answer before you ask. That is the point.
The kicker, which is the whole article in one line: any agency that will not name a single account after you've read this far has answered your question. The records are public, the tools free, the checks fifteen minutes — the only vendors disadvantaged by that arithmetic are those whose portfolios cannot survive it.
Run the Autopsy on us. Our client outcomes are listed; contact us and ask for the links — accounts, disclosures, authorship tables — and we will hand them over before any proposal. That offer is the test every vendor should have to pass.
Download: PDFthe vendor scorecard (PDF) — Autopsy results feed the scorecard: log what each check returned, then score up to three vendors side by side.