The campaign that sold nothing.

Controls on the same instruments: homepage GA4 same-session 3.4%; homepage GA4 user-level 2.06%; homepage Google-click-only sessions in the raw tracker 6.1% reach a thank-you page (97 of 1,594). The session-break explanation is dead — the user-level row follows people across sessions.

That is the signature of YouTube and in-app video/display taps — which is what MN said it bought. GA4, which needs the page to render and the visitor to consent, saw 0.26% of these sessions against 1.1% of homepage sessions: the campaign traffic bounces before the page even finishes loading, four times more often than ordinary traffic.

The page opens on “TAKE BACK THE INTERNET.” over a dark hero, then runs The Restriction Feed — a dated news stream of censorship and surveillance headlines, updating through today — followed by a six-panel editorial essay (identity gating · blocking by default · surveillance by design · platform enforcement · infrastructure control · anonymity removal), each panel structured as the pitch, the drift and the reality check.

Between the feed and the essay sits the only commercial ask on the page: “Be part of the resistance, quietly”, three product lines, and one button — Get Mysterium VPN → /vpn-pricing. The link is correctly targeted. It is one of thirty links on a page whose dominant content is journalism.

Open https://www.mysteriumvpn.com/take-back-the-internet →

The transactions column populates whenever a conversion happens — the homepage proves the mechanism with 11,751 non-null rows out of 201,629. The campaign page has zero non-null rows out of 17,470. The instrument works; the page produced nothing for it to record.

Roughly two-thirds of all Google Ads traffic in 2026 — 426,435 sessions — was sent to a page that has never produced a sale, while the third sent elsewhere converts at 1.4%. That is a page-level result, and it rules out “the ads bought the wrong people” as the sole explanation.

Sessions here move toward pricing 108× less often than elsewhere. Arithmetically right; as a statement about readers, misleading — the denominator is ad taps (see Verified 09-03). Most of these sessions never became a rendered page. The ratio measures the traffic type, not the page.

The people who arrive want to change their TikTok region. The NFL / streaming instinct in the August marketing deck is the evidenced one, and the site's own top organic pages already prove it. The September target-audience document points away from the customer the data shows.

Calibrated to MN's real payroll (public Sodra registry, fetched 08-16): MN Intelligence UAB — the consumer-VPN entity where marketing sits — pays €3,107 average gross to 13 staff (MN Technologijos, engineering, pays €6,365 to 12). With ~1.8% employer contributions over a ~166 h month that is ≈€19/h, so the campaign consumed ≈€63k–€101k of salary time (band €53k–€152k across the monthly spread and the chance some designers sit in the engineering entity). The LT ring's entire payroll is ≈€113k/month, so this programme spent roughly 0.6–0.9 months of whole-company payroll. Three of the original team left during the campaign (partnerships 02-24, SEO ~03, organic lead ~05).

Not in the family, listed so nothing is missed: Black Friday 2025, the Valnet affiliate network (Aug), Fatjoe reviews (MystNodes), Omnisend retention mails, DeleteMe / WhiteBridge partnerships.

Two-thirds of MN's 2026 Google clicks went to the page that returned nothing. The third that went to offer pages returned $63k. The tracking works, Google Ads works, and MN knows how to run a converting campaign — the variable is the destination and the message, nothing else.

The purchase stream even names the winners: "search — Brand — BE/CA/FI/DE/IT/NL/CH/NZ" (58 purchases), "search — Brand — FR" (39), "search — Brand — US" (22), "PMax — US (ROAS) web purchase" (15 across variants). Brand search and PMax sell. Awareness does not.

Read the bottom rows twice: on offer pages — pages whose whole job is to sell — Meta bought 45,396 sessions for 25 purchases and Reddit 50,104 for 2. Paid social fails on the selling pages too, not only on the awareness page. Note what this does not say: the take-back campaign itself was 92.2% Google Ads (426,435 of 462,535 paid sessions), 7.6% Meta, 0.2% Reddit and X. It was a Google buy with a small Meta tail — never a social campaign.

The session tracker sees 4,424 paid-click purchases in the same window. GA4 names a campaign for 176. ~96% of the paid book is invisible in the surface where campaign names live — so the take-back zero could not have been caught from the reporting in use: in that view the entire paid business looks like a couple of hundred sales a year.

Of the 176 that are named, 129 — 73% — are Brand search ("search – Brand – US / UK / FR / BE-CA-FI-DE-IT-NL-CH-NZ"): people typing Mysterium VPN into Google, i.e. demand the brand already created. The rest is generic VPN search and PMax. Not one named campaign is an awareness or persona campaign.

€0.10–0.50 is the ordinary band for YouTube / display / in-app inventory, and it brackets MN's own single disclosed target (Meta awareness at ~€2.50 CPM lands at €0.25–0.50 per click at 0.5–1% CTR). So the Google buy plausibly ran €40k–215k, central ≈€105k. For contrast the offer-page buy, with its expensive search slice, brackets to €120k–350k and returned $62,996 — on these benchmarks both halves of the account lose money at first purchase; take-back is simply the half that returned nothing at all. Every figure in this paragraph is an estimate and dies the moment the Ads cost column is read.

SELECT
  CASE WHEN landing_page = 'https://www.mysteriumvpn.com/take-back-the-internet'
         THEN 'take-back campaign'
       WHEN landing_page = 'https://www.mysteriumvpn.com/'
         THEN 'homepage (control)'
       ELSE 'all other pages' END               AS page_class,
  COUNT(*)                                      AS rows_,
  SUM(no_sessions)                              AS sessions,
  COUNTIF(transactions IS NULL)                 AS rows_tx_null,
  COUNTIF(transactions IS NOT NULL)             AS rows_tx_notnull,
  COUNTIF(transactions > 0)                     AS rows_tx_positive,
  COALESCE(SUM(transactions), 0)                AS tx_sum,
  ROUND(COALESCE(SUM(revenue), 0), 2)           AS rev_sum
FROM `mysterium-bq.claude.vpn_web_sessions`
WHERE session_start_date >= '2025-12-01'
GROUP BY 1

Frozen evidence page, dated 3 September 2026. Every figure was read from the named result files at build time. A later measurement supersedes this page with a newer dated one; it does not edit it.