Non-brand US Search — high-spend keywords by keyword × device, last 12 months + search-term drift

Why keyword × device (not campaign): the campaign/ad-group structure has turned over in the last
year (the v2 exact set replaced older broad campaigns), so a campaign-level year-over-year view is
apples-to-oranges. The keyword text + device combination is the stable unit — it follows the demand
even as it moves between campaigns. Judge on L2 subs / cost-per-sub (Subscription - L2 Purchase via
all_conversions). Window: 2025-07-13 → 2026-07-12. "Now" = last 4 wks (6/15–7/12). Pulled 2026-07-12.

A keyword text+device often ran in several campaigns/ad groups over the year; figures below are summed
across all of them (that's the point). Rep.conv is the inflated Conversions column — directional only.

🔑 Headline: the biggest keywords of the year are GONE

The account's top-spending keywords over the last 12 months are no longer running. The current set
(see the per-campaign files) is a different, more granular exact list. What changed:

~$120K+ of the last year's spend was on keywords that are now switched off. Any historical
"what works" read from old campaign names is stale — use the keyword×device table below.

High-spend keywords × device — last 12 months

Ordered by total 1-yr spend (both devices). Cost/sub = spend ÷ L2 subs for that keyword+device.

Device read across the year: for almost every keyword, Mobile subs cheaper than Desktop
(brain training $139 vs $207; brain games $141 vs $194; free brain training $143 vs $202). Desktop buys
more reported conversions (signup/checkout noise) but fewer real subs per dollar. The exceptions are the
"for adults" / "memory games" game-intent terms, where Desktop is competitive or better.

Search-term drift — then (1yr) vs now (last 4wk)

The retired broad-ish keywords (brain training, free brain training, brain exercises) were broad
enough to spray onto competitor brands and their feature names. That's most of why they were expensive.

🔴 THEN — competitor / rival-feature leakage (1-yr, approx. spend on these queries)

Much of this converted poorly (e.g. double decision* ≈ $2.6K → ~8 subs ≈ $325/sub; staying sharp ≈
$2.3K → ~$135/sub but mostly 0-sub variants). The account was effectively paying to show against BrainHQ,
CogniFit, Elevate and AARP's Staying Sharp on unmanaged broad match.

🟢 NOW — the leakage is much smaller, but the same class persists (last 4wk)

Retiring the broad brain training/free brain training keywords removed the bulk of competitor leakage.
The current set matches far more on-intent (brain games for adults, brain games, memory games for adults). But the surviving broad terms still leak — and negatives still are not applied:

Same story as a year ago (competitor + test intent, 0 subs) — just ~$150/4wk instead of ~$8–10K/yr,
because the broad keywords that caused most of it were switched off. The remaining broad keyword
brain improvement exercises is the current leak source.

On-intent core — then vs now (the queries that actually sub)

The self-query is the profitable core in every case; the drift is the tail. On the broad brain exercises free, the competitor tail simply rotated (staying-sharp/brainhq a year ago → cognifit now).

Actions

1. Add competitor negatives at the account/campaign level — brainhq, brain hq, posit science,
   cognifit, cognitive fit, elevate, double decision, staying sharp, stayingsharp, brainiq,
   luminocity. This was worth ~$8–10K/yr at peak and is still leaking today via the broad keywords.
2. Add test/method negatives — cognitive test, memory test, iq test, iq games, dual n back,
   neuroplasticity, speed of processing — historically high-spend, ~0 sub.
3. Mine the retired winners for re-activation — some retired keywords subbed cheaper than the current
   set: brain training Mobile $139/sub, free brain training Mobile $143, brain training apps
   Mobile $142. If they were cut for competitor leakage (not intent), re-add them as exact/phrase with
   the competitor negatives above rather than leaving the demand on the table.
4. Device: favor Mobile on the training/exercise terms ($139–144 vs $176–207 Desktop); keep Desktop on
   brain games for adults / memory games where it's competitive.
5. The one live broad keyword, brain improvement exercises, is the current leak source — tighten it
   (exact/phrase) or wrap it in the negatives above.

Created 2026-07-12. Method & judging: ../../conversion-events-reference.md. Companion: the per-campaign
files in this folder (current structure) — this doc is the cross-campaign, campaign-agnostic year view.