GEO-130·Issue 11·Authority & Off-site·130 / 132
The Long Game: Why GEO Isn't a Campaign
Why AI visibility decays once a campaign stops, which assets expire fastest, and the five-step maintenance routine that keeps a brand inside the answer.
Issue focusScale It: SOPs, Organization, and the 2027 Roadmap
Consider this case. Last October a DTC pet-fountain brand ran a campaign: twenty review placements, six creator videos, one sponsored slot in a niche publication. Within a month, assistants in English and German were naming its model. In January the budget moved on and the content team was reassigned; nothing was deleted, nothing updated.
This August the same questions went through the same assistants. The brand still turned up, but was no longer recommended: both top slots cited a March guide, and neither brand in it was theirs.
Visibility compounds like an asset, not a spend: going dark slides a brand from recommended to mentioned to absent — a slope, not a cliff, so you have time to catch it.
Decay is a slope, not a switch
A model does not remember your campaign. It rebuilds an answer from what retrieval reaches this minute, so stopping is not one event.
The page stays crawlable; brands rarely vanish overnight. Then the sentence goes stale: price, model, stock, warranty. Once those stop matching the present, a model prefers a page that agrees with it, usually a retailer’s. The retrieval pool is finite, so a rival’s fresher page takes the slot.
The curve moves slowly, then quickly, then flat. The flat stretch is dangerous: still findable, still mentioned, no longer used in a comparison. A mention is an index entry; a recommendation is a decision the model will defend with a sentence. Only one sells.
A quieter version: ten new posts a month while the product page has not been repriced in three years. Publishing velocity and citable inventory are separate curves.
Three asset classes, three depreciation rates
GEO inventory is not one thing; schedule maintenance by layer, not by article.
| Asset | What drives decay | What decay looks like | Maintenance |
|---|---|---|---|
| Crawlable pages: URLs, schema, entity facts | Replatform, template swap, firewall rules | Fetched but unparsed; fields gone | Build check; regression on release |
| Citable facts: price, model, stock, warranty | Master-data change, discontinuation, repricing | Page alive, sentence wrong | Fires on master-data change |
| Quoted sentences and third-party corpus | Rival launches, outlet redesigns, list churn | Citation drops to a source list, then goes | Quarterly re-test; supply material |
Costs are inverted: facts expire fastest and cost least; the corpus is hardest to rebuild. Most teams reverse it: they buy articles while a promo price stays live for three quarters — yet an afternoon of syncing can recover a citation no new article touches. The corpus needs a monthly record: monitoring citations in third-party reviews is that ledger.
Put the re-check on the calendar
“Review it when we have time” is how maintenance dies. Four triggers and one owner are enough.
| Trigger | Fires when | Owner | Done when |
|---|---|---|---|
| Fact change | Price, model, stock, certification, discontinuation | Operations opens, content edits | Page equals master data |
| Page change | Template release, CDN or firewall policy | Engineering | Fields read with scripts off |
| Corpus change | Outlet redesign, list refresh, rival launch | Brand, content | Quoted sentence re-logged |
| Scheduled re-test | One week a quarter, same prompts | Growth, measurement | Movement attributed, not screenshotted |
Three pieces are routinely missing: an owner who can move master data and a front-end ticket, a ledger of URL, quoted sentence, and review date, and a stop line written before the money goes in. Maintenance that dies with a person is not maintenance.
Keep the re-test on the scale you already report, defined in four layers of AI visibility.
Five steps from campaign to standing program
No new headcount; the work is routing maintenance through process.
| Step | Action | Deliverable | Passes when |
|---|---|---|---|
| 1 Build the ledger | Every cited page, quoted sentence, next review date | Asset register | Every URL has an owner and a date |
| 2 Wire the triggers | Four triggers into existing workflows | Trigger list | Someone responds within a week |
| 3 Fix the cadence | Frozen prompts, fixed weekday, quarterly | Re-test record | Two rounds agree before it is a trend |
| 4 Separate the budget | Maintenance stops competing with new content | Budget line | A dedicated annual amount |
| 5 Write the handover | Ledger, triggers, definitions in one document | Handover pack | A new owner still re-tests after a month |
Step 4 gets merged away: new content produces artifacts, maintenance produces a quiet “nothing broke,” so in one pool the loss appears two quarters later. The 90-day roadmap sign-offs are a start, not a finish, and every new page clears the same GEO QA checklist.
Example: nine months dark
Example — brand, outlet, and figures are invented.
An espresso brand ran a campaign in October 2025: eighteen pages, five off-site reviews, two reliably cited models. From January 2026 a part-timer kept the site alive and nothing else. In August, thirty questions returned mentions in eleven answers, recommendations in two.
The causes were not writing quality. The page showed a promo price while master data had reverted to list, so the model cited a retailer page that agreed with the present. The review’s spec table had become an image, pushing the citation into a trailing source list. A rival published a “small kitchens, single-serve” guide in March, a scenario this brand never tested.
Three repairs: sync the price, restore the table as text, ship a sample for the scenario. The citation returned; the scenario stayed unanswered. The nine months skipped were repaid by one repair round and two re-tests — plus two quarters of recommendation slots.
Where this breaks
First, the market is leaving: maintenance is not life support for a dead line, so write the stop line first.
Second, no owner: an unowned “long term” becomes a long time ignored.
Third, revenue lives in a marketplace: listings, reviews, and fulfillment come first, the corpus second.
Fourth, a regulated category: certifications and labels move faster than quarterly; run on the compliance clock.
Fifth, compounding treated as a promise: without a positive return, long term becomes long-term waste.
This line sits under authority and off-site, between making evidence exist and turning citations into orders.
This month, one action: open the ledger and fill three columns — URL, quoted sentence, next review date. In ninety days you will see your decay curve rather than guess at it.