How to Allocate Budget Across Trust Assets

A DTC pet-hydration brand outside Rotterdam opens its spring budget review. Last year’s biggest line was 40 creator videos; the founder wants to double it, the growth lead wants more reviews. Neither side answers the question that decides both: when a buyer in a warm US apartment asks an assistant which fountain stays mold-free with a weekly filter change, which sentence from those videos would the model quote? The videos say the cats love it.

A trust budget should be split by whether each line leaves behind a checkable sentence at the moment of decision — impressions buy memory, evidence buys the recommendation.

Impressions buy memory; evidence buys citations

A model comparing brands collects verifiable statements: who tested what, under which conditions, with which value, and where a stranger can check it. View counts, follower totals, and launch-party photos fit none of that shape.

Three habits look defensible in a weekly report: channel inertia (creator shorts worked, so double them), unit price (cheapest CPM wins), and volume (press releases and awards feel like authority). Each produces activity, none a quotable sentence.

The filter is one question: can a third party re-check this? A lab report naming standard and batch can; a roundup with published inclusion rules can; a clip that says “easy to clean” cannot, however many people watched it.

Inventory the sentences, not the spend

An inventory lists what you kept, not what you did. For every trust line from last year, write the sentence it left behind that stands alone.

Asset	Ask	Counts when	Usually fake
Third-party review	Who tested, under what conditions	Comparison, method, linkable	“Premium experience”
Category roundup	How products were selected	Inclusion rules public	Paid, unlabeled placement
Long-term test	How long, what failed	Named faults, timeline	Unboxing praise
Reviews and Q&A	Are the bad ones answered	Public reply, traceable	Deleted one-stars
Certification	Which body, which standard version	Report downloadable	Badge with no issuer
Bylined expertise	Who wrote it, why them	Author verifiable, relevant background	Brand describing itself

Two brands can spend the same on “reviews” and land in different asset classes: one has “ran 30 days at 68°F, filter changed weekly, method published,” the other “customers love it.” Only the first can be quoted inside a comparison.

Score each row, then fund the gap

Score every row 0–2 on three dimensions, six points total. Quotability: does a sentence survive isolation? Decision proximity: does it surface when the buyer asks “which one,” or only at brand level? Independence: is anyone but you saying it?

Bands decide the money. 0–2: pause the line or rewrite its conditions. 3–4: keep it, replacing adjectives with method, sample size, and numbers. 5–6: add budget until a saturation signal appears.

The score is a sorting tool, not a forecast. It asks the mirror question of the four trust prerequisites for AI recommendation: that piece names the missing prerequisite, this one places the next dollar. Disagreements get settled against live AI answers, not seniority.

Saturation arrives early. The sixth review restating the first five adds noise, not coverage. Three signals: new assets paraphrase each other; independent domains citing you stop rising; everything lands on the same sites. Gap signals are as clear: the model can only cite your own pages for your target questions, or it cites a review of a model you stopped selling. Money works hardest where it fills an absence.

Then set a stop-line per asset class with an observable trigger — two quarters with zero new independent domains — and a fixed cut, like 30% to the gap. A stop-line ends the argument before the meeting.

Example: one quarter, two shapes

Example — brand, amounts, and outputs are invented; only the logic is real.

A $60,000 quarterly trust budget, split two ways.

Plan A puts $52,000 into 35 creator videos and $8,000 into product news. Output: clips calling the fountain easy to clean, reposts, no test conditions, nothing checkable.

Plan B puts $16,000 into three specialist outlets running one shared protocol (same water, same temperature, scheduled filter changes, published method); $12,000 into four long-term testers logging failures over 60 days; $10,000 into answering repeated complaints in public reviews; $8,000 into publishing the lab report as an HTML page; $10,000 into an engineer-bylined piece on pump and filter; $4,000 held back.

Same order of magnitude, opposite asset shapes. Plan A supports one sentence: popular with cat owners. Every line of Plan B can be re-checked and dropped into a comparison — if the sentences reach a page written as an answer unit. Evidence that lives only in a PDF may never be used.

The five-step quarterly playbook

Step	Action	Owner	Passes when
1 Inventory	List trust spend by source; write the sentence each produced	Marketing, content	Every line has a sentence or an explicit “none”
2 Score	Three dimensions, 0–6, tie-broken by live AI answers	Growth lead	Disagreements are settled by transcripts, not seniority
3 Split pots	70% base, 20% offense, 10% experiments	Budget owner	Each pot has named projects, not a ratio
4 Stop-lines	One observable trigger per asset class	Marketing	The trigger can be counted, not felt
5 Rebalance	Move 20–30% per quarter, re-sample the same prompts	Growth, finance	Every move maps to a citation change

The base pot holds what AI asks about any week: reviews, certifications, evergreen tests. Offense funds third-party testing in a new market or category. Experiments try new source shapes, turning recurring buyer questions into structured answers. Move only 20–30% a quarter; full reallocation destroys next quarter’s comparison.

Where this breaks

First, the entity is unresolved: the name is spelled three ways and off-site mentions are near zero. Every euro goes to being identifiable first — see whether your brand entity exists.

Second, revenue sits inside a marketplace, where AI reads listings, reviews, and fulfillment. Off-site reviews are discounted; fund what moves in-platform metrics.

Third, the category is low-ticket and impulse-driven: buyers do not compare evidence, so the table overprices reviews.

Fourth, you are in correction mode: when one wrong claim is spreading off-site, fixing it outranks every allocation rule here.

Fifth, the score becomes a financial promise. It ranks options; it predicts no return. Read it against citations sampled as an AI visibility baseline prescribes, and against orders.

This month, spread last quarter’s trust spend into one table and add the quotable claim each row produced. Rows you cannot fill are the money to stop or re-condition first — pick one, add test conditions, sample size, and a linkable source, and spend again. The rest of the lane is on the strategy hub.