Prioritizing Off-Site Investment: Media, Communities, Creators, Directories

Three proposals land on one table in August. A PR agency wants a three-wave newswire push for a pet-tech brand selling automatic litter boxes in the US and Germany; a creator agency offers twenty unboxing videos; the community lead has a six-month Reddit and Quora calendar. The quotes differ by more than double, and the room debates which channel is “more premium.” Nobody spends the cheapest two minutes available: open an assistant, type “best automatic litter box for two cats,” and read who the answer actually cites.

Order off-site spend by credibility, then citability per dollar, not by how premium a channel sounds — credibility is a gate you cannot buy past, and above the gate you are paying for one specific fact a model can lift.

Off-site spend buys a sentence, not an impression

A synthesis step does not read coverage for tone. It keeps spans that survive lifting out of the page: a subject, a condition, a checkable value.

“Committed to premium pet care” proves the brand exists. “With two cats and four scoops a day, the bin empties every three days, and it runs at about 52 dB under load (Example)” drops into an answer intact. The second sentence is what the money should buy; the first is packaging. If your product pages cannot produce a sentence like that, off-site spend has nothing to amplify — fix the answer unit first.

The four node types produce different evidence, and price barely predicts citability.

Node	What it produces	Where an assistant uses it	Common misuse
Media	Proof you exist, a point of view, original data	“According to…”, entity disambiguation	Launch notices with no value
Communities	Usage context, drawbacks, long-term experience	“Users report…”, “the downside is…”	Paid posts that read as promotion
Creators	Test conditions, measured specs, comparisons	“In testing…”, “measured at…”	Voiceover with no claim to lift
Directories and profiles	Field-level facts: name, category, location	Entity stitching, sameAs checks	Volume over consistency

Media is the most expensive and buys mostly existence; communities cost little and produce the quotable sentences. Sorted by rate card, the order is wrong almost every time.

A reverse example: a brand puts most of its annual off-site budget into two trade outlets, every release saying “industry-leading.” Six months later, ask an assistant how to choose in that category. It cites a forum price thread and an independent roundup, and lists the brand under “other options.” The money worked — the entity stands — but existence was read as recommendation.

Score credibility as a gate, citability as the return

Score each candidate 1–5 on three axes; higher is better.

Axis	Question it answers	1	5
Credibility	Is the source independent to buyers and models?	Paid placement, owned network	Editorial review, user history
Citability	Can you lift a sentence with subject, condition, value?	Slogans and logos	Measured numbers, test conditions, drawbacks
Cost	Cash, lead time, maintenance	Six figures, a quarter to launch	Four figures, live in weeks

Three rules, in order.

First, do not build a weighted total. Credibility below 3 is a no: a cited low-trust source writes a wrong fact into the model’s picture of you, and repairing that costs more than the placement.

Second, above the gate, rank citability against cost. That column is what the budget purchases: one liftable fact, not one impression.

Third, use credibility only to break ties between two sources with the same evidence type.

Map who already speaks for you first with the off-site trust network; directories and profile fields can wait.

Run four stages, and do not skip to stage three

The symptom decides where the first dollar goes.

Stage and owner	Symptom	Spend first	Spend later	Passes when
1 Entity unresolved — growth	Your brand name returns no clear answer	Directories, one verifiable media mention	Creators, communities	Name, category, HQ match on three sources
2 Missing from category answers — content	You never appear in category questions	Community participation, creator tests	Media	Same facts on three independent sources
3 On the shortlist — content	Listed under “other options”	Comparison framing, original data, rankings	Directories	A “versus” answer uses your comparison
4 Defending — ops	Cited prices and models are stale	Correcting old pages	New exposure	Stale figures stop being cited

The expensive misorder is stage one buying stage three: a ranking placement bought before the entity is legible, while assistants still guess what you do.

Split it (Example): 10–15% on the gate, 50–60% on citability, 25–30% on authority, 5–10% on retesting. In stage one the gate can take a third.

Example: one quarter’s budget, reordered

Example — the brand, budget, and scores are invented; no real campaign is described.

A pet-tech brand holds $90,000 for the quarter. Before: media 60%, creators 30%, directories 10%, communities 0. Twenty buyer prompts are re-run. The most-cited sources are two forum threads and an independent roundup; the site appears only on brand-name prompts, and no “best of” list carries it.

Node	Credibility	Citability	Cost	Action
Communities	4	5	4	Increase; participation only
Creator tests	3	4	3	Increase; specs and drawbacks in the brief
Media	5	2	1	Halve; keep only pitches with original data
Directories	4	2	5	Patch inconsistent fields only

After: media drops to 25%, and the savings go to creator tests plus six months of participation. Each brief names one fact set — cats supported, noise under load, monthly consumable cost — and directories patch only disagreeing fields.

Two numbers decide renewal: cost per cited fact, and which node type supplied the new citations. Sample them as an AI visibility baseline would; four weeks cannot judge a quarter.

Where this breaks

First, when checkout happens inside a marketplace, off-site returns are thin. Platform AI reads listings, reviews, and fulfillment — fix those first.

Second, regulated categories — medical, children’s, powered devices — cannot swap certifications for media; compliance answers prefer regulators and test bodies.

Third, a product with no verifiable difference gets its platitudes amplified, and communities amplify drawbacks in the same breath.

Fourth, communities do not sell placement. Paid sentiment is usually identified and removed, and the brand name carries the residue.

Fifth, directories matter in the disambiguation stage only; once the entity is clear, more listings add nothing.

This ordering is work for the strategy hub, sitting with market choice and budget cycles, not an agency’s rate card.

This quarter, run twenty buyer questions through two assistants, copy down every source in the answers, tag each as media, community, creator, or directory, and note whether it holds one liftable fact. Then halve the lowest-ranked budget line and move it to first place — no new budget, no agency revision.