GEO-089·Issue 08·Content & Semantics·89 / 132
Citable Assets: Original Data, Tools, and Templates
Most buying guides are interchangeable. A support-ticket fit table is not. How DTC teams build original data, tools, and templates that AI keeps citing.
Issue focusOff-Site Authority: Digital PR, Communities, and the Knowledge Graph
A US DTC brand selling monitor arms lays three years of guides beside assistant answers. Almost none show up. The support inbox is more damning: the same question every week — “my desktop is 1.2 inches thick with a beveled edge, will the clamp hold?” — and no page answers it. The guides say “sturdy build, tool-free installation, clean cable routing.” The sentence buyers need has been missing for two years.
The content AI keeps citing is rarely the best-written page on the topic. It is the material nobody else can produce: a dataset only you hold, a calculator that turns a vague question into a number, a template a competitor could copy and has not.
Citation is a materials gap, not a writing contest
A model assembling an answer looks for material that supports the sentence it will write. If only your site has it, two options remain: cite you, or drop the claim. Call this citation rigidity: a shortage of supply, not a panel scoring prose. Soft content sits opposite: hundreds of pages already say “lightweight, easy to fold, travel-ready,” so a better sentence is just the next interchangeable one.
The three types are scarce differently. Data is scarce in the sample: only you hold those tickets, return reasons, and warehouse lead times. A tool is scarce in its definition, turning a fuzzy question into computable inputs and outputs. A template is scarce in its distillation, an internal practice cleaned up for reuse. None is written; all are built, which is why one content hire and one quarter cannot copy them. Media and review sites pick them up because they are short of your material, not your opinions; pitching them belongs to the off-site trust network.
The failure mode is obvious. Re-sort public data, add an annual-report cover, skip the sample and the method, and a dozen sites publish the same shape within two weeks. The model has no reason to pick any of them.
Three asset types, three half-lives
Cost and shelf life diverge sharply enough that one budget line hides the trade-off.
| Asset | Typical form | Cost and speed | Half-life | Upkeep |
|---|---|---|---|---|
| Original data | Sample survey, aggregated ops data, bench test | High build, slow uptake: sample, definitions, review | Long: quoted a year on | Re-collect yearly, archive old |
| Tool | Selector, fit checker, duty calculator | Medium build, medium uptake | Long: while the source holds | Data source, interface |
| Template | Checklist, field spec, returns script | Low build, fast uptake | Short: easily copied | Version, license |
They are not three choices but three stages. A template is the quickest proof that this company makes things, not just commissions them. A tool absorbs one specific question at scale. Data holds the ground once a question is asked often enough that only a dataset can settle it. Most teams run it backwards, starting with data — the slowest, most expensive option — and the budget dies halfway.
Five steps from gap to maintained asset
| Step | Action | Owner | Passes when |
|---|---|---|---|
| 1 Find the gap | Run 20 pre-purchase questions through two assistants; mark answers that are only adjectives | Content, growth | One question per gap |
| 2 Choose the form | Decide whether the gap lacks a sample, a definition, or a structure | Content lead | You can say why only you could make it |
| 3 Write the method | Sample, window, calculation, limits, set before the finding | Data, ops | A third party could recompute it |
| 4 Host it | HTML table plus text summary; a conclusion under each chart, not a PDF | Content, front end | Key figures survive JavaScript off |
| 5 Maintain it | Review cycle, version number, archived versions, cut into article, chart, checklist | Content | The page shows a version and a date |
Step 3 is skipped most and decides everything. Data without a method reads like collateral; data with a sample size, a calculation, and a stated limitation reads like a checkable fact. A figure that lives only in a chart is invisible to extraction, as the machine-readable evidence audit shows.
Example: a fit table from the support inbox
Example — brand, figures, and outcomes are invented to show the order of work; no real product is described.
Return to the monitor arms. The tickets cluster into three constraints: desktop thickness, clamp versus grommet mount, monitor weight, and VESA pattern. The team builds one table — thickness range, mount type, weight ceiling, hole spacing, what it cannot hold — plus a checker that turns three inputs into a yes or no, with a reason.
Three details decide whether it gets cited. The dimensions state their basis: thickness behind a beveled lip, weight ceiling with the arm fitted. The tool page names in-house measurement and states the tolerance. Review runs quarterly, and the old version stays at its own URL so old citations do not break.
Within months, assistants answering “which monitor arm fits a thick desk” quote the table, and two review sites link the page. Whether that is progress is a question for share of citation, not a screenshot.
Where this breaks
First, the material is not unique. Re-sorted public data, or an industry average instead of your sample, is something anyone can publish; the citation goes to whoever published first.
Second, the method is missing. A study without sample size, window, or error note is just an opinion piece.
Third, maintenance lapses. Stale prices, discontinued models, and dead compatibility lists become a liability once cited: last year’s conclusion becomes this year’s.
Fourth, the tool is a facade. A login wall or results baked into an image is unreadable, and shoppers leave anyway.
Fifth, the asset is treated as traffic. Being cited is not being bought; the page still owes the reader a next step — buy, compare, ask a person.
Sixth, the category is regulated. Medical, child-safety, and financial claims clear compliance first.
An asset buys next year’s default answer; a campaign buys this quarter’s impressions. With that clear, the rest of the content and semantics line falls into order.
This month, take one question AI answers with adjectives. Build a table from data you already have, state the sample, method, and limits, publish at a stable URL, and in four weeks ask whose page the answer points to.