GEO-086·Issue 08·Authority & Off-site·86 / 132
Digital PR for GEO: Create Citable Facts, Not Press Releases
Releases get syndicated, not cited. A model quotes a number with a sample, a method with a date, a judgment with a name — here is the pipeline to ship them.
Issue focusOff-Site Authority: Digital PR, Communities, and the Knowledge Graph
A smart-lock brand enters the UK and keeps its quarterly PR calendar: new model, design award, a national retail listing. Each ships as a release, gets a day on two aggregators, stops. That week an assistant is asked whether a tenant may change the locks in a rented flat. The answer draws on a tenants’ union guide, a magazine walkthrough, and a landlord association’s note on written permission. Not one sentence comes from the releases.
AI quotes what it can re-check — a number with a sample, a method with a date, a judgment with a name — and a release carries none of the three.
A release answers a question nobody asked
A release is built for distribution, not extraction. Its subject is the company, its verbs are launch, unveil, win, and what follows is a stance, not a condition. It answers what happened here; the buyer asks which one to choose.
Retrieval reaches these pages; synthesis needs another sentence: who, under what conditions, what value, computed how. A release supplies only the who, so the model falls back on the safest line it can support — this brand also sells this category.
Twenty copies of one sentence are still one source
A release moves through a wire network onto dozens of pages unchanged; deduplication collapses them. Twenty restatements are one source with no exclusive information: corpus volume, not authority.
What gets quoted is the rewrite. An editor folds your number into an argument of their own, adds a byline and a date, and that sentence carries an identity the original never had. So the unit is not impressions. It is how many people used your data to write something of their own.
Three materials that get taken
Three kinds of material get lifted, and each fails predictably.
| Material | What it looks like | Who takes it | How it goes wrong |
|---|---|---|---|
| Original data | A customer survey, a cross-market price comparison, a year of returns | Reporters as a news peg; assistants for “roughly how much” | A headline percentage with no sample |
| Method and definitions | Test conditions, grading rules, what “quiet” means in decibels | Review sites building comparison tables | A marketing page with no parameters |
| Signed judgment | A founder on a category shift, an engineer on a trade-off | Feature reporting, trade press | A quotable line with no name or role |
All three need a home on your domain, with a method page, an update date, a downloadable table. Email the dataset and leave “contact us for data” on the site, and nobody has a sentence to lift. Producing them is another discipline — see citable assets — but the test stays fixed: the sentence must survive away from your site.
Five steps from a question gap to a second citation
| Step | Action | Owner | Passes when |
|---|---|---|---|
| 1 Find the gap | 20 buyer questions in the market’s language; log each number’s source | Growth | Recurring claims with no reliable source |
| 2 Pick the question | Four to six weeks, inside the category | Content | One sentence says what the data answers |
| 3 Produce | Sample, window, method, limits; data page, chart, CSV | Content, data | A stranger can reproduce it from the method page |
| 4 Distribute | Review sites, vertical media, data-literate posters; a cut per market | Marketing | Replies ask about method, not just acknowledge |
| 5 Track | Re-ask in three months; log sentence and credit | Growth | Domains listed with the exact sentences |
The order is not cosmetic. Skip step 1 and step 2 becomes what you felt like saying. Skip the method page and steps 3 and 4 fail together: no editor cites an undefined number. Skip step 5 and the next budget is a guess — see attributing PR to AI answers.
One failure earns its own line: a release in the costume of data journalism. The headline says survey, the body adds one figure, the sample and table are missing. The answer stays “this brand also sells this category” — at twice the cost.
Example: one renter survey, one sentence that travels
Example — brand, sample, and figures invented; the sequence is the lesson.
A UK smart-lock team drops the quarterly cycle for two studies a year. Step 1 supplies the topic: “does my tenancy allow a lock change” recurs across twenty English questions, and the answers rest on forum posts and an unsourced “most landlords say no.”
Five weeks, two deliverables: a survey of 500 renters with a method page naming panel, fieldwork dates, and margin of error; and a rental-friendly installation guide with a permission-request template for a letting agent.
No blast. Six tenant publications and two home-tech review sites each get a short email: one finding, the method page, the table, an offer to cut by city.
Result (Example): two outlets publish. One quotes the finding that 38% of renters do not know whether their tenancy permits a lock change; a review site links the guide in a renter buying article. Three months on, that 38% returns in answers, credited to the publication. The brand stays absent — not failure, since the fact enters the answer first.
Where this breaks
Low-consideration categories. Phone cases, cables, storage bins: models rarely compare brands at that depth; a study returns less than fixing listings and reviews.
No production cadence. A one-off survey expires within a year and the data page starts arguing against you. Without a refresh commitment, publish method, not a sample.
Regulated categories. Locks, child products, anything with a compliance line: a data conclusion reads as a regulatory claim. Settle legal wording before fieldwork.
Incentives pointed at the wrong number. Measure releases issued and you get releases; test only rivals on the metrics you win and you buy coverage now and a reputation for unreliable comparison later. Measure sentences lifted and the work changes.
Budget across media, communities, and data follows the off-site investment order; the wider map is the off-site authority hub, after the off-site trust network sets which nodes matter.
This month: write down the ten questions you most want answered, then mark the source behind every number in the answers. One line will recur with nothing checkable behind it. That line is next quarter’s study, and the first sentence of yours a stranger can quote.