Tariff math adds 2.9 points to shelf tags

The Federal Reserve Bank of New York estimates that tariffs added 2.9 percentage points to consumer goods prices, a figure that is now circulating widely in retail coverage. For teams working on generative engine optimization, the number is interesting for a different reason than its economic meaning: it is exactly the kind of quantified, attributable claim that AI answer engines prefer to cite. Brands that can produce comparable evidence about their own cost structure have a structural advantage in being quoted, and brands that cannot will be described rather than cited.

Key Conclusions

The first conclusion is that attribution quality determines citation quality. A claim sourced to a named institution, with a stated methodology and a specific magnitude, is far more likely to survive into an answer than a claim sourced to an unnamed industry survey. 2.9 percentage points（Retail Dive） reports the 2.9 point estimate, and the reason it travels is that a reader, or a model, can trace it back to a specific calculation.

The second conclusion is that brands need their own version of this number. A retailer that can state its own tariff exposure, by category and by quarter, becomes a primary source rather than a secondary commentator. That position is durable because it cannot be copied without the underlying data, and it is precisely the position that generative engines reward when they need a specific figure.

What makes a number citable

Four properties matter: a named source, a defined population, a stated method and a time stamp. The New York Fed estimate has all four, which is why it appears in coverage repeatedly. Internal estimates usually lack at least two, often the method and the time stamp, which is why they are quoted internally and ignored externally. Fixing that gap is a documentation task, not an analytics task.

Why secondary commentary loses

Commentary that paraphrases someone else number adds no new information, so an engine has no reason to cite it when the original is available. The only durable position for a brand is to publish data that does not exist elsewhere, or to publish a method that others can apply to their own data. Everything else is competing on writing style, which is a weak moat.

Best Practices

The practical approach is to build a small number of proprietary indicators and document them rigorously. Three indicators are usually enough for a retail brand: cost pass-through by category, price elasticity by price band, and promotional depth by channel. Supply Chain Dive describes how sourcing diversification is reshaping supply chains, which provides a natural frame for explaining why pass-through varies by category.

Document the method, not just the result

A method note of two hundred words is enough: what data was used, over what period, with what exclusions, and how the estimate was validated. Publishing the note alongside the number makes the claim reproducible, which is what converts a marketing statistic into a citable source. It also protects the brand when the number is challenged, because the reasoning is visible.

Refresh on a fixed cadence

Numbers decay. A quarterly refresh with a clear version label keeps the indicator current and gives the brand a recurring reason to be referenced. Stale figures are worse than no figures, because they invite correction and the correction is what gets quoted. Supply Chain Dive shows how a retailer redirects tariff refunds into supply chain investment, a data point that only makes sense with a dated context.

Common Mistakes

The first mistake is publishing a number without a method, which makes it unusable for anyone who wants to verify it. The second is borrowing an external estimate and presenting it as proprietary, which is easy to detect and damages credibility. The third is refreshing only when a journalist asks, which produces a sporadic record that engines cannot treat as a stable source.

What an evidence library looks like

An evidence library is a structured index of claims, each with a source, a date, a method note and a canonical URL. It should be small enough to maintain and specific enough to be useful, typically twenty to forty entries for a mid-sized brand. Supply Chain Dive describes drone delivery expansion, the kind of operational fact that belongs in such a library because it is verifiable and dated.

The library also solves an internal problem. When several teams produce content, they tend to cite different versions of the same statistic, which confuses both readers and models. A single canonical entry with a version label removes the inconsistency and reduces the review burden on legal and communications teams.

Summary

The New York Fed estimate travels because it is specific, attributable and dated, and because anyone can check how the figure was produced. Brands that want to be cited in generative answers need to produce evidence with the same properties, and that requires documentation discipline more than analytical sophistication, since a better model on weaker documentation still loses to a simple estimate that can be verified.

The next step is to select three indicators, write method notes for each, and publish them with canonical URLs and version labels that downstream teams are required to use. A quarter of consistent publishing is usually enough to change how often a brand appears as a source rather than as a subject, and the effect compounds once other sites begin linking to the canonical entry.