Claim language
Evidence-to-claim language matrix
Organization byline: Measurement Press editorial desk. This page does not list an individual author or reviewer. Read the editorial method and corrections policy.
About 11 min read
A claim is often weakest at the verb. The evidence may be real, but the sentence can still imply more certainty, scale, or cause than the source trail supports.
Use this matrix when a headline, chart, vendor slide, campaign readout, survey finding, or research report needs to be rewritten before it shapes a decision. The goal is not softer language for its own sake. The goal is a sentence that tells readers what is known, what is only suggested, and what still needs a better comparison.
The matrix
Start with the evidence actually in hand. Then choose the strongest claim that keeps the denominator, comparison, and causal design visible.
The matrix is a gate, not a word-swap exercise. First identify the decision the sentence could influence. Then trace the audience, comparison, outcome, source, and uncertainty that would have to be true for the sentence to stand.
| Evidence in hand | Supportable wording | Do not say yet | Next check |
|---|---|---|---|
| One source statement | "The source says..." or "The source alleges..." depending on the source role and documentation. | That the claim is established fact. | Look for the closest record, counter-source, method note, or direct evidence. |
| Public record count | "Recorded cases in this source rose, fell, or totaled X during the stated period." | That real-world prevalence changed without checking reporting rules and the eligible population. | Name record scope, missing universe, population base, and revision risk. |
| Survey or poll result | "In this sample, X% of respondents answered..." | That the public, customers, or voters broadly believe the claim. | Check sample source, weighting, subgroup base, question wording, field dates, and uncertainty. |
| Before-and-after chart | "The metric changed after the event in the shown window." | That the event caused the change. | Check seasonality, starting level, concurrent changes, and whether a comparison was chosen before the result was visible. |
| Attribution report | "The report credited these outcomes under this identity and lookback rule." | That the campaign created the credited outcomes. | Check conversion lag, prior intent, channel blind spots, match rate, and whether a holdout or matched baseline exists. |
| Protected lift test | "The test estimates incremental impact for this audience, outcome, window, and assignment design." | That the result will hold across all markets, channels, creatives, or future budgets. | Check assignment integrity, leakage, outcome capture, uncertainty, and external validity. |
| Marketing mix model | "The model estimates contribution under these controls, priors, calibration inputs, and uncertainty ranges." | That a channel has an exact causal ROI because the model fit is strong. | Check calibration evidence, response curves, sensitivity tests, omitted variables, and decision limits. |
| Brand lift study | "The respondent measure moved in the study population under the disclosed recruitment and weighting rules." | That the campaign drove sales or durable market change. | Check exposed/control balance, respondent quality, question wording, field timing, and business-outcome evidence. |
| Attention or viewability report | "The campaign delivered stronger exposure-quality signals by the reported definition." | That attention alone proves brand or sales impact. | Check measurability, invalid traffic, frequency, placement context, and whether the outcome was tested separately. |
Rewrite clinic
Original: "Complaints surged, proving the policy failed."Cleaner: "Recorded complaints rose in the shown period. To judge the policy claim, readers still need the eligible population, reporting rule changes, prior seasonal pattern, and comparable locations or periods."
Original: "The campaign generated 10,000 sales."Cleaner: "The campaign report credited 10,000 sales under its attribution rule. Incremental sales require a holdout, matched baseline, geo comparison, model counterfactual, or another design chosen before results were visible."
Original: "High-intent audiences performed twice as well."Cleaner: "The high-intent audience produced a higher observed response. The result may reflect pre-existing demand, eligibility, reachability, bid differences, or creative fit unless the comparison controlled for those differences."
Original: "The survey proves buyers want this feature."Cleaner: "In the surveyed sample, respondents preferred this feature under the given wording. Confidence depends on sample source, weighting, field dates, alternatives shown, and whether stated preference predicts behavior."
Original: "The model shows the channel is the best investment."Cleaner: "The model estimates a stronger response for the channel under its assumptions. A budget decision should inspect calibration, uncertainty, saturation, constraints, and whether recent tests agree with the modeled response curve."
Simulated worked example: when a higher count is not a stronger claim
Simulated Worked Example. These numbers are invented to demonstrate the edit method. They are not a report about a real campaign, publisher, organization, or public event.
Imagine a campaign memo that says, “The new placement caused a 25% increase in qualified leads.” The review packet contains 125 recorded leads from 1,000 eligible visits after launch and 100 recorded leads from 1,000 eligible visits before launch. It has no randomized holdout, no matched comparison, and no documented change log for lead qualification.
| Gate | Calculation or record | What the evidence supports |
|---|---|---|
| Observed count | 125 - 100 = 25 additional recorded leads | The packet records 25 more leads in the after window. |
| Comparable rate | 125 / 1,000 = 12.5%; 100 / 1,000 = 10.0%; difference = 2.5 percentage points | The recorded lead rate was higher in this simulated window. |
| Relative change | 2.5 / 10.0 = 25.0% | A 25.0% relative rate increase is descriptive, conditional on the stated denominators. |
| Missing design | No protected counterfactual; qualification rule and lag are not verified | Do not say the placement caused the increase or that it will repeat at another budget. |
The edited sentence is: “The placement window recorded a 2.5 percentage-point higher lead rate, from 10.0% to 12.5%, among the stated eligible visits.” A stronger causal sentence would require a credible comparison, stable outcome definitions, complete follow-up, and uncertainty appropriate to the design. If the denominator, lead rule, or conversion lag changes, recalculate before changing the verb. This follows the CDC distinction between describing patterns and explaining them, and the Census requirement to expose assumptions, limitations, methods, and uncertainty for derived results.
Decision packet: how much should the sentence move?
Use the same packet to test whether a wording change is caused by better evidence or merely by a different denominator. The numbers below extend the simulated example; they are not a claim about a real campaign. Recalculate the rate first, then decide whether the sentence can remain descriptive, become directional, or must be held.
| Packet change | Reproducible check | Allowed wording | Decision |
|---|---|---|---|
| Stated packet | 125 / 1,000 = 12.5%; 100 / 1,000 = 10.0%; +2.5 percentage points; 2.5 / 10.0 = +25.0% | "The after window recorded a higher lead rate among the stated eligible visits." | Descriptive only |
| After-window denominator falls to 900 | 125 / 900 = 13.9%; the base is no longer like-for-like with 100 / 1,000 | "The windows are not comparable until eligible-visit definitions are reconciled." | Hold comparison |
| 20% of after-window leads fail the stated qualification rule | 125 x 0.80 = 100 qualified leads; 100 / 1,000 = 10.0%, equal to the before rate | "The apparent lead-rate increase disappears under the stricter qualification rule." | Revise claim |
| 10% of after-window leads remain unobserved | 125 / 0.90 = 138.9 possible final leads; 138.9 / 1,000 = 13.9% if the missingness assumption holds | "The observed rate is provisional; the conclusion depends on follow-up completeness." | Hold causal wording |
The packet produces one practical rule: a larger count cannot rescue an unstable denominator, outcome definition, or follow-up window. The FTC substantiation source is relevant to matching an objective advertising claim with the support communicated; it is not a license to treat a descriptive campaign report as causal proof. The CDC sources guide the distinction between describing counts and analyzing explanations, while the Census reporting standard supports exposing assumptions, methods, limitations, and uncertainty. None supplies a universal pass threshold for this simulated packet.
Source trail for the edit rules
These references are the basis for the boundaries used in this matrix, not evidence for the simulated numbers above:
- FTC Policy Statement Regarding Advertising Substantiation: objective and implied advertising claims need a reasonable basis before dissemination, and the support should match the level of claim communicated.
- CDC Field Epidemiology Manual: Describing Epidemiologic Data: counts and rates describe what, how much, when, where, and among whom; interpretation should not silently turn description into cause.
- CDC Field Epidemiology Manual: Analyzing and Interpreting Data: apparent associations need checks for spurious explanations and a planned analysis suited to the question.
- U.S. Census Statistical Quality Standard E2: Reporting Results: derived results should make available uncertainty, assumptions, limitations, methods, and how those limits affect interpretation.
Boundary: The FTC statement is an advertising-substantiation policy, not a general editorial accuracy ruling. CDC and Census material helps keep description, uncertainty, and causal interpretation distinct; none of these references validates the simulated figures or determines a claim’s legal status in every context.
Confidence verbs
Verb choice should change when the evidence changes. Do not let the strongest verb arrive before the strongest design.
| Evidence tier | Useful verbs | Reader expectation | Avoid unless stronger evidence exists |
|---|---|---|---|
| Descriptive | shows, reports, records, counts, credits | The source describes what was observed or logged. | proves, caused, drove, changed |
| Directional | suggests, is consistent with, points toward, warrants testing | The pattern is useful but incomplete. | confirms, establishes, settles |
| Comparative | outperformed, lagged, differed, ranked higher | The comparison is visible and reasonably like-for-like. | won, failed, dominated, guaranteed |
| Causal | increased, reduced, caused, generated | A protected comparison estimates what changed because of the intervention. | Use only when assignment, counterfactual, outcome, and uncertainty are visible. |
| Decision-grade | supports, justifies, limits, rules out | The evidence is strong enough for a named decision within stated boundaries. | proves forever, applies everywhere, eliminates risk |
Campaign readout language
Advertising reports often mix delivery, exposure quality, traffic, leads, matched outcomes, and modeled impact. Keep the sentence inside the strongest evidence lane the report actually contains.
| Report field | Useful sentence | Boundary |
|---|---|---|
| Impressions, reach, frequency | "The campaign delivered against the planned exposure target with this reach and frequency profile." | Delivery is not persuasion or sales impact. |
| Viewability, invalid traffic, attention | "The exposure-quality signals met or missed the stated quality threshold." | Quality signals describe opportunity to see, not business lift. |
| Clicks and engaged sessions | "The creative and context produced this observed traffic response." | Traffic response can reflect interest, targeting, placement, or offer strength. |
| Form fills or leads | "The campaign produced these recorded leads with these qualification and follow-up statuses." | Lead count is weak without quality, duplication, and sales-stage context. |
| Matched conversions | "The matchback found these outcomes among matched identities under the disclosed window." | Matched outcomes are not incremental outcomes without a comparison. |
| Holdout or geo lift result | "The designed comparison estimates this incremental effect, with these uncertainty and leakage limits." | The estimate is bounded by design quality and setting. |
Worked downgrade example
Suppose a renewal slide says, "The campaign drove high-value buyers and should be doubled next quarter." The report shows engaged sessions, form fills, and matched conversions, but it does not show a protected holdout, comparable baseline, buyer-side sales-stage quality, or conversion-lag adjustment. The sentence is too strong even if the campaign performed well operationally.
| Review step | What the packet shows | Allowed wording |
|---|---|---|
| Observed response | Traffic quality, engaged sessions, and lead forms improved inside the campaign window. | "The campaign produced stronger observed response in the measured window." |
| Buyer quality | Lead-stage status is partial and follow-up fields are missing for some records. | "Lead quality remains partially observed, so buyer value is not settled." |
| Comparison | No holdout, matched baseline, or pre-stated comparison was protected before launch. | "The report credits outcomes; it does not isolate incremental outcomes." |
| Decision | The result can support renewal of the tested package, not broad budget scale. | "Renew the proven placement set, fix sales-stage tracking, and plan a comparison before claiming lift." |
Six-question edit pass
Use the edit pass when the claim is already drafted and the team needs a practical way to find the overreach. Each question either preserves the current wording or forces the sentence down to a more honest evidence tier.
1. What sentence will readers remember?Find the headline, chart title, slide conclusion, or executive-summary line that carries the strongest interpretation.
2. What type of evidence supports it?Separate source statement, record count, survey result, descriptive dashboard, attribution report, model estimate, and designed experiment.
3. What is the denominator?Name the population, opportunity base, record universe, respondent group, campaign audience, or outcome-eligible group.
4. What comparison is being implied?Identify whether the sentence invites a prior-period, peer, holdout, market, source-role, or counterfactual comparison.
5. Does the verb match the design?Downgrade from causal to descriptive or directional language when the comparison was not designed before results were visible.
6. What would raise or lower confidence?End with the document, method note, base rate, sensitivity check, counter-source, or test design that would most change the judgment.
Use it with the rest of the library
For media claims, pair this matrix with the media claim audit worksheet, headline and source-mix checklist, denominator framing examples, public records and denominator checklist, source role and incentive map, and survey and poll claim checklist.
For advertising measurement, pair it with the campaign readout QA checklist, campaign reporting terms glossary, attribution window and conversion lag checklist, randomized lift test readout checklist, MMM readout QA checklist, and private marketplace measurement checklist.
Takeaway
Good evidence deserves precise language. Weak or incomplete evidence may still be useful, but only if the sentence stops before it turns description into proof, correlation into cause, or response into incrementality.