Annual data asset · Edition 2026
The State of Social Proof 2026
An open benchmark of how social proof actually performs on live websites. The measurement standard is published first — definitions, formulas, panel rules — and the aggregated numbers follow, so every figure can be checked against the method that produced it.
Last reviewed 2026-08-10 · Reporting window open, results pending
10 metrics
Each with a written definition and reproducible formula.
Holdout-only lift
Sites without a holdout are excluded from the lift figure.
Aggregate only
No site, customer, or visitor is identifiable from a figure.
The benchmark metrics
Engagement
| Metric | Definition & formula | 2026 result |
|---|---|---|
| Notification impression rate | Share of measured sessions in which at least one social proof notification was displayed.sessions with ≥1 notification impression ÷ measured sessions | Collecting |
| Notification interaction rate | Share of displayed notifications that received a click or tap.notification clicks ÷ notification impressions | Collecting |
| Dismissal rate | Share of displayed notifications a visitor actively closed.dismissals ÷ notification impressions | Collecting |
Conversion
| Metric | Definition & formula | 2026 result |
|---|---|---|
| Exposed-session conversion rate | Conversion rate of sessions that saw at least one notification.conversions in exposed sessions ÷ exposed sessions | Collecting |
| Unexposed-session conversion rate | Conversion rate of sessions in the same window that saw no notification.conversions in unexposed sessions ÷ unexposed sessions | Collecting |
| Measured lift | Difference between exposed and unexposed conversion rates, reported only for sites running a holdout so the comparison is causal rather than correlational.(exposed CR − holdout CR) ÷ holdout CR | Collecting |
Trust surface
| Metric | Definition & formula | 2026 result |
|---|---|---|
| Review display coverage | Share of participating sites displaying aggregated third-party reviews on a revenue page.sites with reviews on a product, pricing, or checkout page ÷ participating sites | Collecting |
| Testimonial freshness | Median age of the newest testimonial displayed on a participating site.median(days since newest displayed testimonial) | Collecting |
Operations
| Metric | Definition & formula | 2026 result |
|---|---|---|
| Time to first notification | Median elapsed time between script installation and the first live notification.median(first notification timestamp − install timestamp) | Collecting |
| Active campaigns per site | Median number of concurrently active notification campaigns.median(active campaigns per site) | Collecting |
Panel and methodology
Unit of measurement
One participating website with at least 30 days of continuous NotiProof measurement
Reporting window
Rolling 12 months, reported annually
Privacy
Every figure is aggregated across accounts. No individual site, customer, visitor, or personally identifiable record is published or inferable from a reported figure.
Included
- Sites with an installed NotiProof script and a defined conversion event
- A minimum of 1,000 measured sessions in the reporting window
- No manual overrides of notification data during the window
Excluded
- Staging, internal, and test properties
- Accounts active for fewer than 30 days
- Sites where the conversion event changed mid-window (the comparison is not like-for-like)
Questions about the report
What is the State of Social Proof report?
An annual, openly documented benchmark of how social proof actually performs on live websites — the metric definitions, the panel rules, and the measured results, published together so any figure can be checked against the method that produced it.
Most social proof statistics in circulation are untraceable: a percentage with no panel definition, no window, and no way to tell whether the comparison was causal or correlational. That makes them useless for planning and dangerous for forecasting.
This report inverts the usual order. The measurement standard is published first and the numbers follow, so readers can judge the method before they see a result they might want to believe.
Which metrics does the benchmark track?
Ten metrics across four segments: notification engagement, conversion outcomes, the trust surface a site presents, and the operational reality of running campaigns. Each has a written definition and formula so it is reproducible.
Engagement metrics describe whether social proof is seen and acted on at all. Conversion metrics describe whether exposure coincided with — or, where a holdout exists, caused — more conversions.
Trust-surface and operational metrics capture the context that explains the other two: a site with stale testimonials and a single campaign is not comparable to one refreshing proof weekly across the funnel.
How is conversion lift calculated?
Only from sites running a holdout group, comparing exposed sessions against withheld sessions in the same window. Sites without a holdout contribute engagement and trust-surface metrics but are excluded from the lift figure.
Comparing exposed and unexposed sessions without a holdout measures self-selection as much as effect: visitors who browse longer see more notifications and are already likelier to buy. Reporting that difference as lift is the single most common error in vendor-published social proof statistics.
Excluding non-holdout sites shrinks the sample and makes the headline figure less flattering. That is the intended trade-off.
Who is included in the panel?
Live sites with at least 30 days of continuous measurement, a defined conversion event, and a minimum of 1,000 sessions in the window. Staging properties, short-lived accounts, and sites that changed their conversion event mid-window are excluded.
Inclusion rules are published before each wave so they cannot be tuned after the fact to produce a better headline. Any change to the rules between editions is stated in the edition notes rather than applied silently.
Every figure is an aggregate. No individual site, customer, or visitor is identifiable from a published number.
How should teams use these benchmarks?
As a sanity check on your own measurement, not as a forecast. A benchmark tells you whether your engagement and conversion figures are in a plausible range for comparable sites; it cannot predict what a change will do on your traffic.
If your interaction rate sits far below the benchmark range, the likely causes are placement, timing, or relevance of the notification — problems worth fixing before you attribute anything to the channel.
If it sits far above, verify that the events feeding your figures are counted the way the definitions here describe. Unusually good numbers are usually a measurement artefact.
When are the numbers published?
Each edition publishes its measurement standard first, then the aggregated results once the reporting window closes and the panel meets the minimum size for every reported metric. Metrics that do not meet it are marked as such rather than reported thinly.
A metric with too few contributing sites is reported as unavailable for that edition. Publishing a fragile figure and revising it later does more damage to a benchmark than an honest gap.
Edition-over-edition changes will be shown with both the current and prior definitions where a definition changed, so trends are not read across incompatible methods.