Brand Lift Index

Quantify brand visibility movement with confidence intervals. Know when a shift is statistically significant versus random variation.

What it measures

The Brand Lift Index answers a deceptively simple question: is your brand becoming more or less visible in AI-generated responses, and is that change real or just noise?

Weekly (or biweekly), the system probes a curated corpus of prompts across LLM platforms and measures how often your brand is cited. Raw citation rates are then processed through three layers of statistical rigour — Wilson confidence intervals for precision, EWMA smoothing for trend clarity, and CUSUM change detection for alert reliability.

Statistical methods

Three complementary statistical techniques work together to separate real signal from noise.

Wilson Score Interval

Confidence interval for citation rates

When we observe that your brand was cited in 7 out of 10 probes on ChatGPT, the raw rate is 70%. But with only 10 samples, how confident are we in that number? The Wilson score interval provides a 95% confidence interval that accounts for small sample sizes. With a handful of samples the interval is wide; with a hundred it narrows considerably. This prevents overreacting to volatile small-sample results.

EWMA Smoothing

Separating signal from noise in weekly measurements

Each weekly measurement is noisy. Exponentially Weighted Moving Average smooths the time series by giving more weight to recent observations and less to older ones. This reveals genuine trends without overreacting to single-week spikes.

CUSUM Change Detection

Detecting when brand visibility actually shifts

CUSUM (Cumulative Sum) control charts detect sustained shifts in a process mean. We track two accumulators — one for upward shifts and one for downward shifts. When either accumulator exceeds a threshold, we flag a statistically significant change. This distinguishes a genuine brand visibility shift from random week-to-week variation. A CUSUM flag means something real has changed — not just noise.

How the pipeline works

Corpus Configuration

You select 10–20 prompts from your tracked set to form a measurement corpus. These should be a representative mix of awareness, consideration, and decision prompts. The corpus runs on a weekly or biweekly cadence across your chosen platforms (ChatGPT, Gemini, Claude).

Probe Execution

Each run sends every corpus prompt to every platform multiple times — repeated sampling is what makes the statistics meaningful. Each probe records whether your brand was cited in the response.

Per-Prompt Statistics

For each prompt-platform combination, we compute the Wilson score interval: given the probes observed and the citations recorded, what is the 95% confidence interval for the true citation rate? Prompt-platform pairs with too few samples are flagged as insufficient.

Platform Rollups

Prompt-level stats are aggregated into per-platform rollups (ChatGPT, Gemini, Claude) and an overall "all platforms" rollup. Each rollup includes: total probes, total citations, visibility percentage, and the 95% confidence interval.

EWMA Smoothing

The current run's visibility percentage is blended with the historical EWMA. This produces a smoothed trend line that filters out week-to-week noise while responding to genuine shifts.

CUSUM Alert Detection

The smoothed value is fed into the CUSUM control chart. If the cumulative deviation from the historical target exceeds the threshold, a Brand Lift alert is fired with direction (increase or decrease), from-percentage, and to-percentage. These alerts appear in the Brand Lift hero card and the alerts page.

Frequently Asked Questions

What does the Brand Lift Index percentage mean?

The headline percentage is your brand's visibility rate: the fraction of probes where your brand was cited in the LLM response. A visibility of 45.2% means your brand appeared in roughly 45 out of every 100 prompts sent to AI platforms.

What does the confidence interval [CI] mean?

The 95% confidence interval tells you the range within which the true visibility rate almost certainly falls. If the CI is [38.1% – 52.3%], there is a 95% probability that the real citation rate is between 38.1% and 52.3%. Narrower intervals mean more precise measurement, driven by larger sample sizes.

What triggers a CUSUM alert?

A CUSUM alert fires when the smoothed visibility moves far enough from its historical target that the change is statistically unlikely to be noise. The system tracks both upward and downward deviations, so you are alerted to both improvements and declines.

How many prompts should I include in the corpus?

We recommend 10–20 prompts representing a mix of your key topics. Too few prompts produce wide confidence intervals; too many increase cost without proportional precision gains.

What does the EWMA alpha control?

Alpha determines how responsive the smoothed trend is to new data. A higher alpha reacts faster to changes but is noisier; a lower alpha is smoother but slower to reflect real shifts. The platform default balances responsiveness with stability for weekly measurement cadences.

How much does a run cost?

Each probe involves an LLM API call. Costs vary by platform and corpus size, but a typical run costs approximately $2–5 USD. Actual costs are tracked and displayed on the dashboard.

Can I trigger an on-demand run?

Yes. From the Brand Lift section, you can trigger an immediate run outside the scheduled cadence. This is useful for checking the impact of a specific campaign or content change without waiting for the next scheduled window.