A clear expected result
State what you are changing, which audience or question set it should affect, and how you expect visibility to move.
AI Prompt Experiments
Test whether a change is associated with a repeatable visibility lift.
Compare two prompt groups on one AI platform, repeat the measurement automatically, and see whether an observed visibility difference repeats. Treat the result as evidence of association, not proof of causation.
A fair, repeatable test
Separate a promising movement from a repeatable result
A dashboard trend can tell you that visibility moved. An experiment records what you expected, compares two prompt groups on the same AI platform, and repeats the measurement before treating one side as a dependable lead.
Two prompt groups
Assign at least five tracked or custom prompts to the control and variant groups so both can be measured repeatedly.
One selected AI platform
Choose the AI platform where the change matters. Each experiment uses one platform so the result stays clear.
Three-step setup
Build the test from prompts already in your workspace
The setup keeps the expected result, prompt groups, AI platform, and final launch review together.
Define the hypothesis
Give the experiment a name, explain what you are changing, and state why it should move visibility.
Pick the prompt groups
Choose control and variant prompts from your tracked set, or add a custom prompt with its market context.
Pick the AI platform and launch
Select an available model, review the prompt counts and expected result, then launch or save the experiment as a draft.
One visibility score, three components
See what created the difference between the two sides
The headline lift shows the overall result. The component breakdown shows whether the difference came from appearing more often, appearing more prominently, or gaining ground in the competitive answer.
Mention Frequency
How consistently the brand surfaces across the responses in each experiment side.
Mention Quality
How prominently the brand appears when the selected model includes it in an answer.
Share of Voice
How much attention the brand receives compared with other brands in the selected prompt group.
Repeated runs + confidence
Watch the result build instead of judging one answer
Watch both scores develop as new runs arrive and see when the difference becomes consistent enough to trust.
01Control and variant visibility scores
02Percentage lift and point difference
03Visibility trend across collected runs
04Progress toward the minimum runs and the next scheduled run
05Whether the difference is consistent enough to trust
06The current leading side
When the evidence is ready
Turn the result into a clear next step
When the difference becomes consistent enough to trust, the experiment shows the leading side and the next choices available to your team.
Use the stronger direction
Use the leading side as evidence for the next implementation or testing cycle, without treating prompt-group differences as proof of causation.
Run for longer
Keep collecting repeated test runs when the result needs more evidence.
Conclude as inconclusive
Keep the learning without treating either side as a proven winner.
Tracking + experimentation
Keep daily monitoring separate from a scoped test
Your custom prompt set remains the everyday view of current AI visibility. An experiment gives one expected result its own control and variant groups, AI platform, repeated runs, and confidence check.
Ongoing
Prompt tracking
Monitor the questions that matter to the business and open the original answers behind each result.
Scoped test
Prompt experiment
Test a clear expected result with control and variant prompt groups on one selected AI platform.
Continue with friction AI
Choose the right questions, make the change, and inspect the proof
Start with your custom prompt set, use Actions to find an improvement worth making, and keep every experiment connected to the original AI answers.
Choose the questions to test
Build and edit the prompt set that tracks the customer decisions most important to your business.
Explore featureFind a change worth testing
Audit your website, profiles, channels, and third-party coverage to identify a practical improvement.
Explore featureInspect the answers behind the result
Open the customer question, original AI answer, platform, competitors, and supporting sources.
Explore featureFAQ
How AI Prompt Experiments work
What is an AI prompt experiment?
An AI prompt experiment compares control and variant prompt groups on one selected AI platform. friction AI repeats the measurement, shows the visibility difference, and tells you whether that difference is consistent enough to trust.
How many prompts does each side need?
Each experiment requires at least five control prompts and five variant prompts before it can start. Prompts can come from the tracked workspace or be added as custom prompts with a market.
Does one experiment run across every AI platform?
No. Each experiment runs on one selected AI platform. ChatGPT, Claude, Gemini, and Perplexity options are available according to the account plan. Use separate experiments when you need to test another platform.
Which metric does the experiment track?
Current experiments track visibility. The results also break that visibility score into Mention Frequency, Mention Quality, and Share of Voice so you can see what created the difference.
How quickly does an experiment start?
After launch, the first run is scheduled within four hours. The experiment then repeats automatically and shows the next scheduled run, progress toward the minimum, and current results.
How should we interpret the result?
Each side needs at least five prompts. Scheduled runs compare group-level visibility on one selected AI platform and report each group score, the observed lift, collected runs, and current confidence. Because the groups can contain different prompts, the result shows an association between prompt group and visibility, not proof that a content change alone caused the difference.
What statistical information is reported?
When enough scheduled runs have been collected, the analysis records the test type used, control and variant sample sizes and means, variability, effect size, p-value, and the configured significance threshold. Repeated AI responses are measurements within scheduled runs, not independent observations of customer behavior.
Does friction AI publish the content change?
No. Your team makes the change in the relevant site or channel. friction AI provides the experiment structure and measures the visibility outcome on the selected AI platform.
Stop guessing which change moved visibility
Define the hypothesis, launch the prompt experiment, and let repeated runs build the evidence.