Aplo is a sampling pipeline: a fixed prompt panel, repeated runs across engines, raw responses retained, structured metrics out the other end.
You have probably already scripted a version of this and found the hard part is not the API call. It is that responses vary, so single samples are noise; that changing a prompt silently invalidates your history; and that parsing citations out of prose reliably is its own project.
The mechanism, not the feature list. If any of these are wrong, the data is wrong.
Each prompt and model pair runs several times per event. The measured quantity is probability of recommendation, so a single call is not a data point.
Panel questions cannot be edited, only superseded, and the constraint is enforced in the database rather than by convention. Otherwise model drift and question drift are indistinguishable.
Raw responses are retained and re-parseable. Extraction is versioned, so improving the parser re-enriches history instead of only improving new data.
Provider APIs, not consumer-app automation. Search-grounding behavior is provider-specific and is exactly what is being measured, so requests are not routed through an aggregating gateway. Automating the consumer apps would also put the core loop on the wrong side of provider terms.
Changes are diffed against competitors in the same panel over matched windows, with adjacent categories as a second control. Effect sizes carry uncertainty, and we report lift observed rather than caused.
Webhook notifications to Slack and Discord, plus CSV and JSON export, are in scope for the first release. A programmatic API with a webhook suite — querying responses, pulling citations, piping metrics into your own dashboards — is a later roadmap item and is not available yet. Google Search Console and analytics sync, so AI visibility sits next to conventional search data, is planned.
We would rather tell you the API is not built than let you plan an integration around it.
Slack & Discord webhooks
Event notifications on schedule
CSV & JSON export
Bulk pull of responses and metrics
Programmatic API
Query responses, pull citations, pipe to dashboards
GSC & analytics sync
AI visibility next to conventional search data
Publishing limitations is the single strongest trust signal available. Here are ours.
Real assistant sessions carry memory, personalization, geography, and interface experiments an API call does not reproduce. Consistency over time is the value, not perfect simulation. We intend to publish the calibration between API sampling and real browser sessions rather than leave you to assume it.
Referrer detection misses zero-click answers and stripped referrers, and AI influence often resurfaces later as branded search. Treat any attributed number as a lower bound.
In-panel competitors are affected by your changes too — publishing a comparison page alters their environment. This is why effect sizes come with uncertainty rather than a verdict.
The waitlist is small and curated. You will get methodology notes and architecture decisions before the product is public — and we will not waste your time with marketing emails.
