Skip to content
Coeff IQ

AI

Reading AI answers and their limits

Coeff IQ's AI answers are built to be checked. This page explains how to read them and what they cannot tell you.

Every statement shows its evidence

Every statement in an answer cites the figures it rests on. Those figures come from your synced data, not from the AI's wording, so you can compare each sentence with the number beside it.

An answer that the AI cannot support with its evidence is rejected. It is not shown, not saved and not charged.

Observations, explanations and next steps

Answers keep three kinds of statement apart:

  • What the data shows: observations taken from your figures.
  • Possible explanations: ideas worth checking. The data does not prove a cause, and the AI does not claim one.
  • Suggested next steps: things to look into.

Treat explanations and next steps as suggestions for you to judge, not as conclusions.

What limits an answer

Each answer lists anything that affects it, such as:

  • a data source whose last sync failed or is out of date;
  • missing days: days with no stored data are never treated as zero;
  • incomplete recent days: if the period includes today, or days a provider has not reported yet, volume changes are not compared;
  • which integrations it covers.

Read these before acting on an answer. Resolving them, for example by fixing a failed sync, can change the next answer.

Coverage and time zones

  • AI uses only data synced into Coeff IQ, for the client you are viewing.
  • Unusual-day detection and performance findings cover Google Analytics 4 and Search Console only.
  • Dates are UTC calendar days. GA4 days stay in your property's time zone, and Search Console days in US Pacific Time.
  • AI Chat uses the 30 days ending yesterday (UTC). A report's AI summary uses only that report's saved data, so later syncs never change it.

Older results

AI results created before answers were checked against evidence are labelled not validated. Generate a new result if you need one with evidence.

Mistakes

AI can still be wrong. If an answer contradicts its own evidence, or the evidence looks wrong, contact support with the client name, the feature, and the time of the answer. If a reference is shown, include it too.

We use necessary cookies to run this site, and — only with your consent — Google Analytics 4 to understand aggregate usage. See our Privacy Policy and Cookie Policy.