The OE Checks tab.
This is the one manual check in the whole tool: verifying the open-ended answers. Once the OE flags are verified, every risk score on the Respondent Risk tab can be trusted at face value. Let the AI do the first pass.
The three AI checks
| Check | Use it on | Cost | Notes |
|---|---|---|---|
| AI relevance | Every text question, every survey | Cents per 1,000 answers | Flags answers that do not answer the question. |
| AI written | Every text question with answers over 3 words | Cents per 1,000 answers | Flags machine-written text. Scores as a soft amber signal, never red on its own. |
| AI brands | Brand awareness questions only | Tens of cents per 1,000 answers | Searches the web to confirm an answer is a real brand. Expensive and accurate: use it when relevance is not sharp enough on brand lists. |
The routine: for every non-brand text question, always run AI relevance and AI written before reading anything yourself. On a typical survey that first pass costs cents and removes most of the reading. (AI written needs more than 3 words per answer; under that there is nothing to judge authorship on.)
Counts are filters
Every count on the AI toolbar is clickable: passed, flagged, unchecked. Click a number to see exactly those answers and nothing else.
Strictness: the four verdicts
The relevance check returns one of four verdicts per answer: Definite yes and Maybe yes (both pass, green), Maybe no (a red flag by default) and Definite no (always red). If a question comes back too strict, flip the Maybe no switch on that question: Maybe no becomes a yellow Borderline (AI) flag, scored far more gently. A question with 86 hard fails can drop to a handful of reds plus the rest borderline. The switch is per question, and Definite no stays red regardless.
Staying in charge
- Reflag anything, in either direction: AI-flagged to clean, or clean to irrelevant. You do not have to check every AI verdict; scan the groups, spot-check, and correct what is wrong.
- Click an answer to see every respondent who gave it, along with all of their other text answers: the fastest way to judge a borderline case in context.
- Answers sort A to Z inside every flag group, so near-duplicates and keyboard mash sit together and scanning stays fast.