The AI summary of survey results

What it is

When a survey you sent to your team comes back with enough responses to actually say something, Verinode reads the results and writes you a short, plain-English paragraph: how the team feels, whether there is a clear pattern or a split opinion, and one specific thing to do about it. That paragraph is the AI summary, labeled AI Analysis on the results screen.

It is not a replacement for the raw numbers below it, and it is not Verinode making the call for you. The per-question ratings, the yes/no splits, and every written answer your team gave still sit right underneath it, in full. The summary is a fast first read: something to glance at before you scroll, not the only thing you look at.

Where to find it

1. Open Forms from the sidebar (/forms).
2. Make sure the Surveys tab is selected, it sits alongside Audits and Reviews at the top of the page.
3. Click any survey card to open its detail view.
4. Switch to the Results tab (next to Setup).

If the survey qualifies (see below), the AI Analysis panel appears near the top of the Results tab, right under the Responses received line and above the question-by-question breakdown.

When it appears

The AI Analysis panel only shows up when both of these are true:

• The survey is a full survey, meaning it was not sent as a Lightning pulse. A Lightning survey carries a small amber Lightning badge next to its type badge at the top of the detail view; a full survey does not. Lightning results show a simple per-person rating list and an aggregate distribution instead, there is no AI narrative on those, because a Lightning pulse is a single 1-5 tap, not a multi-question survey with room for a written synthesis.
• At least two teammates have responded. With zero responses, the Results tab shows the empty state below instead. With exactly one response, you see the per-question and per-person breakdown, but no AI panel, one respondent isn't enough for Verinode to describe a pattern versus an outlier.

What you see

Analyzing responses…, in italics, while Verinode is generating the summary. This usually takes a few seconds. Once it's ready, the loading line is replaced by the paragraph itself: typically three to four sentences covering team sentiment, any notable pattern (a clear consensus, or one respondent well off from the rest), and one concrete, actionable takeaway. It reads like a colleague reading the results over your shoulder, not a chart caption.

If the summary can't be generated (no eligible responses at the moment it runs, or a hiccup reaching the model), the panel simply doesn't render, there's no error banner, no placeholder. If you open the Results tab and see nothing between the response count and the question list, that's what happened, refreshing or re-opening the survey after a moment will usually produce it once responses are in.

How it works

Every time you open the Results tab on a qualifying survey, Verinode checks whether it already generated a summary in roughly the last hour. If it did, it hands you that cached version instantly, no extra wait, no extra cost. If the summary is missing or has gone stale, it reads through the responses on file for that survey, along with the survey's title, subject, and type, and asks its analysis model for a short synthesis: team sentiment, patterns worth flagging, and one thing to act on, capped under 100 words so it stays a summary and not a report.

The freshly generated paragraph is saved against the survey, so the next person who opens the same survey within that hour sees the same read rather than triggering a new one. As more responses come in, opening Results again after the cache has aged will produce an updated summary that reflects the fuller picture.

Generating (or regenerating) a summary draws a small amount of your monthly Intelligence Capacity, the same allotment your IQ chat conversations use. Viewing a cached summary you've already generated doesn't use any more of it.

How to use it

Read the AI Analysis paragraph first, it's built to tell you in one breath whether this survey came back the way you expected or surfaced something you need to look at. Then use it as a pointer into the detail below, not a substitute for it:

• If the summary calls out a split or an outlier, switch to the per-person view (the name pills above the question list) and find who's driving it. The individual view highlights any rating that sits meaningfully above or below the group average, so you can confirm the summary's read with the actual number.
• If the summary mentions a theme in the written answers, scroll to that question's block and read the quoted responses yourself. Verinode's job here is to point you at the right place fast, your judgment on what to do with a specific comment is still yours.
• Use Export CSV at the bottom of the detail view if you want the full response set, one row per respondent, in a spreadsheet, and (once the survey is closed) Export PDF for a shareable record.

Empty states

No responses at all. If nobody has responded yet, the Results tab shows: "No responses yet" with "Results will appear here once team members start responding." No AI panel, no question breakdown, nothing else renders until the first response lands.

One response. The Results tab shows the full question-by-question breakdown for that single respondent, but the AI Analysis panel does not appear. Wait for a second response, or nudge the remaining invitees, before expecting a synthesis.

Lightning surveys. These never show an AI Analysis panel, regardless of response count. Instead you get a per-person rating list sorted low to high, an average, and (once there's at least one rating) an aggregate 1-5 distribution.

Best-practice example

You send a Carrier Assessment to five teammates who work claims with a specific carrier. Three respond within a day. Opening the Results tab, the top line reads the response count and rate, then the AI Analysis panel loads: something like a read on overall satisfaction being mixed, one respondent flagging slow approvals while the other two are neutral to positive, and a suggested next step, perhaps raising the approval-turnaround pattern directly with your adjuster contact rather than waiting for the fourth and fifth response. You switch to that respondent's individual view, confirm their approval-process rating is well below the group average, and read their written comment on what specifically slowed the claim down. That's the summary doing its job: pointing you at the one thing worth a closer look, with the underlying data one click away to back it up.

Related articles

• The decision workspace
• Acting on decisions
• Clients and carriers
• Connecting your data

Data sources