Player insights

Find out why they got stuck, in their own words

A HedgeHunt is a trail of steps that sends people around real places on their phones. Eight separate plays stalled on step 2 of yours and you have no idea why. Ask for an analysis and you get the answers players actually gave, how many times each one came up, and one clear thing to change. Often the problem is a single word.

Part of Pro, and of a Hunt Pass for a single hunt. It needs five plays before it has anything to read, so start the hunt first.

The AI Insights tab in the builder: two steps players struggled with, each showing the number of sessions affected, a suggested edit, and the real answers players gave with counts
A real analysis of a Barcelona hunt. In the builder this is the AI Insights tab, inside Hunt activity.

How to trust it

The numbers are your players. The advice is AI.

These two jobs are kept apart on purpose, and it is the most important thing to understand about this feature.

Your players wrote the evidence

Every count and every quoted answer listed under a step is computed from real submissions by our servers. The AI is handed those counts and is never asked for a number of its own, so no figure in that evidence is its invention.

The AI writes three short things

A cause, in a few words. A sentence on what is going wrong. And one concrete edit to make. That is the whole of its job.

You make the call

It is advice, and it can be wrong. You have the real answers sitting underneath it, so you can check its reasoning yourself in about ten seconds.

Where it comes from

Every report says how many plays it read

Under the last step, every analysis tells you how many plays it read, when it sampled them, how many steps produced any evidence at all, and that the advice is AI while the responses and counts are your players' own. So you can always see the ground it is standing on.

The end of an insights report: two more struggling steps with the answers players gave, then a footer stating how many sessions it was based on and that the analysis is AI-generated advice
Two more struggling steps, and under them the lines that close every report.

What it reads

Eight ways a step goes wrong

Every stumble a player can have on a step, grouped and counted by how many separate sessions hit it.

Wrong answers

What they typed or picked instead, and how many said the same thing.

Written answers

Free writing, shown one by one, because no two people phrase it the same way.

Scans that did not match

QR codes players scanned that were not the one you meant.

Check-ins too far away

Plays that tried to check in outside the area you drew, with how near and how far they stopped when that can be measured.

Rejected photos

How many pictures the check turned down. The photos themselves never leave the session record.

Rejected recordings

The same count for voice notes on an audio mission, again without the recordings.

Problem reports

When a player tells you something is wrong, in their own words.

Skip requests

The moment somebody gave up and asked to move on.

The diagnosis

What it says is causing it

Each diagnosed step gets one of these, plus a sentence explaining it and one edit to make. Often the fix is a handful of words, though a blocked or unreachable spot is not a wording problem at all.

Prompt is ambiguous

Your wording can be read more than one way, and players took the other one.

Answer is too narrow

People knew it, but your accepted answer or your grading note is stricter than it needs to be.

Target is unobservable

You are asking about something they cannot actually see from where they stand.

Players may be blocked

Something in the real world is in the way, like a locked gate.

Where it stops

What it needs, and what it will not claim

One more thing it will not do: tell you that an edit of yours worked. It reads what players submitted, not the history of your edits, so that answer comes from the next report, after people have played the new version.

It needs five plays

Below five plays there is nothing solid to read, so it says so rather than guessing. You can run it 20 times an hour once you are over that line.

It covers at most five steps

Ranked by how many separate plays each one tripped up, and it skips any step it cannot genuinely explain. Some analyses come back saying no clear pattern.

It reads the last 90 days

Plays by people who were not signed in are kept for 90 days. A hunt that was busy last year has nothing left for it to read.

It runs when you ask

This is not a live dashboard. You press a button, it reads the recent plays, and it writes you a report.

How to get it

Pro

It comes with Pro, alongside the rest of the creator analytics: completion rates, per step difficulty, and every play in detail. If you only want it for one hunt, a Hunt Pass unlocks that hunt on its own as a one-time purchase, bought inside the builder when the hunt is ready. On the free plan you still get a preview of your hunt activity, just not the full analysis.

See plans and pricing →

Questions people ask

How much play does it need first?

Five play sessions, and one person who plays twice counts as two. Below that the analysis will not run, because a handful of attempts cannot tell a real pattern from bad luck. It reads the most recent 200 sessions.

Is this included in the free plan?

No. Player insights is part of Pro, and it is also included in a Hunt Pass if you want to unlock one hunt on its own. Access follows the hunt owner, so collaborators on a Pro owner’s hunt see it too.

Does the AI make up the numbers?

No, and it cannot. Every count and every quoted answer is calculated by our servers from real player submissions before the AI sees anything. The AI only writes the cause, the explanation and the suggested edit.

Will it tell me whether my fix worked?

No. It reads what players submitted, it never puts two versions of your hunt in time order, and it is instructed never to claim an edit fixed or improved anything. Where one step was played under two different wordings it can note that one is associated with more struggle, never that your change caused it. If you want to know, look at the next report after people have played the new version.

Does it show me everything players typed?

No. It covers at most the five worst steps, and up to five different answers per group, each with how many submissions matched it. That is enough to see the pattern without burying you in a list.

Do you show players’ personal details?

Email addresses and phone numbers are removed from quoted answers before anything is stored or sent to the AI. Beyond that, what a player types is what you see, so treat it as you would any message from them.

See the rest of creator analytics, where this sits in the whole flow, or the check you can run before you go live.

Build one, watch people play it, make it better

Start free, share your hunt, and let what players actually do tell you what to fix.

Create your hunt, free