> For the complete documentation index, see [llms.txt](https://docs.askyazi.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.askyazi.com/monitor-fieldwork/response-trust.md).

# Response Trust

Understand how Response Trust scores each response, set its sensitivity, and review, override or exclude flagged responses.

Response Trust checks every finished response for signs of low effort or fraud and gives it a verdict, with a written reason. It never stops anyone taking part. It helps you decide which responses to keep, which to look at, and which rewards to approve.

{% hint style="info" %}
**In short:** Response Trust is on by default for new studies at **Balanced** sensitivity. Set it in **Build > Study configuration > Response quality**. Each response gets a verdict such as **Approved**, **Flagged** or **Rejected** with a reason, shown on the Participants tab, in Table Data and in exports. You can override any verdict and add a note. **Auto-exclude rejected responses** removes Rejected participants from analytics.
{% endhint %}

## How it works

1. A participant finishes the study (or an activity).
2. Within about 5 to 10 minutes, Response Trust checks the response against its signals.
3. The response gets a verdict and a short reason, such as "All checks passed."
4. You see the verdict on the [Participants tab](https://docs.askyazi.com/monitor-fieldwork/participants-tab.md) (**Trust** and **Reason** columns), in Table Data (**Trust status** and **Trust note**), in the reward approvals list (as a **Trust score** tag) and in exports.

Until the check has run, a response shows **Not yet scored**.

<figure><img src="https://358667285-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FzMRQUXiGQlBwWdg6WEmG%2Fuploads%2FwaXi4DY8iXwX260TriXj%2Ftrust-column.webp?alt=media&amp;token=9b6ab905-760f-4461-93cd-cea04f4996a5" alt="The Participants table with the Trust column and the Survey done filter selected"><figcaption></figcaption></figure>

## Verdicts

| Verdict            | What it means                                        | What to do                                                             |
| ------------------ | ---------------------------------------------------- | ---------------------------------------------------------------------- |
| **Approved**       | No problems found                                    | Keep it                                                                |
| **Flagged**        | Something looks off. Worth a look.                   | Read the reason and the transcript, then keep it or change the verdict |
| **Rejected**       | Strong signs of a low-quality or fraudulent response | Usually exclude it and decline its reward                              |
| **Not yet scored** | The check has not run yet                            | Wait a few minutes                                                     |

## Settings

Open **Build > Study configuration > Settings > Response quality**.

* **Response Trust**: on or off. "Catch low-quality and fraudulent responses automatically."
* **Signals**: which checks run, shown as for example "5 of 5 active". Tick each one on or off:
  * **Response speed**: flags participants who finish much faster than the rest of the panel.
  * **Straight-lining**: catches people picking the same option again and again across multiple-choice questions.
  * **Duplicate images**: detects the same photo reused across several questions or sent in by different participants.
  * **Voice match**: spots the same voice answering under more than one participant ID.
  * **AI judge**: an AI model reads the questions and answers and flags nonsensical or inconsistent answers.
* **Sensitivity**:
  * **Lenient**: only flag the most obvious cases.
  * **Balanced**: recommended for most studies.
  * **Strict**: casts a wider net; expect more false positives.
* **Auto-exclude rejected responses**: automatically removes participants marked as Rejected from analytics.
* **Enable Question Actions**: switches on AI validation of media answers. This is a different feature, explained below.

<figure><img src="https://358667285-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FzMRQUXiGQlBwWdg6WEmG%2Fuploads%2FB5FlW1eC3zonmlJ9sgJV%2Fresponse-quality-panel.webp?alt=media&amp;token=5af14601-b1e8-492f-b24c-a3aaf128c1dd" alt="Response quality settings: Response Trust switched on with its five signals and the sensitivity choice"><figcaption></figcaption></figure>

New studies, duplicated studies and studies created with AI or Ask Yazi start with Response Trust on, all signals active and sensitivity at Balanced. Studies created before this default keep whatever they had.

## Review and override a verdict

A verdict is never final.

* On the Participants tab, change a person's verdict and add a note. The note appears under the AI's reason, with your name and the date.
* In Table Data, filter by trust status and update many rows at once with the bulk verdict control.

Use overrides when you know something the check does not, for example a short answer that is genuinely complete.

## Excluding responses from reporting

* Switch on **Auto-exclude rejected responses** to drop Rejected participants from charts and analytics automatically.
* There is no one-step option to leave Flagged responses out of exports. Review them and change the verdict to Rejected if you want them excluded, or filter by trust status in Table Data before you export.
* Exports include a readable trust status column (Approved, Flagged, Rejected, Not yet scored) with the reason, so you can apply your own rule.

## Response Trust and AI media validation

These two are often confused. They do different jobs.

|                            | Response Trust                            | AI media validation (**Enable Question Actions**)                             |
| -------------------------- | ----------------------------------------- | ----------------------------------------------------------------------------- |
| When it runs               | After the response is finished            | In real time, on one question                                                 |
| What it checks             | The whole response, against fixed signals | One answer (text, photo, video or voice note) against criteria you write      |
| What the participant sees  | Nothing                                   | A message asking them to redo the answer, or a screen-out                     |
| Can it stop a participant? | No                                        | Yes. It can hold them at the question or screen them out.                     |
| Where to set it            | **Response quality**                      | **Response quality > Enable Question Actions**, then criteria on the question |

AI media validation judges the audio of a video, not the picture. A participant rejected by it gets a rejection message and has no way back into the study, so keep criteria tight and test them on a small batch. See [Response quality and validation](https://docs.askyazi.com/build-a-study/response-quality-and-validation.md).

## How Yazi limits fraud

* Taking part needs a real WhatsApp account on a real phone number, which blocks most bot and duplicate-account fraud.
* Each WhatsApp number can only take part in a study once.
* Response Trust then checks what people actually answered.
* For high-risk studies, add evidence questions (for example a photo of a receipt or a shopfront) with AI media validation.

## Limits and things to know

* Response Trust only scores studies run fully inside Yazi. If participants are sent to another survey tool, no trust scores travel with them.
* Verdicts are a guide. A cautious, careful participant can be flagged and a clever cheat can pass. Read the reasons.
* Removing a participant also clears their trust score. If they join again, they are scored fresh.
* Response Trust uses AI. If your organisation has AI switched off, check with Yazi whether trust scoring runs.

## Related

* [Participants tab](https://docs.askyazi.com/monitor-fieldwork/participants-tab.md)
* [Response quality and validation](https://docs.askyazi.com/build-a-study/response-quality-and-validation.md)
* [Set up rewards](https://docs.askyazi.com/reach-participants/set-up-rewards.md)
* [Table Data](https://docs.askyazi.com/results-and-reporting/table-data.md)
* [Exports](https://docs.askyazi.com/results-and-reporting/exports.md)
