## Why It Matters

An insight reads **Based in Poznań: No** on a role that requires being in the office. The AI is right — the resume says Kraków. It still hasn't told you the thing you need to know, which is whether this candidate would move.

Until now your only options were to guess, to overwrite the field with your own assumption, or to write the candidate a one-off email and keep the answer in your inbox. Asking through Kit gets you "No — but I'm moving there in Q1", stored in the field, visible to everyone reviewing the application. No amount of re-extraction could have produced that sentence, because it isn't in the resume.

## How It Works

From the Candidate Insights panel on an application, you pick the fields you want confirmed, say why each one matters, and send one email. The candidate opens a link that shows only those fields, answers each one, and their answer writes back into the field marked as coming from them.

The round trip is four steps:

1. **Pick fields** — any insight field visible to you, whether it's filled in or empty
2. **Give a reason per field** — the candidate sees this, so it explains rather than interrogates
3. **Send one email** — one message covering every field you picked, not one per field
4. **The answer lands in the field** — tagged **Candidate**, and safe from future re-extraction

## Sending a Request

Open the application and find the **Candidate Insights** panel. In its menu, choose **Ask candidate to clarify…**. The **Ask the candidate to clarify** dialog opens.

Under **What should the candidate confirm?** you get one row per insight field. Check the ones you want answered. Empty fields read as "please fill this in"; filled fields read as "please confirm or correct this". Both are fair to ask about — a confidently wrong value is worth more of your attention than a blank one.

Checking a row reveals a **Why you're asking** box, prefilled with wording that fits the field's type. Rewrite it in your own words whenever the default is too generic — this is the sentence that decides whether the candidate feels consulted or audited. **Suggest reasons** hands the checked fields to AI and replaces the boxes with tailored wording you can still edit.

Two optional message boxes sit below the fields:

| Box | Who sees it |
|-----|-------------|
| **Message to the candidate** | The candidate, above the questions |
| **Internal note** | Your team only — saved to Notes, never sent |

**Send request** mails the candidate.

> [!WARNING]
> The **Internal note** is never sent to the candidate. The **Message to the candidate** always is. Both boxes look the same in the dialog, so read the label before you type anything you would not want forwarded.

Fields restricted to managers are never offered in this dialog, so a field your wider team cannot see can't reach a candidate-facing email either. See [Candidate Data Fields](/docs/candidate-data-fields) for how field visibility is set.

## What the Candidate Sees

The link opens a short page headed **A few quick questions about your application** — only the fields you asked about, nothing else from their application, and no login.

Each field shows its label, your **Why we're asking** line, a control matching the field's type, and a free-text box:

| Field type | Control |
|------------|---------|
| Boolean | Yes / No / **Not sure** |
| Select | The field's own choices |
| Rating | A 1-to-max scale |
| Tags | Comma-separated text |
| Number, date, URL, text | The matching input |

The free-text box, **Anything we should know?**, is always there under **Your answer**. That box is where the Poznań answer actually lives: the candidate picks **Not sure**, which leaves the value alone, and types "moving there in Q1". You get the nuance without a wrong value being written.

**Send my answers** submits everything at once.

## Declining Is Allowed

The page says so plainly, and **I'd rather not answer** is a real button. A candidate who uses it has their application untouched: no flag, no "declined" badge on your side, no effect on how they're evaluated. You simply see the request close without answers.

This is deliberate. A clarification request is a question, not a stage — treat a non-answer as one.

## Where the Answer Lands

An answered field shows a **Candidate** source badge instead of the AI one, and the value is the candidate's.

The badge is not decoration. AI re-extraction will never overwrite a candidate-sourced value, exactly as it never overwrites one your team typed by hand. Running **Re-extract** after a candidate answers is safe. See [AI Extraction for Candidate Data](/docs/ai-extraction) for what re-extraction does touch.

The value the AI originally found is kept on the record, so you can still see what it read before the candidate corrected it.

## Tracking a Request

Three places tell you where a request stands:

| Where | What it shows |
|-------|---------------|
| Candidate Insights panel | An **Asked** chip on pending fields, **Answered** once the answer arrives, and **No answer yet** in place of a value |
| Notes | Your internal note, plus who sent the request and when |
| The email thread | The exact message the candidate received |

You also get a notification when the answers come in, so you don't have to keep checking the panel. Everything is visible to anyone who can open the application — see [Reviewing Applications](/docs/reviewing-applications).

## Links, Reminders, and Resending

The link in the email works for **14 days**. If the candidate hasn't answered after **5 days**, Kit sends one reminder — one, not a sequence.

Only one request can be open per application at a time. Try to send a second and Kit tells you there's already an open request and to cancel it first. That limit is what stops a candidate collecting five emails from three teammates about the same application.

Two controls handle the rest:

- **Resend email** — same questions, fresh link, for a candidate who lost the email
- **Cancel request** — closes the round and stops the emailed link from working

> [!NOTE]
> Answering, declining, cancelling, or resending all invalidate any link sent earlier. Only the most recent email in a round has a working link.

## Quick Checklist

- [ ] Open the application and use **Ask candidate to clarify…** in the Candidate Insights menu
- [ ] Pick only the fields whose answers would change a decision
- [ ] Rewrite each **Why you're asking** line so it reads as a question, not an audit
- [ ] Check that anything sensitive went in **Internal note**, not **Message to the candidate**
- [ ] Expect the answer within 14 days, with one reminder at day 5
- [ ] Confirm the answered field now carries the **Candidate** badge
- [ ] Treat **I'd rather not answer** as a neutral outcome, not a signal