# HUNCH: the probability that the answer is yes

`=HUNCH(A2, "Is the customer asking for their money back?")` returns the probability that the answer is yes, as a number from 0 to 1. In Excel the same function is `=HUNCH.ASK`. Point it at a range and it fills the whole column in one go.

Updated 28 September 2026. Source: https://hunchsheet.app/docs/hunch

## Syntax

```
Google Sheets: =HUNCH(text, question)
Excel:         =HUNCH.ASK(text, question)
```

| Argument | Required | What it is |
| --- | --- | --- |
| `text` | Yes | A cell, or a one-column range such as `A2:A500`, holding the text to judge. Blank cells come back blank and cost nothing. |
| `question` | Yes | A yes/no question in plain English with the definition inside it. It can be a cell reference or a string you build with `&`. |

The result is a number between 0 and 1: the probability that the answer to your question is yes for that text. A range in gives a column of numbers out, one per row, in the same order. In Excel the column spills down from the formula cell.

## Examples with real outputs

```
=HUNCH(A2:A7, "Is the customer asking for their money back?")
```
In Excel: `=HUNCH.ASK(A2:A7, "Is the customer asking for their money back?")`

| Message | Money back? (0–1) |
| --- | --- |
| Please cancel my plan and refund last month, I never used it. | 0.98 |
| Can you resend the invoice for March? Our accounting lost it. | 0.03 |
| The export keeps failing on files over 10 MB. | 0.05 |
| I was charged twice this month. Please reverse the second charge. | 0.94 |
| Honestly the price is steep for what we use, thinking of downgrading. | 0.08 |
| Love the new filters, saved me an hour today. | 0.02 |

The refund request and the double charge score 0.98 and 0.94. The invoice request is about money without asking for any back, so it gets 0.03. The price complaint scores 0.08 because the customer talks about downgrading and never asks for a refund. The wording of the question is what draws that line.

## How to word the question

- Put the definition in the question. "Is this lead a decision maker?" leaves the model to guess what you count. "Can this person approve a software purchase for their team without asking someone else?" does not.
- Ask one thing. "Is this urgent and about billing?" mixes two conditions into one number. Ask them separately, or use [HUNCHMULTI](https://hunchsheet.app/docs/hunchmulti) to get both columns from one read of the text.
- Phrase it so yes is the case you want to find. Sorting the column descending then puts those rows on top.
- Keep the text in the cell and the question in the formula. A question can point at a cell (`=HUNCH(A2, $B$1)`), which makes it easy to try versions side by side.

## Reading the number

A value near 1 means the model is confident the answer is yes, and near 0 that it is no. Values around 0.5 mean the text supports both readings or does not say. Those rows are the ones worth reading yourself.

Jev is trained so its probabilities are calibrated: across many rows that score about 0.8, roughly 8 in 10 should be yes. Check that on a sample of your own data before you set a threshold. Sort by the column, read the rows on each side of the cut, and move the cut until it matches your judgment.

```
=FILTER(A2:A500, HUNCH(A2:A500, "Is the customer asking for their money back?") >= 0.7)
```
In Excel: `=FILTER(A2:A500, HUNCH.ASK(A2:A500, "Is the customer asking for their money back?") >= 0.7)`

> Tip: When a column is final, freeze it. In Sheets use Extensions › Hunch › Freeze selection to values; in Excel copy the column and paste values. Frozen numbers never recalculate and never spend credits again.

## Cost and limits

- One credit per answered row. Blank cells and repeated texts inside the same request are free.
- Answers are cached for six hours, so reopening or recalculating the sheet does not spend credits.
- Rows travel in requests of 40. In Google Sheets one formula handles up to 1,500 unique texts within Google's 30-second limit on custom functions. For bigger columns use Extensions › Hunch › Ask about selection, which writes values.
- The model judges text. It does not do arithmetic, count items or compare dates, so keep those in normal formulas. English works best.

The messages a cell can show, and what to do about each, are on the [errors page](https://hunchsheet.app/docs/errors).

## Questions

### Is 0.8 the same as 80 percent accuracy?

No. 0.8 is the probability that the answer is yes for that one row. Accuracy depends on where you set your threshold and on your data. Score a sample, read it, then pick the cut.

### Can the question come from another cell?

Yes. `=HUNCH(A2, B$1)` reads the question from B1, and `=HUNCH(A2, "Is this about " & C$1 & "?")` builds it. Changing that cell re-asks every row, which spends credits for the new question.

### Why does the same text always get the same number?

The model gives a stable answer for the same text and question, and Hunch caches it for six hours. Change the wording of the question to get a new judgment.

### Does it read links, files or other sheets?

No. It judges the text in the cell and nothing else. Paste what matters into the cell, or join columns with `&` into one text.