> ## Documentation Index
> Fetch the complete documentation index at: https://slatehq.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Fact check

> Compare what AI answers say about your brand with your brand facts, find the claims they get wrong, and see which pages those claims come from.

> [!NOTE]
>
> Fact check is available to selected workspaces. Ask your Slate contact to turn it on for your workspace.

Fact check compares what AI answers say about your brand with the brand facts you add, such as your prices and features. It lists the claims AI gets wrong and the pages AI cited next to them.

## Why It Matters

A buyer who asks an AI assistant about your pricing can get an old price or a plan you no longer sell. Fact check finds those claims and their sources, so you know which page to fix.

## Where to Find It

Open [AI Visibility](https://slatehq.com/docs/ai-tracker/overview.md#where-to-find-it), then select **Fact check** under **Brand**.

Fact check uses your Brand Kit. If your brand or competitors aren't set up yet, the page shows **Unlock your brand insights**. See [Set up AI Visibility](https://slatehq.com/docs/ai-tracker/setup.md).

Until your workspace has its first AI answers, the page shows read-only sample data for a made-up brand, marked **Sample data**.

## Key Terms

* **Brand fact**: a one-line statement about your brand, such as "The Pro plan costs \$49 a month." Facts live in **Brand Kit → Brand Facts**. Only facts that are on get checked.
* **Claim**: something an AI answer says about your brand. One answer can make several claims.
* **Verdict**: **Wrong** (one of your facts says the claim is wrong), **Correct** (the claim matches a fact), or **Not covered** (no fact speaks to the claim).

## Set Up Fact Check

Before the first results, the page shows three setup steps:

1. **Add your brand facts**: click **Add facts** to write facts or upload a CSV file. To draft facts from your website, click **Draft from website** in **Brand Kit → Brand Facts**.
2. **Choose prompts to check**: click **Choose prompts**. See [Choose Prompts](#choose-prompts).
3. **Read the results**: claims show up after the chosen prompts run again.

The page then shows **Waiting for the next run**, with the next run's time in UTC. Only new answers are checked. Claims show a few minutes after a run ends.

## Choose Prompts

In the **Choose prompts to fact-check** window, tick the prompts whose answers you want checked, then click **Save**, for example **Save 12 prompts**. The tabs are **Suggested** (when available), **All active**, and **Selected**.

* Only active, recurring prompts can be checked. One-time and paused prompts don't run again.
* Each answer costs 1 credit, charged only after the check completes and results are saved.

To change the list later, click **Choose prompts** next to the page title.

## Read the Results

Three numbers sit at the top. They follow your filters.

| Number | What it means |
| - | - |
| **Wrong claims** | Different claims that one of your facts says is wrong. A claim repeated in many answers counts once. |
| **Accuracy** | Correct claims ÷ (correct + wrong claims). Not covered claims and claims marked **Not relevant** aren't counted. |
| **Not covered** | Claims about your brand that none of your facts speak to. Click **Add facts** to cover them. |

When a comparison is on, each number shows its change. If some runs aren't checked yet, a line above the numbers says why, for example that you were out of credits.

**Accuracy over time**, at the bottom, shows the share of checked claims that were correct, per day or week.

## The Claims Card

The **Claims** card has three tabs: **Wrong**, **By fact**, and **Not covered**. Use the search box to find a claim or a fact.

### Wrong

Wrong claims, grouped by their fact's category.

| Column | What it shows |
| - | - |
| **Claim** | What the AI answer said, the fact it breaks (**Breaks:**), and its reach, for example "In 9 answers · 3 prompts · ChatGPT, Perplexity · last seen 26 Oct". |
| **Source** | The site cited most often next to the claim in the latest answer, with its type, such as **Our page** or **Third-party**. **+N more** counts the other cited sites. |
| **Status** | **Open** until you resolve the claim, then **Resolved**. |

### By fact

One row per fact, with **Category**, **Wrong**, **Correct**, **Accuracy**, and **Last seen**. Click **Wrong**, **Correct**, **Accuracy**, or **Last seen** to sort.

* Click a fact to see its wrong claims on the **Wrong** tab.
* **No matching claim found in checked answers** means no checked answer mentions the fact. Click **Review prompts** to check your prompts.
* **Edit facts** opens your facts in the Brand Kit.

### Not covered

Claims none of your facts speak to:

* **Add a fact** opens **Add facts** with the claim filled in. The claim is checked from the next run.
* **Not relevant** leaves the claim out. Click **Undo** in the message to bring it back.

## Open a Claim

Click a claim to open its panel:

* **The fact it breaks** (or **The fact it matches**), with **Edit fact**. If the fact changed after the check, the panel says **Fact edited since this check**.
* **In the answers**: the latest two answers that made the claim. Click **Open answer** to read one, or **View all** for every answer.
* **Where AI got it** (wrong claims only): the pages cited with the claim, closest first. Click a page to see what it says. **No source cited** means the claim most likely comes from older training data.

### Correct the checker

Click the **⋯** menu at the top of the panel:

* **Accept as correct**: the wrong claim is right after all. **Remove "Accept as correct"** puts the checker's verdict back.
* **Not relevant**: leave the claim out. **Mark as relevant** brings it back.
* **Edit the fact**: open the fact in Brand Facts.
* **Resolve**: mark the claim as handled once the page is fixed. **Reopen** sets it back to **Open**, and so can a new wrong answer.

The message after a change may offer **Undo**.

## Filters

The filter row has **Platform**, **Topic**, **Prompt**, **Location**, **Period**, and **Compare to**.

* Pick one platform or **All Platforms**, and one topic or **All Topics**. If several are selected, the page asks you to **Pick one platform and one topic**.
* **Prompt** lists only the prompts Fact check checks.
* **Period**: **Last 7 days**, **Last 30 days** (default), **Last 3 months**, **Last 6 months**, **Last year**, or **Custom**.
* **Compare to**: **Off**, **Previous period** (default), **Same period last year**, or **Custom**.

## When the Page Shows a Message

* **Fact check isn't available for this workspace yet.**: ask your Slate contact to turn it on.
* **Fact check isn't on the Launch plan.**: click **Upgrade plan**.
* **No wrong claims in this period.**: try a longer period or fewer filters.

## Tips

* Price and comparison prompts work best for Fact check.
* When a wrong claim's source is **Our page**, fix that page first.
* Add facts for the **Not covered** claims buyers care about.

## What's Next

* [Objections](https://slatehq.com/docs/ai-tracker/objections.md): the doubts and complaints AI answers raise about your brand
* [Sentiment](https://slatehq.com/docs/ai-tracker/sentiment-analysis.md): positive, neutral, and negative statements about your brand
* [Brand Kit](https://slatehq.com/docs/brandkit.md): your brand, competitors, and brand facts
* [Answers](https://slatehq.com/docs/ai-tracker/prompt-history.md): the full AI answers behind each claim


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.