# Now Live: Cross Check Verifies What AI Says About Your Brand

- Canonical: https://limelit.co/blog/now-live-cross-check-verifies-what-ai-says-about-your-brand
- Published: 2026-09-05
- Publisher: Limelit (https://limelit.co)

> Cross Check is live in Limelit: it extracts the factual claims AI answers make about your brand, verifies each against sources you control, and drafts corrections.

Cross Check is live for every Limelit account. It reads the AI answers Limelit already tracks, pulls out the factual claims they make about your brand, and verifies each one against a Knowledge Source you control. Every claim gets a verdict: accurate, inaccurate or unsupported, with a confidence and the exact source version pinned. Wrong claims become drafted corrections that a person reviews before anything ships.

## Why sentiment was not enough

Most AI visibility tools score tone. Positive, neutral, negative. That tells you how an engine feels about you and nothing about whether it is right. In our tracking, the claims that cost deals are rarely negative. They are confidently wrong: an old price, a feature you removed two years ago, an integration you never had, a competitor's capability attributed to you.

A tone score cannot catch those. A fact check can, but only against a source that says what is true today. That is why Cross Check starts with Knowledge Sources rather than with the answers.

## What it does

- **Knowledge Sources.** Add the pages that define your product: your pricing page, a fact sheet, a comparison page. Each source is versioned. When you refresh it, the new version is kept beside the old one, so a verdict always says which version it was checked against.
- **Claim extraction.** Cross Check reads the tracked answers that mention you and extracts the factual claims: prices, features, integrations, limits, comparisons.
- **Verification.** Each claim is checked against your sources and returns accurate, inaccurate or unsupported, with a confidence. Unsupported means your sources do not say, which is often the real finding: the engine is filling a gap you left.
- **Overrides.** If a verdict is wrong, you can override it and the override is recorded with the claim.
- **Corrections.** An inaccurate claim becomes a plan action: a drafted, human-reviewed correction aimed at the source the engine trusted. Limelit never publishes on its own.

## How to use it

1. Open Cross Check from the app and add two or three Knowledge Sources. Your pricing page is the one to start with; it is where engines go wrong most.
2. Run a check. Claims appear with their verdicts and the answer excerpt each one came from.
3. Work the inaccurate and unsupported list. Accept the drafted correction, edit it, or dismiss it.

Cross Check is also reachable from your own agent. The MCP server, free with every account, exposes the check, the claims, the verifications and the plan actions, so an agent in Claude or Cursor can run the loop and hand you the review.

## What it costs

Extracting and verifying claims uses model calls, which are metered in credits and shown before the run. A spend cap stops Limelit before it is exceeded. Adding and refreshing Knowledge Sources is free.

## FAQ

**What counts as a Knowledge Source?**
A page or document you control that states facts about your product. Pricing pages, fact sheets and feature pages are the usual ones. Each is versioned so a verdict is pinned to the exact text it was checked against.

**Does Cross Check change my content?**
No. It drafts a correction as a plan action for a person to review. Nothing is published without approval.

**How is this different from sentiment analysis?**
Sentiment scores tone. Cross Check tests truth. A claim can be warmly phrased and wrong; Cross Check catches that, a sentiment score does not.