# How we check

Every card shows where its facts come from and when we last checked them. So you can check us yourself, not just trust us.

1. [See the common mistake.](#cards) On every card, you see what usually goes wrong before you rely on it.
2. [Open the sources.](#where) On every lesson and card, you can open the pages the facts come from and see when they were last checked.
3. [See how current it is.](#cadence) Each card shows when its sources were last checked. Spot a mistake? Tell us, and if we got it wrong, we fix the page.

## What every card gives you

- **The common mistake.** What usually goes wrong with this idea, named before you rely on it.
- **Sources.** Links to the pages each fact comes from: the publisher, then the page's exact title.
- **Last checked.** The date we last re-opened those sources and confirmed the card still matches them.
- **What changed.** Dated notes on how the card has changed, newest first. You'll find dated updates across the site on [What's new](/changelog).

Start here lessons give you sources and a last-checked date too.

## See it on a real card

Here is part of the [Hallucination](/hallucination) card. On the card itself, Last checked is at the top, What changed sits under the title, Common mistake comes right after How to apply, and Sources is in the side panel (below the card on a phone).

Last checked 2026-10-17

Example

Source: The warehouse closes at 18:00. Weekend hours are not listed.

Claim: "The warehouse closes at 18:00 and is open on Sunday." Not supported. The source does not mention Sunday. That claim is a hallucination.

**Common mistake:** Treating a confident or specific sentence as proof that the claim was in the source.

**Sources**: [Anthropic, Reduce hallucinations](https://platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/reduce-hallucinations) · [Gemini API, Safety and factuality](https://ai.google.dev/gemini-api/docs/safety-guidance) · [OpenAI, Optimizing LLM accuracy (LLM means large language model)](https://developers.openai.com/api/docs/guides/optimizing-llm-accuracy)

Want to run the same check on any AI answer? [Know which parts of an answer you can rely on](/start/check-the-output) shows you how.

## Where our facts come from

1. **First choice:** the maker's own pages (docs, help and privacy pages, release notes), and official guidance from public bodies such as NIST.
2. **Then:** research papers, with the lab or journal named.
3. **Then:** named press or independent evaluations. We say who found it.
4. **Never:** forums, social posts or AI-written summaries. If one appears, it's a labelled example, not a source.

Facts about a tool come from its maker's own pages, or a standards body. That covers what it can do, its limits, its settings and its privacy.

## No maker pays us

We're not paid by any maker. When we name a tool, it's an example of a kind of tool, not an endorsement.

## How current each card is

Each card shows the date its sources were last checked, so you can see how current it is. Start here lessons show it too.

## Found a mistake?

Email **[support@buildsparkai.com](mailto:support@buildsparkai.com)** with the page and the line. If you have a source, include the link. We check it against the source. If we got it wrong, we fix the page. On a card, the fix is noted under What changed.

## What the labels mean

- **Example:** we made it up to show the idea. It isn't real data or a real reply.
- **Not a live model / Simulated:** the reply in the demo was written by us or put together by our own code. No AI tool is running.

## Why we show cost in relative words, not amounts

You'll read "costs more than…" or "can be much cheaper when…", never a price. Prices change often and differ from plan to plan. The reason something costs more stays true much longer, and it's the part you can act on.

[Start here →](/start)