# Data-freshness tiers

Data-freshness tiers with caveat language

Also called agent data freshness.

**In short:** Add a “Data freshness” section saying how often volatile facts change, plus a “Never confirm …” rule for facts agents must not state as current.

- Category: Agent Readiness · weight 20 of 100 within the category
- Agent Readiness is worth 30 in technical, 25 in generic and commerce, 20 in regulated (points of 100, by scoring profile)
- Audit ID A2 (as shown on reports) · rubric v2.1

## What it checks

Whether volatile facts are labelled with how fresh they are, and how to caveat them.

## Why it matters

Prices, stock and rates change. Without freshness tiers an agent repeats yesterday’s number as fact; with them it knows to caveat or to check.

## When it applies

Applies when the file itself suggests volatile data: freshness wording, a never-confirm rule, or a documented API (wording such as “endpoint”, “OpenAPI”, “API reference”, GET / or POST /). So a file that documents an API but says nothing about freshness is scored, and scores 0. Otherwise it’s Not applicable.

An audit that doesn’t apply, or that we couldn’t verify, is removed from your score and its weight shared among the Agent Readiness audits that do apply. It is never scored as a zero.

## How it’s scored

0.7 for data-freshness wording (“data freshness”, “freshness tier”, “near real-time”, “content volatility” or “volatile”) · 1.0 when the file also has a never-confirm rule (“never confirm”, “do not confirm”, “must not state / assert / confirm”) · 0 with no freshness wording.

> **Measurement.** Today wording stands in for the tiers themselves: which kinds of volatile data you cover, and how specific their caveats are, isn’t read. Scoring coverage of volatile data classes × caveat specificity is planned.

## How to fix

1. Add a section headed “Data freshness” listing each kind of volatile data and how often it changes.
2. For each, give caveat language the agent should use, and where to check the live value.
3. Say plainly which facts an agent must never confirm on your behalf, in those words: “Never confirm …”.

## Example

```
## Data freshness

- Prices: change weekly. Say “prices from our site as of <date>”; confirm at https://acme.com/pricing
- Never confirm a price, fee or availability as current; link the page that states it.
```

## Frequently asked questions

### What are data-freshness tiers, and what should they say?

A list of each kind of volatile fact on your site (prices, stock, rates, schedules), how often it changes, the caveat an agent should use when repeating it, and where to check the live value. For example: “Prices: change weekly. Say ‘prices from our site as of <date>’; confirm at https://acme.com/pricing”. Add a plain line naming anything an agent must never confirm on your behalf.

Sources: [LLMSRank — Data-freshness tiers audit](https://llmsrank.dev/docs/audits/data-freshness-tiers.md)

### How is the data-freshness check scored?

Today the scanner scores freshness wording: phrases such as “data freshness”, “freshness tier”, “near real-time” or “volatile/volatility” earn 0.7, and adding a never-confirm rule (“never confirm”, “do not confirm”, “must not state”) lifts it to 1.0. A file that qualifies but has no freshness wording scores 0.

Sources: [LLMSRank — Data-freshness tiers audit](https://llmsrank.dev/docs/audits/data-freshness-tiers.md) · [LLMSRank — How scoring works](https://llmsrank.dev/docs/method.md)

### When is the data-freshness check Not applicable?

The rubric applies it only to sites with volatile data; if every fact is stable there are no tiers to declare and its weight moves to the other agent-readiness checks. Today the scanner infers volatile data from the file itself: freshness wording, a never-confirm rule, or a documented API (an endpoint, OpenAPI spec, `GET /` or `POST /`). So a file that documents an API but says nothing about freshness is scored, and scores 0.

Sources: [LLMSRank — Data-freshness tiers audit](https://llmsrank.dev/docs/audits/data-freshness-tiers.md) · [LLMSRank — How scoring works](https://llmsrank.dev/docs/method.md)

### Should I put live prices or stock levels in llms.txt itself?

No. llms.txt v2 says the file should stay small enough to fit in context, with the detail behind links fetched only when needed. Link the page or API that has the live value, and use the freshness tier to tell the agent how to caveat it.

Sources: [llmstxt.org — The /llms.txt file (v2, 2026-08-10)](https://llmstxt.org/) · [LLMSRank — Data-freshness tiers audit](https://llmsrank.dev/docs/audits/data-freshness-tiers.md)

### In our research, how many sites label how fresh their data is?

Almost none. In our 187-site corpus (captured 2026-10-10, scored with rubric v2.1) the check applied to 81 files and was Not applicable to 106; of the 81, only 2 earned full marks and 84.0% scored zero, with no freshness wording at all.

Sources: [LLMSRank — How scoring works](https://llmsrank.dev/docs/method.md) · [LLMSRank — Data-freshness tiers audit](https://llmsrank.dev/docs/audits/data-freshness-tiers.md)

See also: [all audits](https://llmsrank.dev/docs/audits.md) · [how scoring works](https://llmsrank.dev/docs/method.md)