How is AI usage accounted for in Trace?
AI tool subscriptions are accounted for under Scope 3 Category 1, using a spend-based "software subscriptions" emissions factor that's applied automatically to your AI-related spend.
If your team pays for ChatGPT, Claude, Gemini, or a similar tool, that spend is already being captured in your Trace inventory. You don't need to add anything separately, and there's no dedicated "AI" line in the platform, because it isn't needed for this to be reflected in your footprint.
Where does AI fit in the GHG Protocol?
AI subscriptions are a purchased service, so they fall under Scope 3 Category 1 (Purchased Goods and Services), alongside your other software and SaaS spend. The GHG Protocol Scope 3 Standard sets out a hierarchy of calculation methods here: supplier-specific data, hybrid, and average-data (spend-based). Spend-based, using environmentally-extended input-output (EEIO) factors, is a recognised and legitimate method, particularly where supplier-specific data isn't available. That's the situation with most AI providers today.
How Trace applies this
Trace applies a spend-based "software subscriptions" emissions factor to your AI platform spend. This factor is drawn from sector-average data and already reflects typical software and cloud emissions, including AI, at an economy-wide average level. It's applied automatically wherever AI-related spend is coded to a software or subscriptions cost centre, the same way any other software spend is treated.
Why we don't currently use a provider-specific factor
Although we'd prefer a more precise number, many providers don't yet publish their emissions publicly. Provider disclosure is inconsistent (some AI providers have published detailed methodology, others have said very little), and even where a customer's spend is concentrated in a single provider, there's no credible official figure to substitute in. As with other spend categories, we therefore default to verified EEIO spend-based factors. It's worth noting that spend-based factors do lag real-world activity (sometimes by a year or more) which does mean that the full extent of AI-related software emissions today may be underreported. Once validated datasets "catch up", however, AI emissions will be more accurately reflected. This is a known and widely accepted gap since AI is moving so quickly (and, therefore, the associated emissions generated).
If you want to understand what is and isn't known about each provider's environmental impact, and how to build your own more detailed estimate, see Understanding the environmental impact of your AI usage (detailed estimation guide)
Reference table
| Scenario | Category | Trace treatment |
|---|---|---|
| Direct subscription to ChatGPT, Claude, Gemini, or similar | Scope 3, Category 1 | Spend-based "software subscriptions" factor, applied automatically |
| AI features bundled into a broader SaaS platform (e.g. Microsoft 365 Copilot, Salesforce Einstein) | Scope 3, Category 1 | Spend-based "software subscriptions" factor (spend isn't separable by vendor) |
| Free-tier or unpaid AI tool usage | Not currently captured | No spend line exists, so the spend-based method doesn't apply. See our detailed estimation article if this is material to you. |
| Cloud compute for running your own AI models | Scope 3, Category 1, or Scope 2 if self-hosted | Spend-based factor, or actual utility data if self-hosted |
Key points
- Free-tier usage that's material isn't captured by the spend-based method at all, since there's no spend to apply a factor to.
- This is a fast-moving area. As providers publish more verified data, Trace will look to introduce more precise, provider-specific factors.
AASB S2 Considerations
Where AI spend is a material part of your Category 1 emissions, AASB S2's measurement uncertainty requirements apply. You don't need a bespoke methodology to comply, just the ability to describe your estimation approach (spend-based, EEIO-derived) and acknowledge the limitation created by the lack of provider-specific data. This is consistent with how AASB S2 expects entities to disclose the basis and limitations of Scope 3 estimates generally.