ChatGPT
OpenAI’s CEO publicly stated that an average ChatGPT query uses about 0.34 Wh of energy and about 0.32 mL of water. We could not verify a comparable official OpenAI CO2-per-query figure.
Partial disclosure
Sustainability at Mentionpath
Honest about AI’s footprint. Practical about what we do next.
AI has a footprint. We’re not pretending otherwise.
Mentionpath uses AI to help brands understand how they appear across modern answer engines and AI search platforms. That creates real value, but it also has a real environmental cost. AI systems use electricity, consume water, and depend on data-center infrastructure whose impacts are not always transparently disclosed at the prompt level.
Public reporting is improving, but it is still inconsistent across providers. That is why our approach has three parts: acknowledge the footprint honestly, reduce avoidable waste where we can, and invest in nature-based solutions through Greenspark alongside those efforts.
We do not present nature-based solutions as a substitute for reducing unnecessary compute. We see them as one part of a broader responsibility model.
Our approach
Acknowledge
Be honest about AI’s cost.
Reduce
Avoid unnecessary compute and digital waste where practical.
Invest
Support nature-based solutions through Greenspark.
What we found
What is the footprint of AI search and answer platforms?
There is no single clean number. A few companies have published prompt-level figures. Most have not. In many cases, the public documentation explains how products search, reason, or use tools, but does not publish a directly comparable per-query energy or carbon figure. That matters because a plain text prompt is not the same as a search-heavy or research-heavy workflow.
Updated: April 2026
| Platform | What we could verify | Status | Source |
|---|---|---|---|
| ChatGPT | OpenAI’s CEO publicly stated that an average ChatGPT query uses about 0.34 Wh of energy and about 0.32 mL of water. We could not verify a comparable official OpenAI CO2-per-query figure. | Partial disclosure | |
 Gemini | Google published a prompt-level estimate for a median Gemini Apps text prompt of 0.24 Wh, 0.03 gCO2e, and 0.26 mL of water. | Official disclosure | |
 Google AI Overviews | Google documents that AI Overviews may use query fan-out and do not trigger on every search. We could not verify a separate public per-query footprint number for AI Overviews. | No public prompt-level disclosure found | |
| Google AI Mode | Google says AI Mode uses a custom Gemini model and query fan-out across related searches and data sources. We could not verify a separate public per-query footprint number for AI Mode. | No public prompt-level disclosure found | |
 Perplexity | Perplexity documents a difference between Fast Search and Pro Search, and also offers Deep Research across many sources. We could not verify a public per-query environmental disclosure. | No public prompt-level disclosure found | |
 Claude | Anthropic documents that web search can repeat multiple times in a single request and can be token-intensive. We could not verify a public per-query environmental disclosure. | No public prompt-level disclosure found | |
 Grok | xAI documents real-time web search, browsing, and multi-agent research. We could not verify a public per-query environmental disclosure. | No public prompt-level disclosure found | |
 DeepSeek | DeepSeek publicly documents reasoning mode and tool usage, but we could not verify an official public per-query energy or CO2 disclosure. | No public prompt-level disclosure found |
Gemini
Google published a prompt-level estimate for a median Gemini Apps text prompt of 0.24 Wh, 0.03 gCO2e, and 0.26 mL of water.
Official disclosure
Google AI Overviews
Google documents that AI Overviews may use query fan-out and do not trigger on every search. We could not verify a separate public per-query footprint number for AI Overviews.
No public prompt-level disclosure found
Google AI Mode
Google says AI Mode uses a custom Gemini model and query fan-out across related searches and data sources. We could not verify a separate public per-query footprint number for AI Mode.
No public prompt-level disclosure found
Perplexity
Perplexity documents a difference between Fast Search and Pro Search, and also offers Deep Research across many sources. We could not verify a public per-query environmental disclosure.
No public prompt-level disclosure found
Claude
Anthropic documents that web search can repeat multiple times in a single request and can be token-intensive. We could not verify a public per-query environmental disclosure.
No public prompt-level disclosure found
Grok
xAI documents real-time web search, browsing, and multi-agent research. We could not verify a public per-query environmental disclosure.
No public prompt-level disclosure found
DeepSeek
DeepSeek publicly documents reasoning mode and tool usage, but we could not verify an official public per-query energy or CO2 disclosure.
No public prompt-level disclosure found
The clearest public prompt-level numbers we could verify are from Google for Gemini, and more partially from OpenAI for ChatGPT energy and water. For most other platforms, the public record is mostly about product behavior rather than a directly comparable CO2-per-query figure.
Our view
Use AI, but be honest about the cost.
We do not think the credible answer is to ignore AI’s footprint. We also do not think the credible answer is to invent a fake precision number and pretend the problem is solved.
Our view is simpler than that. Mentionpath uses AI, so we acknowledge that cost directly. We aim to avoid unnecessary compute where we can. We keep our own digital experience lean where practical. And we invest in nature-based solutions through Greenspark alongside those efforts.
That means we treat sustainability as an operational discipline, not a marketing shortcut.
What we are saying
- AI has a real cost
- Disclosure is incomplete
- Reduction matters
- Restoration still plays a role
- Credibility comes from honesty
Leaner products
Lower-waste digital choices
We believe better digital products should also be leaner products.
For Mentionpath, that means aiming for efficient page design, thoughtful use of third-party scripts, optimized media and assets, and product choices that avoid unnecessary reruns or wasteful analysis patterns. We support dark mode across our site and product, as dark webpages use less energy. We see it as one small part of a broader effort to build a lighter digital experience.
Lighter pages
We try to keep our marketing site and product surfaces fast, clear, and free from unnecessary bloat.
Smarter compute
We avoid avoidable reruns, duplicate analysis, and wasteful workflows where practical.
Thoughtful tooling
We prefer product and infrastructure choices that support efficiency rather than constant always-on overhead.
Dark mode as a detail
We support dark mode, but we treat it as a supporting design choice.
Infrastructure choices
Optimising for more sustainable providers
Mentionpath is hosted on Google Cloud and deployed through Netlify. We frame this as a lower-waste infrastructure choice, not a blanket claim of “fully sustainable hosting.”
Google Cloud publishes region-level carbon data and has a stated goal to match its energy consumption with carbon-free energy every hour in every region by 2030. Google also says it has matched 100% of its global electricity use with renewable energy purchases since 2017, and reported that its data center energy emissions fell 12% in 2024 despite increased electricity consumption driven by business growth, including AI.
Netlify’s sustainability position focuses on efficient Jamstack-style delivery, serverless and event-driven computing, autoscaling, and the sustainability policies of its major cloud providers. We use that as part of a broader lower-waste delivery approach, rather than as a claim that the web has no footprint.
Google Cloud
Google operations
Netlify
Region-aware hosting choices
Efficient delivery patterns
Autoscaling over always-on overhead
Lower-waste framing, not zero-impact framing
How we invest
Nature-based solutions through Greenspark
Alongside our efforts to reduce unnecessary digital waste, we invest in nature-based solutions through Greenspark.
By nature-based solutions, we mean actions that protect, conserve, restore, and sustainably manage ecosystems while also supporting resilience, biodiversity, and human well-being. We do not treat this as a license to ignore the environmental cost of AI. We treat it as part of a broader approach: acknowledge the footprint, reduce what we can, and contribute to restoration alongside that.
Acknowledge
Reduce
Invest
How impact works in Mentionpath
How impact works in Mentionpath
We want our impact model tied to real customer actions, not random clicks. That is why Mentionpath funds impact through milestones that reflect real engagement and long-term use.
| Milestone | Trees (Greenspark) |
|---|---|
| Demo booked | 1 |
| Complete onboarding | 1 |
| Starter monthly plan | 1 |
| Pro monthly plan | 3 |
| Scale monthly plan | 10 |
| Starter annual plan | 2 |
| Pro annual plan | 6 |
| Scale annual plan | 20 |
| Referral | 3 |
| Partner signup | 5 |
Demo booked
1
Complete onboarding
1
Starter monthly plan
1
Pro monthly plan
3
Scale monthly plan
10
Starter annual plan
2
Pro annual plan
6
Scale annual plan
20
Referral
3
Partner signup
5
Why these milestones
Demo bookings reflect real intent
Onboarding reflects real adoption
Annual plans support long-term commitment
Referrals and partner signups reward trusted growth
Frequently asked questions
Quick answers about AI’s footprint, how we research providers, and how Greenspark fits our approach.
Why talk about AI’s footprint so directly?
Because it is real, and because public disclosure is still incomplete. We think acknowledging uncertainty is more credible than pretending there is a perfect, universal number.
Why don’t you publish one single CO2 number for all prompts?
Because the products are not directly comparable. A plain text prompt is different from a query that triggers web search, query fan-out, repeated tool calls, or deep research across many sources.
Are you claiming trees cancel out AI use?
No. We are saying two things at once: AI has a footprint, and we invest in nature-based solutions through Greenspark alongside our efforts to reduce unnecessary waste.
Why include nature-based solutions at all?
Because we want our sustainability work to include real-world ecological restoration alongside digital-efficiency efforts.
Does dark mode make AI sustainable?
No. Dark mode is a small design choice within a broader lower-waste product philosophy. It is not the core of the claim.
Will this page be updated?
Yes. We label the research section with an update date and aim to refresh sources as disclosures change.
Methodology
Methodology
This page is based on public provider disclosures, official product documentation, and a limited number of independent reference sources where provider-level prompt disclosures were unavailable.
- Prefer official company disclosures where available.
- Use official product documentation to understand how a system behaves when no environmental disclosure exists.
- Avoid fake precision where the data is incomplete.
- Keep restoration separate from reduction in how we explain sustainability.
Key references
Google Cloud
Sam Altman, The Gentle Singularity
Google Search Central
Google
Perplexity Docs
Anthropic Docs
xAI Docs
DeepSeek API Docs
Google Cloud
Google
Netlify
UNEP