{
  "model": {
    "id": "dttt-environmental",
    "name": "DTTT AI Environmental Impact Model",
    "version": "0.6",
    "release_date": "2026-03-16",
    "release_type": "minor",
    "status": "published",
    "description": "A framework for assessing and communicating the indicative environmental impact of AI use across advisory, marketing and operational work. Applicable to any organisation or individual using general-purpose AI tools. Version 0.5 introduces the A\u2013E grading scale, three-axis scoring logic and updated methodology sources. Version 1.0 will mark the first stable release following committee review and methodology verification.",
    "url": "https://www.thinkdigital.travel/ai-transparency/environmental/index-html",
    "data_confidence": "Estimated",
    "disclaimer": "Energy and carbon estimates for AI tool usage are indicative and based on the best available published research. Precise per-model, per-query figures are not yet available for most closed AI systems. Scores produced by this model are strategic disclosure tools, not audited environmental accounts. DTTT will update methodology sources at each version as first-party disclosure from AI companies improves, including in response to requirements under the EU AI Act."
  },
  "scope": {
    "intended_users": [
      "DTTT team members assessing advisory and research work",
      "Destination management organisations assessing internal AI workflows",
      "DMO marketing and content teams assessing campaign and content production",
      "Any organisation or individual using general-purpose AI tools who wishes to disclose environmental impact"
    ],
    "applicable_outputs": [
      "Advisory reports and strategy documents",
      "Marketing campaigns and content",
      "Articles, blogs and editorial",
      "Social media content",
      "AI-generated images and visual assets",
      "AI-generated video and multimedia",
      "Internal workflow and operational AI use",
      "Research and data analysis"
    ],
    "note": "This model assesses what was done and how intensively, not who did it or for what purpose. It applies equally to DTTT's own work and to the independent work of member organisations and other users."
  },
  "scoring_axes": {
    "description": "Every assessment is based on the self-reported combination of three factors. Where a session involved multiple task types, score based on the highest-intensity task used.",
    "axis_1": {
      "id": "task_type",
      "name": "Task type",
      "description": "What the AI was asked to do.",
      "levels": [
        {
          "id": "text",
          "label": "Text",
          "description": "Drafting, research, analysis, editing, brainstorming, translation or code generation.",
          "plain_language": "Writing, researching, editing or analysing text"
        },
        {
          "id": "image",
          "label": "Image generation",
          "description": "AI generation of images, illustrations or visual assets of any kind.",
          "plain_language": "Creating images or visual assets using AI"
        },
        {
          "id": "extended_reasoning",
          "label": "Extended reasoning or agentic",
          "description": "Long chain-of-thought tasks, deep research modes, multi-step autonomous workflows or extended iterative sessions requiring sustained compute.",
          "plain_language": "Using an AI tool's advanced research or reasoning mode, or running multi-step automated workflows"
        },
        {
          "id": "video",
          "label": "Video generation",
          "description": "AI generation of video content of any kind.",
          "plain_language": "Creating video content using AI"
        }
      ]
    },
    "axis_2": {
      "id": "model_category",
      "name": "Model category",
      "description": "The type of AI tool used.",
      "levels": [
        {
          "id": "on_device",
          "label": "On-device or lightweight",
          "description": "Local inference, small open-source models, on-device assistants, lightweight writing or editing tools running on modest infrastructure.",
          "plain_language": "A lightweight or on-device tool (such as on-device grammar checking or a small local model)"
        },
        {
          "id": "standard_cloud",
          "label": "Standard cloud",
          "description": "General-purpose cloud AI assistants at typical usage levels. Includes everyday text-based use of ChatGPT, Claude, Gemini and equivalent tools for standard tasks.",
          "plain_language": "A general AI assistant such as ChatGPT, Claude or Gemini used for everyday tasks"
        },
        {
          "id": "frontier",
          "label": "Frontier or large",
          "description": "The most capable and largest available models, extended context sessions, advanced reasoning or deep research modes, and high-compute specialised tools.",
          "plain_language": "An advanced or specialist AI tool, or using a general assistant in its most powerful mode (such as deep research or extended reasoning)"
        }
      ]
    },
    "axis_3": {
      "id": "usage_intensity",
      "name": "Usage intensity",
      "description": "How much and how continuously AI was used.",
      "levels": [
        {
          "id": "light",
          "label": "Light",
          "description": "A small number of discrete queries for a single task. Under roughly 30 minutes of active AI use.",
          "plain_language": "A small number of queries for one task, or a brief session"
        },
        {
          "id": "moderate",
          "label": "Moderate",
          "description": "Regular use across a working session, covering multiple tasks. Roughly one to two hours of active AI use.",
          "plain_language": "Used across a working session for several tasks"
        },
        {
          "id": "heavy",
          "label": "Heavy or continuous",
          "description": "AI used throughout a working day or across an extended project. Long iterative sessions, high query volume, or continuous use as a primary work tool.",
          "plain_language": "Used continuously or throughout most of a working day or project"
        }
      ]
    }
  },
  "scoring_logic": {
    "step_1": {
      "description": "Determine base grade from the combination of task type (Axis 1) and model category (Axis 2).",
      "base_grade_grid": [
        {
          "task_type": "text",
          "on_device": "A",
          "standard_cloud": "A",
          "frontier": "B"
        },
        {
          "task_type": "extended_reasoning",
          "on_device": "B",
          "standard_cloud": "C",
          "frontier": "D"
        },
        {
          "task_type": "image",
          "on_device": "B",
          "standard_cloud": "C",
          "frontier": "D"
        },
        {
          "task_type": "video",
          "on_device": "D",
          "standard_cloud": "D",
          "frontier": "E"
        }
      ]
    },
    "step_2": {
      "description": "Apply the usage intensity modifier (Axis 3) to the base grade.",
      "modifiers": [
        {
          "intensity": "light",
          "effect": "Move one grade down (minimum A)",
          "rationale": "Brief or infrequent use substantially reduces session-level energy consumption relative to moderate baseline use."
        },
        {
          "intensity": "moderate",
          "effect": "No change. Base grade applies.",
          "rationale": "Moderate use represents the baseline assumption for the base grade grid."
        },
        {
          "intensity": "heavy",
          "effect": "Move one grade up (maximum E)",
          "rationale": "Continuous or high-volume use substantially increases session-level energy consumption relative to moderate baseline use."
        }
      ]
    },
    "multiple_task_types": "Where a session involved more than one task type, assess based on the highest-intensity task used. Record all task types in the disclosure."
  },
  "example_assessments": [
    {
      "scenario": "Brief text query to on-device tool",
      "task_type": "text",
      "model_category": "on_device",
      "intensity": "light",
      "base_grade": "A",
      "final_grade": "A"
    },
    {
      "scenario": "Standard ChatGPT or Claude session, single text task",
      "task_type": "text",
      "model_category": "standard_cloud",
      "intensity": "light",
      "base_grade": "A",
      "final_grade": "A"
    },
    {
      "scenario": "Standard ChatGPT or Claude, full working session across multiple text tasks",
      "task_type": "text",
      "model_category": "standard_cloud",
      "intensity": "heavy",
      "base_grade": "A",
      "final_grade": "B"
    },
    {
      "scenario": "Frontier model (e.g. Claude, GPT-4o) used for text tasks, light use",
      "task_type": "text",
      "model_category": "frontier",
      "intensity": "light",
      "base_grade": "B",
      "final_grade": "A"
    },
    {
      "scenario": "Frontier model, text tasks, moderate session",
      "task_type": "text",
      "model_category": "frontier",
      "intensity": "moderate",
      "base_grade": "B",
      "final_grade": "B"
    },
    {
      "scenario": "Frontier model, continuous daily use across a project (e.g. DTTT advisory work)",
      "task_type": "text",
      "model_category": "frontier",
      "intensity": "heavy",
      "base_grade": "B",
      "final_grade": "C"
    },
    {
      "scenario": "Image generation, standard tool, one-off asset",
      "task_type": "image",
      "model_category": "standard_cloud",
      "intensity": "light",
      "base_grade": "C",
      "final_grade": "B"
    },
    {
      "scenario": "Image generation, frontier tool, campaign production at moderate volume",
      "task_type": "image",
      "model_category": "frontier",
      "intensity": "moderate",
      "base_grade": "D",
      "final_grade": "D"
    },
    {
      "scenario": "Deep research or extended reasoning mode, frontier model, moderate use",
      "task_type": "extended_reasoning",
      "model_category": "frontier",
      "intensity": "moderate",
      "base_grade": "D",
      "final_grade": "D"
    },
    {
      "scenario": "Deep research or extended reasoning mode, frontier model, heavy use",
      "task_type": "extended_reasoning",
      "model_category": "frontier",
      "intensity": "heavy",
      "base_grade": "D",
      "final_grade": "E"
    },
    {
      "scenario": "Video generation, standard cloud tool, moderate production",
      "task_type": "video",
      "model_category": "standard_cloud",
      "intensity": "moderate",
      "base_grade": "D",
      "final_grade": "D"
    },
    {
      "scenario": "Video generation, frontier model, heavy campaign production",
      "task_type": "video",
      "model_category": "frontier",
      "intensity": "heavy",
      "base_grade": "E",
      "final_grade": "E"
    }
  ],
  "grades": [
    {
      "grade": "A",
      "label": "Low energy intensity",
      "colour_hex": "#00703c",
      "colour_name": "Green",
      "definition": "Low energy use at session level. Disclose as part of standard transparency reporting.",
      "typical_use": [
        "A short research or drafting session using a general AI assistant",
        "On-device grammar or writing tools used throughout a working day",
        "A single brief query to any text-based AI tool"
      ]
    },
    {
      "grade": "B",
      "label": "Low-moderate energy intensity",
      "colour_hex": "#5a8a3c",
      "colour_name": "Mid green",
      "definition": "Low to moderate energy use. Disclose as part of standard transparency reporting.",
      "typical_use": [
        "A full working day using Claude or ChatGPT for text-based advisory work",
        "Occasional AI image generation using a standard cloud tool",
        "A frontier model used lightly for a single text task"
      ]
    },
    {
      "grade": "C",
      "label": "Moderate energy intensity",
      "colour_hex": "#f4a100",
      "colour_name": "Amber",
      "definition": "Moderate energy use. Disclose in deliverable transparency notes.",
      "typical_use": [
        "Continuous daily use of a frontier model (such as Claude or GPT-4o) across a multi-day project",
        "A marketing campaign using standard cloud image generation at moderate volume",
        "Light use of a deep research or extended reasoning mode"
      ]
    },
    {
      "grade": "D",
      "label": "High energy intensity",
      "colour_hex": "#e05206",
      "colour_name": "Orange",
      "definition": "High energy use. Disclose and consider lower-intensity alternatives for future work.",
      "typical_use": [
        "A campaign using AI image generation via a frontier model",
        "Extended use of deep research or autonomous workflow modes",
        "Video generation at any production scale",
        "A large-scale content operation using frontier models heavily across multiple modalities"
      ]
    },
    {
      "grade": "E",
      "label": "Very high energy intensity",
      "colour_hex": "#b10e1e",
      "colour_name": "Red",
      "definition": "Very high energy use. Disclose and actively consider lower-intensity alternatives before committing to this approach.",
      "typical_use": [
        "Heavy AI video production for a destination campaign",
        "Continuous use of frontier reasoning or deep research modes at high volume",
        "Large-scale automated content generation combining video, image and extended reasoning tasks"
      ]
    }
  ],
  "disclosure_format": {
    "fields": [
      {
        "field": "environmental_grade",
        "label": "Environmental impact grade",
        "type": "single_select",
        "options": [
          "A",
          "B",
          "C",
          "D",
          "E"
        ]
      },
      {
        "field": "task_types_used",
        "label": "AI task types used",
        "type": "multi_select",
        "options": [
          "Text",
          "Image generation",
          "Extended reasoning or agentic",
          "Video generation"
        ]
      },
      {
        "field": "model_category",
        "label": "Model category",
        "type": "single_select",
        "options": [
          "On-device or lightweight",
          "Standard cloud",
          "Frontier or large"
        ]
      },
      {
        "field": "usage_intensity",
        "label": "Usage intensity",
        "type": "single_select",
        "options": [
          "Light",
          "Moderate",
          "Heavy or continuous"
        ]
      },
      {
        "field": "conscious_choices",
        "label": "Conscious choices made",
        "type": "multi_select",
        "note": "Records thoughtful choices made during this work. Does not adjust the grade. The grade reflects what was done; this field records how consciously it was approached.",
        "options": [
          "Chose a lighter model where a more powerful one was not necessary",
          "Limited queries to what was needed for the task",
          "Avoided video or image generation where text would have served as well",
          "Used a provider with verified renewable energy infrastructure",
          "None noted"
        ]
      },
      {
        "field": "disclosure_note",
        "label": "Disclosure note for deliverable footer",
        "type": "text",
        "template": "AI tools used in producing this work have been assessed against the DTTT AI Environmental Impact Model and received a grade of [environmental_grade] ([label]). Task types: [task_types_used]. Scores are indicative. See thedttt.com/ai-transparency for methodology."
      }
    ]
  },
  "grading_scale": {
    "status": "introduced",
    "note": "The grading scale was introduced at v0.5. Grade boundaries and scoring logic will be reviewed by the DTTT governance committee and may be tightened at v1.0 as first-party disclosure data from AI companies improves under EU AI Act requirements. The current boundaries are deliberately calibrated to be accessible and indicative rather than artificially strict."
  },
  "methodology_sources": [
    {
      "source": "IEA",
      "title": "Electricity 2024: Analysis and Forecast to 2026",
      "year": 2024,
      "url": "https://www.iea.org/reports/electricity-2024",
      "relevance": "Data centre electricity consumption projections including AI workloads."
    },
    {
      "source": "Goldman Sachs",
      "title": "AI is Poised to Drive 160% Increase in Data Center Power Demand",
      "year": 2024,
      "relevance": "AI power demand projections and infrastructure investment context."
    },
    {
      "source": "Google",
      "title": "Gemini environmental impact disclosure",
      "year": 2025,
      "relevance": "First major first-party per-query environmental disclosure by an AI company. Reports median text query energy use of 0.24 Wh. Provides the most credible single anchor point for standard cloud text task energy estimates currently available."
    },
    {
      "source": "Epoch AI",
      "title": "How Much Energy Does ChatGPT Use?",
      "year": 2025,
      "url": "https://epoch.ai/gradient-updates/how-much-energy-does-chatgpt-use",
      "relevance": "Third-party estimate of GPT-4o query energy at approximately 0.3 Wh, broadly consistent with Google's first-party Gemini figure. Provides context for frontier model text task energy intensity and explains the role of mixture-of-experts architecture in efficiency."
    },
    {
      "source": "Hugging Face / Luccioni and Alves da Costa",
      "title": "What kind of environmental impacts are AI companies disclosing? (And can we compare them?)",
      "year": 2025,
      "url": "https://huggingface.co/blog/sasha/environmental-impact-disclosures",
      "relevance": "Critical analysis of 2025 environmental disclosures from Google, OpenAI and Mistral. Establishes that current company disclosures use incompatible methodologies and cannot be meaningfully compared. Supports the decision to base DTTT scoring on task-type and model-category bands rather than per-model figures."
    },
    {
      "source": "Bailey / Medium",
      "title": "The Energy Impact of LLMs \u2014 Where Are We in 2025?",
      "year": 2025,
      "url": "https://nathanbaileyw.medium.com/the-carbon-impact-of-llms-where-are-we-in-2025-6b6551e193ac",
      "relevance": "Synthesis of task-type energy differentials. Estimates image generation at up to 60 times more energy than simple text queries, and video generation at approximately 2,000 times more. Supports the task-type hierarchy in the base grade grid."
    },
    {
      "source": "MIT Technology Review",
      "title": "We did the math on AI's energy footprint",
      "year": 2025,
      "url": "https://www.technologyreview.com/2025/05/20/1116327/ai-energy-usage-climate-footprint-big-tech/",
      "relevance": "Documents the lack of first-party disclosure from closed model providers and the structural reasons why per-model verified figures are not currently available. Supports the disclosure-based rather than measurement-based approach of this model."
    },
    {
      "source": "EU AI Act / European Commission",
      "title": "EU General-Purpose AI Code of Practice",
      "year": 2025,
      "relevance": "Signed by OpenAI and Anthropic in July 2025. Requires comprehensive energy consumption metrics as part of model documentation. Enforcement begins August 2026 for new models. Expected to materially improve first-party disclosure data available to update this model's methodology sources at v1.0."
    }
  ],
  "governance": {
    "licence": "Creative Commons Attribution 4.0 International",
    "contact": "hello@thedttt.com",
    "committee": true,
    "versioning_policy": "Pre-release versions (0.x) are published openly but clearly labelled as in development. v0.5 introduces the grading scale and three-axis scoring logic. v1.0 marks the first stable release, to be published following committee review of grade boundaries and updated methodology sources reflecting improved first-party AI company disclosure.",
    "committee_review_notes": "The committee is invited to review: (1) whether the grade A definition adequately communicates that negligible does not mean zero impact; (2) whether the heavy usage intensity modifier of one grade is appropriate or should carry greater weight in specific task-type combinations; (3) grade boundary tightening at v1.0 as EU AI Act disclosure data becomes available."
  },
  "changelog": [
    {
      "version": "0.6",
      "date": "2026-03-16",
      "type": "minor",
      "summary": "Grade labels renamed to energy intensity framing throughout. Disclosure language standardised across all five grades to encourage consistent disclosure at every level, not only at higher grades.",
      "changes": [
        "Grade A label: Negligible impact \u2192 Low energy intensity",
        "Grade B label: Low impact \u2192 Low-moderate energy intensity",
        "Grade C label: Moderate impact \u2192 Moderate energy intensity",
        "Grade D label: High impact \u2192 High energy intensity",
        "Grade E label: Very high impact \u2192 Very high energy intensity",
        "Grade definitions rewritten: all five grades now carry a consistent 'Disclose' instruction rather than escalating urgency towards higher grades",
        "Colour key updated: 'Green \u2014 negligible' \u2192 'Green \u2014 low intensity'"
      ]
    },
    {
      "version": "0.1",
      "date": "2026-03-07",
      "type": "pre-release",
      "summary": "Initial pre-release. Establishes scope, energy intensity categories and disclosure format. Grading scale deferred to v0.5.",
      "changes": []
    },
    {
      "version": "0.5",
      "date": "2026-03-07",
      "type": "pre-release",
      "summary": "Introduces the A\u2013E grading scale with green-to-red colour direction, three-axis scoring logic (task type, model category, usage intensity), and updated methodology sources. Scope expanded to cover DMO marketing, content and operational use as well as DTTT advisory work. Mitigation field reframed as conscious choices disclosure field with no grade adjustment effect.",
      "changes": [
        "A\u2013E grading scale introduced with grade definitions, colour values and typical use examples",
        "Three-axis scoring logic introduced: task type (text, image, extended reasoning, video), model category (on-device, standard cloud, frontier) and usage intensity (light, moderate, heavy)",
        "Base grade grid introduced mapping task type and model category to a starting grade",
        "Usage intensity modifier introduced: light moves one grade down, heavy moves one grade up",
        "Twelve example assessments added covering common DTTT and DMO scenarios",
        "Scope section added defining intended users and applicable output types",
        "Colour direction set to green (#00703c) through amber (#f4a100) to red (#b10e1e), distinct from the Transparency Model green-to-blue gradient",
        "Mitigation field removed as grade modifier and replaced with conscious choices disclosure field, recorded alongside grade without adjusting it",
        "Disclosure format updated: energy category and carbon band fields replaced by environmental grade, task types used, model category, usage intensity and conscious choices fields",
        "Disclosure note template updated to reference DTTT Environmental Impact Model by name and version",
        "Methodology sources expanded: Google Gemini disclosure, Epoch AI GPT-4o analysis, Hugging Face comparative disclosure analysis, MIT Technology Review and EU AI Act Code of Practice added",
        "Grading scale status updated from deferred to introduced",
        "Committee review notes added flagging items for consideration ahead of v1.0"
      ]
    }
  ]
}