Agent-Evaluable Transparency
Your environmental data is only truly transparent when an AI agent can correctly answer a comparative question about it.
AET is the first named operational framework to apply Generative/Answer Engine Optimization (GEO/AEO) to corporate environmental transparency — with a self-diagnostic test.
First published June 2026
What is AET?
If a company publishes a sustainability report a human can read but an AI assistant cannot retrieve, parse, and reason over — relative to the market — then in an agent-mediated web that data is functionally opaque.
Transparency stops being "is it published?" and becomes "can a machine answer for it?" AET turns that into a testable property and gives you free, copy-paste prompts to test your own organization and fix it.
Why it matters now
Machines read the web more than humans
Bot traffic has overtaken human traffic; AI crawler traffic grew ~187% in 2025. Increasingly, the reader of your sustainability page is an agent acting for a human.
AI crawlers do not run JavaScript
AI crawlers (such as GPTBot, ClaudeBot or PerplexityBot) fetch the raw HTML and move on. Content that only appears after JavaScript is invisible to them.
"Machine-readable" is not "machine-answerable"
XBRL/ESRS tagging and ESAP make filings parseable, not answerable in natural language. AET targets that exact gap.
The four AET conditions
An indicator is Agent-Evaluable only when all four hold. Miss one, and it is not.
Reachable
The AI crawler is not blocked by robots.txt or WAF and receives HTTP 200.
Materialized
The value is in the raw HTML (SSR/SSG), not injected by JavaScript.
Structured
Labeled with machine context — JSON-LD with units, period and scope — consistent with the visible content.
Comparable
Accompanied by a benchmark or methodology reference so a relative judgment is possible.
The Ask-Your-AI Transparency Test
Do not ask your AI for a number. Ask a qualitative, comparative question — and judge whether it can answer correctly, with market context.
Prompt
“Considering indicator X, what is its variation relative to the market? Do our data indicate a favorable or unfavorable trend for the environment, compared with the market benchmark and methodology?”
✅ If the AI answers correctly, with benchmark context
Your data is Agent-Evaluable.
❌ If it cannot, hedges, or invents
The link between what should be communicated and what is machine-communicated is broken. Your data is not LLM-friendly.
Engineering gate (master test)
Simulate an AI crawler: fetch the raw HTML with JavaScript disabled. If your key indicators are absent from the response, the page is client-rendered and fails AET — fix with SSR/SSG first.
curl -sL -A "GPTBot" https://YOUR-SITE.com/sustainability \ | grep -i "<h1\|application/ld+json\|kWh\|PUE\|CO2\|water"
Two prompt families — test, then fix
AET ships as copy-paste prompts so any company can self-serve. No need to hire anyone.
Paste into your company AI assistant
Corporate, neutral, qualitative-comparative questions. If your own AI cannot answer them, your data is not agent-evaluable.
Energy intensity vs. the market
Acting as our internal sustainability analyst and using ONLY information that is publicly retrievable about our organization, answer: considering our energy intensity per request (or per unit of AI workload), what is its variation relative to the market benchmark? Is the trend favorable or unfavorable to the environment? State the specific disclosed figures and the source you relied on. If you cannot find the data, say so explicitly and explain what is missing.
Water use (WUE) vs. peers
Using only publicly available information about our company, compare our reported water-use efficiency (WUE) with the industry methodology and peer benchmark. Are we better, worse, or in line with peers, and is the direction favorable for the environment? Cite the exact source and reporting period you used. If the data is not retrievable, state that clearly.
Rate our transparency tier
Based strictly on what is publicly retrievable about our organization right now, classify our environmental transparency into a tier — A (full, per-unit metrics disclosed), B (partial, aggregate metrics only), or C (limited/none) — and justify the classification with the specific evidence you found. If you had to guess due to missing data, say so.
Carbon trend & direction
As an external analyst limited to public sources, describe the year-over-year direction of our carbon emissions (Scope 1/2/3 where disclosed) relative to our stated targets and the market benchmark. Is the trajectory favorable or unfavorable, and how confident are you given the data available? Cite sources and flag any gaps.
Can an AI even read our page?
Fetch the raw HTML of our public sustainability page WITHOUT executing JavaScript (the way AI search crawlers read it). Tell me which of our key environmental indicators (energy, carbon, water, PUE/WUE) are actually present in that raw HTML versus only appearing after JavaScript runs. List anything that is invisible to a non-JavaScript AI crawler.
Reading it: a confident, sourced, benchmark-aware answer = pass. "I don't have that data" or a generic answer = fail.
Paste into your AI coding agent
Ready instructions for your AI coding agent to apply AET to your own site — materialize, structure, expose, verify.
Master prompt: audit → plan → fix
You are an AI coding agent. Apply the AET (Agent-Evaluable Transparency) methodology to THIS website so our environmental/sustainability data becomes readable and answerable by AI agents that do NOT execute JavaScript (GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot, PerplexityBot). Reference methodology: AET = environmental data is only transparent if an AI agent can retrieve it from raw HTML and answer a benchmark-relative question about it. The four conditions per indicator are: (1) Reachable, (2) Materialized in raw HTML (SSR/SSG, not JS-injected), (3) Structured with machine context (JSON-LD schema.org incl. units, reporting period, scope), (4) Comparable (benchmark/methodology reference present or linkable). Do this in three steps and show me each before proceeding: 1) AUDIT — fetch the raw HTML of our public sustainability/ESG pages as a non-JS crawler. List which environmental indicators (energy, carbon, water, PUE/WUE, renewable %) are present in raw HTML vs only after JS. Check robots.txt and any WAF/CDN rules for AI-bot blocks. Report failures against the four AET conditions. 2) PLAN — propose the minimal changes to make each critical indicator pass all four conditions: which pages move to SSR/SSG, what JSON-LD to add (with units/period/scope), the robots.txt edits, an llms.txt, and any content rewrites into a direct-answer + benchmark-framed format. 3) IMPLEMENT — apply the changes. Then verify with: curl -sL -A "GPTBot" <URL> | grep -i "<h1\|application/ld+json\|kWh\|CO2\|PUE\|water" and confirm the indicators now appear in raw HTML. Constraints: keep JavaScript only for non-essential enhancement; do not invent or estimate any environmental figures — only expose what is actually disclosed; keep JSON-LD consistent with visible content. Summarize what passed and what remains.
Add JSON-LD with units/period/scope
Add valid schema.org JSON-LD to the raw (server-rendered) HTML of our sustainability pages so AI agents can parse our environmental indicators with full machine context. For each disclosed indicator (energy per request, renewable %, PUE, WUE, CO2/Scope figures), include the numeric value, unit, reporting period, and scope, plus a reference to the measurement methodology. Use an Organization node linked (sameAs) to our verified external profiles. Ensure the JSON-LD exactly matches the visible content (no divergence) and validates against the Schema.org / Rich Results validators. Do not include any metric we do not actually disclose.
Fix robots.txt + add llms.txt
Update our robots.txt to explicitly allow the AI bots we want to read us (GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot, Claude-User, PerplexityBot, Google-Extended) and confirm none are blocked by our WAF/CDN (they must receive HTTP 200, not 403). Then create a /llms.txt at the site root following the spec: a single H1 (our name), a one-sentence blockquote summary, curated ## H2 sections with "- title: short description" link lists pointing to our key sustainability/methodology pages, and an ## Optional section. Keep it a curated map, NOT a second sitemap — no link dumps.
Materialize data in raw HTML (SSR/SSG)
Our critical environmental data is currently rendered client-side, so it is invisible to AI crawlers that do not run JavaScript. Refactor our sustainability and key data pages so all critical content (indicator values, comparisons, headings, meta tags, JSON-LD) is present in the server-rendered raw HTML via SSR or SSG. Reserve client-side JavaScript for non-essential enhancement only (charts, animations, widgets). Verify with: curl -sL -A "GPTBot" <URL> | grep -i "kWh\|CO2\|PUE\|water" — the values must appear without JS.
The loop: Family 1 reveals the gap → Family 2 closes it. Prompts are in English (universal for AI agents).
Do your own research
Paste into any chat AI to run your own research on this company.
Act as an independent environmental-transparency correspondent applying GreenCodeAI's Agent-Evaluable Transparency (AET) methodology — data is only transparent if an AI agent can retrieve it and answer comparative, benchmark-relative questions about it. Stay neutral and evidence-based; you are applying GreenCodeAI's open method, not speaking for GreenCodeAI. To ground your analysis you may first read the public method at https://greencodeai.eu/aet and https://greencodeai.eu/methodology and reuse the same lens and benchmark framing. Subject: research the company "[company name or website URL]" and assess how transparent its environmental/sustainability data is. First, disambiguate the subject: confirm exactly which organization you are analyzing. Names can be ambiguous, so assume the organization in the AI industry by that name (the one relevant to AI's environmental impact), and state the legal entity and official website you used. If you find more than one plausible match, pick the AI-industry one and note the others. Start with a 3-line verdict — transparent / partially transparent / opaque, with one line of reasoning — before the detail below. Then: 1. Find its publicly disclosed environmental indicators — energy/kWh per request, carbon (CO₂, Scope 1/2/3), water (WUE), PUE, renewable energy %. Cite the exact source and date for each; PRIORITIZE primary sources (the company's own reports/pages) over third-party aggregators, and say explicitly when a metric is NOT disclosed. 2. For each indicator, tell me whether it is favorable or unfavorable for the environment compared with the market/peer benchmark, and how confident you are. 3. Assess machine-readability: could an AI agent read these figures from the company's raw HTML (without running JavaScript) and answer comparative questions about them? Flag anything hidden in PDFs, behind JavaScript, or simply absent. 4. Give an overall verdict: is this company transparent, partially transparent, or opaque about its AI environmental impact — and what should it disclose to improve? Rules: be specific and cite your sources; explicitly LABEL any figure you could not verify against a source; state clearly whenever data is missing rather than guessing. If you do NOT have web access, say so plainly and do NOT invent numbers — instead list exactly what should be checked and where.
Implementation checklist for your site
A page passes AET when all mandatory items pass. Use it as a PR / QA gate.
Mandatory
- Critical environmental indicators appear in raw HTML (master curl test, JS off).
- Single correct <h1>; semantic structure and heading order.
- Valid JSON-LD in raw HTML with units, reporting period and scope per indicator.
- robots.txt explicitly allows the AI bots you want (GPTBot, ClaudeBot, PerplexityBot…).
- AI crawlers are not blocked by WAF/CDN (logs show 200, not 403).
- Each indicator is benchmark-framed (comparison or link to methodology/peers).
- The Ask-Your-AI Transparency Test passes for your top 3 indicators.
Recommended
- /llms.txt follows the spec (H1, summary blockquote, curated H2 links, Optional).
- Pages open with a direct answer; named statistics, dates and source citations.
- E-E-A-T signals in HTML (author, credentials, update date).
- JavaScript reserved for non-essential enhancement only.
- Measure signals you own (AET pass rate, AI-crawler access in your logs) — not unverifiable cross-AI "visibility" scores.
We apply AET to ourselves first
Honest gap (June 2026): some of our ranking data is fetched client-side, so by our own master test it is not yet visible to AI crawlers. Fixing this is our canonical AET demonstration.
Integrity and proof: we publish our own AET checklist and close the gaps in the open.
Free for the community
AET is an open, free methodology. Test your organization, fix your site, and help build a transparent, agent-readable market for AI sustainability — no contract required.
Read the full methodology