Why we build for LLMs, not just Google
More people are starting their research in ChatGPT, Claude, Perplexity and Gemini than ever before.
Whitepaper
We tested the websites of 453 UK environmental and purpose-led organisations. They include Wildlife Trusts, Rivers Trusts, National Parks, community energy groups, sustainability consultancies and sustainable architects, many of them working to cut emissions.
A heavy website rarely does great harm on its own, but it works against the organisation’s mission. It is also usually slower and harder to use, and it can be less visible in search. This report sets out the evidence and what to do about it.
Organisations with serious climate commitments usually measure their buildings, travel and supply chains. Their website is rarely on the list.
The 2026 Charity Digital Skills Report found that 75% of UK charities take at least one step to reduce the environmental impact of their digital use. Most of those steps are about other things: 59% use digital tools to cut printing and travel, and 39% extend the life of their devices. Only 8% design their websites and online services for energy efficiency, and only 5% have ever carried out a digital sustainability audit.
Certification schemes don’t single websites out either. B Lab’s new B Corp standards, published in May 2025, split the climate requirements by company size:
In both cases, a website only gets counted if the organisation chooses to include it.
This matters because digital technology has a real footprint. A peer-reviewed study by Freitag, Berners-Lee and colleagues (Patterns, 2021) estimates that information and communications technology accounts for about 2.1–3.9% of global greenhouse gas emissions. For comparison, commercial aviation accounts for about 2%. The authors stress that these figures carry significant uncertainty.
Websites are also getting heavier. According to the HTTP Archive Web Almanac 2025, the median mobile home page grew from 845 KB in July 2015 to 2,362 KB in July 2025, almost three times as much data to load.
| Rating | Emissions per page view |
|---|---|
| A+ | 0.040 g or less |
| A | 0.079 g or less |
| B | 0.145 g or less |
| C | 0.209 g or less |
| D | 0.278 g or less |
| E | 0.359 g or less |
| F | 0.360 g or more |
Global average page transfer size
The grades come from the HTTP Archive’s crawl of the web on 1 June 2023. The F threshold matches the global average page transfer size of 2.4 MB, so an average web page already earns an F. The authors say they “intentionally set the bar high” to push the industry towards ambitious reductions.
These figures are estimates, not meter readings. The model’s authors say it is likely to overestimate rather than underestimate. That makes it most useful for comparing sites and tracking improvement over time.
Figures per page view are small, so it helps to scale them up. The table assumes a site with 10,000 page views a month (120,000 a year), with every view costing the same as its home page:
| Site | Per year |
|---|---|
| A+ site | 5 kg |
| Median site | 48 kg |
| Heaviest 10% | 154 kg |
| Heaviest site | 906 kg |
For most organisations, even the heaviest figure is a small part of their total footprint. That is why the case for lighter websites is about more than tonnes of carbon. A lighter site keeps an organisation’s digital work consistent with its mission, and it is faster and easier for everyone to use.
| Rating | Share of sites (%) |
|---|---|
| A+ | 2 |
| A | 3 |
| B | 8 |
| C | 8 |
| D | 10 |
| E | 13 |
| F | 57 |
| Type of organisation | Sites | Rated F | Rated B or better | Median page (MB) | Green hosted | Median mobile score |
|---|---|---|---|---|---|---|
| Sustainability consultancies | 29 | 76% | 3% | 3.4 | 45% | 62 |
| National Parks | 15 | 73% | 0% | 3.6 | 60% | 53 |
| Rivers Trusts | 43 | 70% | 9% | 3.5 | 67% | 58 |
| Sustainable architecture practices | 24 | 67% | 13% | 3.4 | 42% | 65 |
| Environmental charities | 43 | 65% | 0% | 3.6 | 72% | 57 |
| National Landscapes | 44 | 64% | 5% | 3.6 | 75% | 60 |
| Other purpose-led organisations | 59 | 61% | 12% | 3.5 | 61% | 59 |
| Water and marine charities | 45 | 53% | 18% | 2.7 | 69% | 65 |
| Wildlife Trusts | 44 | 52% | 0% | 2.4 | 9% | 45 |
| Community energy groups | 107 | 36% | 30% | 2.1 | 79% | 64 |
| All | 453 | 57% | 13% | 2.9 | 62% | 59 |
Three patterns stand out:
The picture is similar across the UK. 58% of sites in England and Wales were rated F, as were 57% in Northern Ireland and 48% in Scotland.
Which tool a site is built with matters less than how it is built. Among WordPress sites, those built with a page builder and those without had similar median weights (3.3 MB each) and similar F ratings (62% and 59%). The weight comes mostly from images, scripts and third-party code, which careful building controls whatever the tool.
The 22 sites rated A+ or A don’t use special technology. They simply load much less:
| Metric | A+ and A sites | F sites |
|---|---|---|
| Sites | 22 | 256 |
| Median home page | 0.37 MB | 4.4 MB |
| Median number of files loaded | 17 | 89 |
| Images as a share of page weight | 45% | 56% |
| Median mobile performance score | 90 | 56 |
| On verified green hosting | 82% | 57% |
| Built with a page builder | 23% | 36% |
The best sites aren’t text-only or plain. Half of them run on WordPress, and images still make up almost half of what they load. The difference is discipline: fewer and smaller images, fewer scripts and plug-ins, and less third-party code. The result is a home page roughly twelve times lighter, which also loads fast enough for Google to class it as good. Most of the A-rated sites belong to community energy groups, which tend to have small, simple sites.
The same pattern appears in other research. A 2026 study in the journal Sustainability used the same v4 model to test the home pages of 12 European universities, and half were rated F. It found that design and hosting choices explained the differences better than each country’s electricity grid.
Method. WINC Studio tested 465 home pages on 24 September 2026. Page weight, speed and accessibility came from Google PageSpeed Insights (Lighthouse, mobile). Carbon ratings came from CO2.js using the Sustainable Web Design v4 model. Hosting was checked with the Green Web Foundation’s green check. Google could not load 12 of the pages, leaving 453 results.
The organisations were taken from public directories: The Wildlife Trusts, UK National Parks, the National Landscapes Association, The Rivers Trust, community energy networks, and the national environment link bodies. The list was topped up with 59 organisations WINC Studio had already identified. Their results matched the rest: excluding them, 56% were rated F.
Each home page was tested once, and scores vary a little between runs. This is a structured sample of UK environmental organisations, not a random sample of all UK websites.
Nearly half the sites we tested (47%) scored below 90 in Lighthouse’s automated accessibility check, and a higher score doesn’t mean a site passes. Automated tools only catch some problems.
The Government Digital Service showed this by building a page with 143 deliberate accessibility failures and running ten tools over it. The best single tool found 37% of the failures. All ten tools together found 71%, and the rest could only be found by a person. An automated score shows the least a site gets wrong, not how accessible it is. Checking with a keyboard and a screen reader, and testing with users, is still needed.
The legal position. Public sector bodies must meet the WCAG 2.2 AA standard and publish an accessibility statement. That includes National Park Authorities and councils, and charities that are mostly publicly funded or provide essential services. In England, Scotland and Wales, the Equality and Human Rights Commission enforces these rules; in Northern Ireland, the Equality Commission does. Every organisation also has duties under the Equality Act 2010 to make reasonable adjustments for disabled people.
How it connects to sustainability. Lighter pages load faster on older phones and slower connections, so they reach people that heavy sites leave behind. The W3C guidelines say sustainable practice tends to make sites more accessible as a by-product. Our data shows the link is loose, though. A+ and A sites had a median accessibility score of 92.5, compared with 90 for F-rated sites. Accessibility needs its own attention, not just a lighter page.
Green hosting is worth doing, but on its own it will not move a heavy site out of an F.
What “green” means. The Green Web Foundation keeps the largest open database of verified green hosting providers. To be listed, a provider must show evidence that it has “taken steps to avoid, reduce or offset 100% of the greenhouse gas emissions caused by using electricity to provide their services.” Anyone can check a website for free with the Foundation’s Green Web Check.
How common it is. The Web Almanac 2024 sustainability chapter found that only 10–12% of all websites measured were on verified green hosting. Among the 1,000 most visited sites the figure was 55–56%. In our survey, 62% of the 453 sites were verified green, so environmental organisations are well ahead of the web as a whole. Even so, more than a third were not, and among the 44 Wildlife Trusts only 4 were.
| Type of organisation | Green hosted (%) |
|---|---|
| Sustainability consultancies | 45 |
| National Parks | 60 |
| Rivers Trusts | 67 |
| Sustainable architecture | 42 |
| Environmental charities | 72 |
| National Landscapes | 75 |
| Other purpose-led | 61 |
| Water and marine | 69 |
| Wildlife Trusts | 9 |
| Community energy | 79 |
| All | 62 |
The rules are getting stricter. From 1 October 2026, the Green Web Foundation will no longer accept carbon offsets as the basis for verification. Providers will need renewable energy certificates or similar evidence matched to the time and region where the energy is used. Hosts that are already verified must meet the new criteria when they are next re-verified, in February/March or June/July 2027. Some hosts that call themselves green today may lose that status.
What hosting does and doesn’t cover. In the v4 model, green hosting removes the operating energy of the data centre. That is about 18% of the model’s total (0.055 of 0.300 kWh per GB). Networks, visitors’ devices and the emissions from manufacturing hardware are not affected. For example, a 2.9 MB home page (our survey’s median) comes to about 0.45 g CO2e per view on standard hosting and about 0.37 g on green hosting. Both are F ratings. The biggest gains come from sending less data, with green hosting on top.
The university study found a large gap: green-hosted home pages had a median of 0.16 g per view, against 0.76 g for the rest. That is more than hosting alone can explain. Our survey found almost no gap, 0.38 g against 0.43 g, because the green-hosted sites were slightly heavier (median 3.0 MB against 2.8 MB). Choosing green hosting and building a light site are separate decisions, and it takes both.
A lighter website does help search visibility, but only indirectly. The link runs through speed and user experience, not through a “green” ranking signal, and Google has no such signal.
What Google says. Google’s own documentation states that “Core Web Vitals are used by our ranking systems.” These are three measures of real visitors’ experience. Google’s targets are: main content visible within 2.5 seconds, a response to interaction within 200 milliseconds, and minimal layout shift (under 0.1). Pages that send less data generally load faster, which makes the first target easier to reach, especially on mobile.
What Google doesn’t say. Speed is not a shortcut to the top of the results. Google is explicit that it “always seeks to show the most relevant content, even if the page experience is sub-par”. It also says good scores “doesn’t guarantee” top rankings. Useful, relevant content still matters most. Where competing pages are similarly relevant, a better page experience can help.
The knock-on effects. Speed also changes what visitors do once they arrive. In a study for Google, Deloitte and 55 analysed 30 million sessions on 37 European and US brand sites. Making mobile pages 0.1 seconds faster was linked to 8.4% more retail conversions. It was also linked to a 21.6% improvement in lead-generation visitors progressing to a form submission. The study covers commercial sites, but the pattern is relevant to any site that wants donations, sign-ups or enquiries.
The wider picture. The W3C Web Sustainability Guidelines are a Group Draft Note published in November 2025. They say that following them “will often make web content and services more accessible, usable, performant, and secure as a by-product.” Our survey shows the link clearly. Sites rated A+ for carbon had a median mobile performance score of 100, B-rated sites 77, and F-rated sites 56. 211 of the sites had enough real-visitor data in Google’s Chrome UX Report; of those, 27% missed Google’s 2.5-second target for main content. For most of these organisations, reducing page weight would improve their carbon rating and their page experience together.
None of these rules mentions websites, and most small organisations aren’t directly covered by them. But each one means more organisations are measuring the emissions from services they buy, and a website is one of those services.
| Change | Who it affects | When |
|---|---|---|
| Green Web Foundation stops accepting offsets as evidence of green hosting | Hosting providers, and anyone relying on a "green" label | New verifications from 1 October 2026; existing providers at re-verification in 2027 |
| B Corp standards require larger companies to measure Scope 1, 2 and 3 emissions | B Corps and companies seeking certification | Published May 2025, applied as companies recertify |
| UK Sustainability Reporting Standards published | Voluntary for now; the FCA is consulting on requiring them for listed companies | Published 25 February 2026; proposed from 1 January 2027 |
| PPN 006 requires a Carbon Reduction Plan | Suppliers bidding for central government contracts worth over £5 million a year | In force under the Procurement Act 2023 |
Most organisations can make a real difference with three changes, in this order.
Fix images first. At the heavy end of the web, images make up about 70% of page emissions, according to the Web Almanac 2024. Resize them, use modern formats such as WebP or AVIF, and lazy-load anything below the first screen.
Cut unused code. Remove plugins, trackers, fonts and page-builder code you don’t need. The Web Almanac found that the heaviest tenth of mobile pages ship more than 800 KB of JavaScript that is never used. For comparison, the best sites in our survey load a median of 17 files, and the F-rated sites load 89.
Check your hosting. Use the Green Web Check. If your host relies on offsets, ask how it will meet the new criteria that apply from 1 October 2026. Green hosting helps, but it works on top of a light site, not instead of one.
Answer yes or no to each question. The Website Carbon Calculator, Google’s PageSpeed Insights and the Green Web Foundation’s Green Web Check are all free, and between them they will give you most of the answers in a few minutes.
Seven or fewer yeses? Your website is probably working against your mission, and the free audit below will show you where to start.
WINC Studio offers a free audit to purpose-led organisations. You get a short written report within five working days covering:
There’s no obligation. Book at email or call 0117 911 9464.