The Green Blind Spot Edition 2026-Q4 · 113 companies · measured 2026-10-02

How the Index is measured

I built The Green Blind Spot to answer one question: how much CO2 does a company's website emit per visit? Brands that make public sustainability claims often don't see the footprint of their own websites; this index measures it. Here is exactly how I get that number, and where it falls short.

A starting point

I built this index to start a conversation. The model is not perfect, and it will get better. Each time I change it, I will publish what changed, what I got wrong and what held up, so everyone can learn from it.

I am not an expert on ecological footprints. I am an SEO, and I kept running into this in my work with clients: many brands don't realise their website has a footprint at all.

What I measure

Each company's 30 most-visited pages, based on estimated monthly search traffic from DataForSEO. I open each page in a real Chromium browser at a normal desktop screen size, and scroll down it five times so that content which only loads as you scroll (images, ads, embedded widgets) gets counted too. Then I add up every byte the page transferred.

From that byte count I get:

  • CO2 per page view: one page's transferred bytes converted to grams of CO2 with the Sustainable Web Design Model version 4 (the same model websitecarbon.com uses): 0.30 kWh per GB transferred across data centres, networks and devices, at the global average 494 g CO2 per kWh, about 0.148 g of CO2 per MB.
  • CO2 per visit: this is the number the ranking is sorted by, and it is not just one page's figure. I take the CO2 per view of every page I measured for that company and average them, weighting each page by how many of the company's visits actually go to it. A page almost nobody visits barely moves this number, even if it happens to be heavy; a page that gets most of the company's traffic moves it a lot. This is meant to reflect what a typical visit to the company's site actually costs, not the worst or best single page.
  • Annual CO2 for a page: that page's CO2 per view multiplied by its estimated monthly visits, multiplied by 12.
  • Annual CO2 for a company: the sum of that figure across all the pages I measured for it. I estimate this internally, but the ranking published here sorts by CO2 per visit, not by annual footprint.
  • Traffic coverage: what share of the company's total estimated visits the pages I measured actually account for. I publish this next to every company's number so you can see how representative it is.

Fewer than 10 pages, no number. If I could only measure fewer than 10 pages for a company, or its estimated traffic is under 100 visits a month, I don't publish a number for it. It stays out of the ranking rather than being shown with a number I don't trust.

How I rank companies

Lightest first, by CO2 per visit (rank 1 is the lightest site). This treats a small brand and a global one the same way, because it doesn't matter how popular the site is, only how much it costs the planet each time someone loads it.

Why I show a letter grade and a peer band, not just one number

Every company also gets a letter grade, A+ through F, using the Sustainable Web Design Model's own rating scale (A+ is up to 0.040 g per visit, F is anything above 0.359 g). In this edition, measured 2026-10-02, the median company here emitted 1.78 g per visit, about 5.0× the cutoff for the lowest passing grade, so almost everyone gets an F. That is a real finding about how heavy sustainability-branded websites tend to be.

I don't adjust this scale to flatter the companies in this Index. Inventing a softer scale just for this group would be the first thing a company could point to if it disputed its result.

Because almost everyone fails the letter grade, I also show a peer band: where a company sits against the others in this same edition, from lightest 25% up to heaviest 25%. I calculate this after removing companies I couldn't measure fairly, and I recalculate it every edition, so a company's peer band can shift because the companies around it changed, not only because its own site changed.

How the letter grades are set

Every grade is a fixed cutoff in grams of CO2 per visit, from the Sustainable Web Design Model version 4 rating scale. I don't move these to make the results look better or worse.

The cutoffs come from page weights in the HTTP Archive, a public record of how millions of web pages are built. A+ is the lightest 5% of web pages. The line between E and F is the median web page, about 2.4 MB.

GradeCO2 per visit
A+up to 0.040 g
Aup to 0.079 g
Bup to 0.145 g
Cup to 0.209 g
Dup to 0.278 g
Eup to 0.359 g
Fabove 0.359 g

Who is included

Companies that publicly position themselves on sustainability: certified B Corps, companies with a public net-zero or science-based target, consumer brands that market themselves as sustainable, or companies whose core product is a climate or environmental technology. I curate and version this list myself, and a company's inclusion is based on its own public claims, never on whether it is a client of mine.

Anyone can suggest a company, including their own. I review each one against these criteria and decide. Submit a site for review.

How the cost is worked out

Each ranked company page shows one yearly euro figure for what heavy, slow pages cost. It adds two parts. These are the only outside numbers:

Visitors at risk. The ad value of the company's search traffic for a year, times the share of its page loads that happen on phones, times the extra seconds those phone loads take beyond 2.5 seconds, times the share of visitors lost per second. The ad value is what the same search visitors would cost as Google Ads clicks, in the company's main country.

Hosting for the extra weight. The weight of a typical page above 1 MB, times the company's yearly search visits to the pages I measured, times the delivery price per GB.

What this can and cannot prove. Google measures how slowly pages load for real Chrome users, not why. Page weight is one cause; servers, scripts and images are others. A lighter page usually loads faster, but cutting weight alone may not win all of this back.

I chose the cautious number at most steps: desktop left out, slow loads counted as 5 seconds at most, search visits only. The exception is the ad value: DataForSEO prices every search visit at what the click would cost as an ad, including searches for the company's own name, which a well-known brand could buy back for less. For the largest brands that makes the figure higher than what they would really pay.

Page weight comes from the 2 October 2026 crawl; search ad value was fetched on 3 November 2026; phone speed is Chrome UX Report September 2026; dollars converted at the ECB rate of 3 November 2026.

What this does not capture (read this before citing a number)

  • Only estimated organic search traffic counts. Visits from direct links, paid ads, social media and other sites referring people in are not included, so the total annual CO2 figures understate the real total. This applies evenly across every company, so comparisons between companies still hold.
  • Only the top 30 pages, not the whole site. I publish the traffic coverage percentage for each company so you can judge how complete that picture is.
  • No adjustment for green hosting yet. The model lets a site on renewable-powered servers remove the data-centre share of its figure, which would make it at most about 18% lower. The 2026 editions leave it out, so they can be compared with each other. I plan to add it from the January 2027 edition and will say on this page what changed. websitecarbon.com already applies it, so a green-hosted site can score a little lower there than here.
  • Cookie banners: measured before the visitor clicks, until the November 2026 crawl. Up to the October 2026 crawl I measured each page as it loads before anyone answers the cookie banner. On sites with ads, the ads and trackers only load after a visitor clicks "Accept all", so those sites come out lighter here than the page most of their readers get. On 7 October 2026 the Guardian's homepage was 4.3 MB before accepting and 12.8 MB after. Sites without a banner are measured in full, so the comparison is uneven. From the November 2026 crawl I accept the banner first and measure what a visitor who clicked "Accept all" downloads. Earlier crawls can't be recalculated for this, so sites with a banner will come out heavier. The January 2027 ranking is the first built on these crawls, and I will mark it as the point where the method changed.
  • Pages here come out heavier than a basic load. I scroll each page to load lazy content, while the HTTP Archive behind the grade scale measures pages without scrolling.
  • No allowance for browser caching. Every visit is measured as if it were a first-time visit, not a returning one where the browser already has some files stored.
  • Measured against United States, English-language search by default. Some companies have a different market set in my configuration, but for any company without one, and whose real audience is mostly outside the US, its traffic and total footprint will be understated. Again, this is applied consistently, so relative ranking is still meaningful.
  • Not the same numbers as my client reports. For client work I run a separate tool calibrated to match Screaming Frog, which uses an older conversion. The Index does not use it.
  • These are estimates, produced the same way for every company. The Index is built to compare companies against each other fairly, not to state the exact carbon cost of a single visit to any one of them.

What running this Index costs

Roughly 6 kg of CO2 per monthly crawl. Almost all of it is the data the crawl downloads; the computer doing the work adds a fraction of a percent.

The crawl runs on a Raspberry Pi in my home. It is a Raspberry Pi 5, the small computer that also runs my home's smart-home system, so it is switched on anyway. Once a month it opens every page in the Index the way a visitor would.

Figures for the October 2026 crawl, on 2 October from 01:00 to 06:57 Amsterdam time:

  • Data: 40.7 GB downloaded across 3,278 page loads for 134 companies. Put through the same model I use for every company, that is 6.0 kg of CO2. The model counts the data centres and networks that serve the pages as well as the device that loads them.
  • The Pi's own electricity: at most about 0.05 kWh. The Pi has no meter of its own, so this comes from its built-in power readings: about 2 watts at rest and 8 watts flat out, over the six hours of the crawl.
  • Battery or grid: about half each. The crawl ran at night, so no solar power came in directly. My plug-in home battery, charged by the solar panels on my roof the day before, supplied about 0.33 kWh while the crawl ran, and the grid about 0.34 kWh. These are readings for the whole house; the Pi was one of the things drawing on them.
  • The grid part comes from Vattenfall. My contract carries Vattenfall's "Groen uit Nederland" label: 90.9% Dutch wind and 9.1% Dutch solar in 2025, according to Vattenfall's electricity label. A label like this works through certificates of origin. The electricity that physically reaches my house at night is the Dutch grid mix, which averaged 254 g of CO2 per kWh in 2025 (Ember).

Even counted entirely at the Dutch grid average, the Pi's own electricity comes to about 13 g of CO2, a fraction of a percent of the 6.0 kg the model gives for the data.

What else runs the Index, and whether I count it:

  • AI help: 3 to 6 kg of CO2 in October 2026, about as much as the crawl. I build and run the Index with Claude, an AI assistant. My logs for 1 to 28 October show 1,286 replies on this project, which also covers my client crawls, so this is an upper bound for the Index alone. The low end follows EcoLogits (text written only); the high end counts every word read and written, on the US grid average. Most of it falls in months I change the Index, not months it only runs.
  • Hosting this site: about 15 g of CO2 a month. Netlify served 135 KB in a sample hour on 28 October; over a month that is about 0.1 GB, put through the same model as the companies.
  • The database: left out. The Index's results take up 1.4 MB in Google BigQuery. Storing and reading that is too small to show up.
  • Traffic data from DataForSEO: left out. I buy about $3.60 of search traffic estimates a month. DataForSEO publishes no energy figures, so I can't put a number on it.
  • Domain names at Cloudflare: left out. Cloudflare answers the lookups for this site's address and forwards greenblindspot.org here. It publishes no figure per customer.

What changed, and when

From the 2027-Q1 edition, each company is measured in its main country. Up to and including 2026-Q4, I measured most companies on United States, English-language Google search, and only some on their own country. From the January 2027 measurement, every company has a main country, and its 30 pages and their traffic come from Google in that country. For a European company these are often different pages than before, for example its Dutch or French pages instead of its English ones. The edition shown on this site is still measured the old way.

Rankings can move because of this, not because a website changed. A company's grams per visit, grade and place can all shift between 2026-Q4 and 2027-Q1 for this reason alone. Compare those two editions with that in mind. From 2027-Q1 on, editions are measured the same way again and compare directly.

The same edition adds a page per country, next to the pages per sector and per claim.

How often this runs

Monthly measurement, quarterly ranking. I re-measure every company's website monthly, and publish a new headline ranking every quarter. Companies I couldn't measure fairly are re-checked at every monthly crawl. From the 2027-Q1 edition, each company page and the home page show the monthly measurements as a line, starting November 2026. A quarterly ranking smooths out month-to-month swings in search traffic estimates; the line shows them.

If something looks wrong

Tell me and I'll re-measure. Use the dispute form on that company's page, or email letsgrow+greenblindspot@baba-seo.com. Tell me what looks off and I'll re-measure and reply.

Thank you

This method is better because people challenged it. I'd rather have it questioned than admired, so here is who did, and what changed because of it.

  • Dawn Regan (GreenSEO) questioned the Guardian's number at the GreenSEO meetup on 7 October 2026, the evening I published it. That question is why the cookie-banner limit above exists, and why sites with a banner are measured after "Accept all" from the November 2026 crawl.