1.1. UNFCCC COP

COP has been central to global climate governance since 1995, announced under the UNFCCC in 1994 [6]. Established at the 1992 Rio Earth Summit (formally the United Nations Conference on Environment and Development (UNCED), the UNFCCC now has 198 member states, serving as the primary framework for climate change action, coordinating the global response by negotiating commitments, refining mitigation and adaptation strategies, and providing financial and technological support [5,6]. It has produced landmark agreements, including the 1997 Kyoto Protocol, setting GHG emission reduction targets for developed nations [7,8], and the 2015 Paris Agreement, which committed nearly all member states to limiting global warming to 1.5°C above pre-industrial levels [2].

COP is held annually, structured around high-level plenaries and technical negotiations. The member states’ host country selected from United Nations regional groups, plays a crucial role in shaping the agenda. The UNFCCC operates on the principle of “agreement by consensus,” ensuring broad participation while also allowing any country to block progress [8], underscoring both the collaborative and contentious nature of climate diplomacy.

1.2. Good COP, Bad COP

As the urgency of climate action has intensified, so too has the scale and complexity of COP. What began as a relatively small diplomatic gathering has expanded into a global event, a phenomenon the European Capacity Building Initiative (ECBI), which supports developing countries in building capacity for effective participation in international climate negotiations, has referred to as “Mega-COP” [9]. According to the UNFCCC, the number of in-session attendees has grown exponentially by 1,664%, from 3,969 participants at COP1 (1995) to 70,002 at COP28 (2023) [10]. This growth is not attributable to an increase in workload: prior research demonstrated that the size of agendas and the number of parallel sessions has remained “relatively stable” [10]. Nevertheless, participation figures continue to rise in a pattern termed “summit nostalgia” [10]: we argue that this may be attributed to the “ratcheting effect,” a form of systemic expansion that resists de-escalation, observed in research on deescalating the impact of computing and the internet [11].

The scale of COP has drawn increased scrutiny, particularly regarding its environmental impact. The selection of host countries with vested fossil fuel interests [12], concerns over human rights violations [13], the use of private jets, and the involvement of industry figures with potential conflicts of interest have all contributed to growing “climate scepticism” [8,14]. Even the UNFCCC acknowledges this vulnerability: “civil society and the media will raise tough questions on the sustainability aspects of all organisational matters relating to the conference, from emissions offsets to waste management. Host countries will need to be well prepared with answers” [15].

1.3. COP Host Country Websites and Environmental Impact

Each selected COP host country takes on the responsibility of logistics, infrastructure, and digital communications, including the official COP host country website, serving as the central platform for event registration, agenda dissemination, public engagement, and media coverage [15]. Despite the broad guidelines set by the UNFCCC for hosting COP, as detailed in the HCA, a legal accord that formalises the responsibilities of each host nation, and in the How to COP handbook [15–17], there is a lack of specific directives regarding the standardisation of host country online presence, leading to significant variability. The handbook encourages host countries to “identify opportunities to deliver environmental and social value” [17] but does not include SWD. Similarly, the COP30 HCA contains no provisions addressing the sustainability of the host country website for COP30 [18].

The recently published Digital Economy Report 2024 by the UNCTAD highlights the current and growing environmental impact of computing and the internet [1], acknowledging it “had failed to recognise the relationship between information and communications technologies (ICTs), the Internet and sustainability,” and much of digitalisation has progressed “without taking into consideration its direct environmental impact [1].” However, it does not specifically address strategies for mitigating this impact through SWD.

The absence of a COP sustainable web policy has led to unintended negative consequences, both reputational and environmental (see Section 2.2). This lack of clear guidance contributes to “COP fatigue,” further eroding trust in climate action initiatives—”green-scepticism” [8]. Without definitive directives, host countries risk perpetuating unsustainable digital practices that contradict COP’s climate commitments, ultimately neglecting the environmental impact of websites.

The ECBI has called for a reduction in the scale of COPs and greater oversight of their organisation [10]. It argues that “the current arrangements… are not fit for the current purpose,” with potentially severe consequences for the multilateral climate regime [10]. Discussions on COP have primarily focused on its physical scale of the conference, however the digital footprint remains largely unexamined (see Section 3).

2.1. The environmental impact of the internet

Since the first proposal for a hypertext project called WorldWideWeb in 1989 [19], the internet has evolved from a lightweight academic tool into a complex, media-rich platform. Technological advancements have unshackled prior constraints on network and device speeds, increasing accessibility to the internet. However, rather than fostering efficiency, these advancements have led to unchecked growth, in line with Jevon’s Paradox, where increased resource efficiency and availability leads to higher consumption [20].

“Digitalisation’s promise of dematerialisation has not yet materialised”, as the internet, the world’s largest machine, now contributes an estimated 1.5% to 3.2% of GHG emissions [1], double that of the aviation industry [21]. With 5.5 billion internet users globally (68% of the world’s population) growing at an annual rate of 4%, this trend continues to accelerate [22]. Each byte added to the web accelerates an extractive systems that deplete vital resources such as water, land, and abiotic materials, compounding environmental harm across the technology supply chain [1,23]. Webpage sizes—the total file size of all components delivered as part of a webpage, including HTML, style sheets (CSS), scripts, images, fonts, and other assets—have increased considerably due to increasingly sophisticated, feature-rich, and media-dense designs, as well as the growing use of third-party services and analytics tools [24]. Desktop webpage sizes have increased by 465% and mobile webpage sizes by 1530% since 2010 [24]. By 2024, the average webpage size had reached 8 MB on desktop and 7.2 MB on mobile, exceeding the recommended targets of under 1 MB, ideally around 500 KB [25]. As a result, the average webpage now earns an F rating (≥ 0.847g CO₂e) according to Website Carbon (available at https://websitecarbon.com), which assesses the environmental impact of websites based on their energy consumption and GHG emissions.

The proliferation of resource-intensive websites not only exacerbates environmental harm but also deepens digital inequities, disproportionately affecting the 47% of the world’s poorest 40% who rely on mobile devices with slower network speeds [26]. This, in turn, limits progress on the United Nation’s Sustainable Development Goals (SDGs) [27]. Establishing robust website sustainability policies is an imperative for maintaining the reputational integrity and efficacy of climate action efforts and minimising the environmental impact of the internet.

2.2. Sustainable web design and “Greenwashing”

The SWD movement has emerged in response to growing awareness of the ecological impact of websites. SWD “is an approach to designing web services that puts people and planet first” [28]. The movement advocates for websites that are designed to lower emissions associated with the delivery and use of web content. The impact of websites is measured using metrics such as total page weight (in kilobytes KB) (the total size of all files that must be downloaded to load a page, including images, scripts, style sheets, etc.) and equivalent carbon emissions per pageview (in grams of CO₂e). These can now be assessed using widely accessible tools such as Website Carbon (available at https://websitecarbon.com) and CO2.js by The Green Web Foundation (available at https://observablehq.com/@greenweb/co2-js-playground).

Key SWD principles rely on simple interventions that can significantly reduce environmental impact while also improving performance and accessibility. These include compressing images, using modern image formats, minimising the use of client-side scripts, reducing unnecessary third-party requests, optimising typography by reducing and limiting font file size, enabling efficient navigation structures, and limiting page weight through a page weight budget [29,30].

In 2024, the World Wide Web Consortium (W3C) (the main international standards body for the internet) published the first version of the Web Sustainability Guidelines (WSG) (available at https://w3c.github.io/sustainableweb-wsg/), outlining evidence-based best practices for “making websites and products more sustainable” [29]. The guidelines centre on simple, practical interventions that can significantly reduce the environmental impact of websites while also improving performance and accessibility. However, despite its potential, the lack of enforcement mechanisms and limited organisational adoption, compounded by the “fog of enactment” [31], continues to impede substantive progress [32], while also creating fertile ground for both direct and indirect “greenwashing” [33]: defined as “the intersection of two firm behaviours: poor environmental performance and positive communication about environmental performance” [34], and more commonly as “the deceptive practice where companies exaggerate or falsely represent their environmental efforts, misleading stakeholders into believing that their products, services, or operations are more environmentally friendly than they truly are [35].” “Greenwashing” hinders meaningful progress in reducing environmental impact [36]. “Indirect greenwashing” fosters a climate of “green-scepticism”, making it difficult for stakeholders to distinguish between genuine sustainability efforts, unintentional misrepresentation, and strategies deliberately designed to mislead [37]. As a result, credible sustainable web practices may be undermined by growing distrust, reducing the effectiveness of green communications, “climate fatigue” [8,37], even with tougher guidelines announced in 2024 by the International Auditing and Assurance Standards Board (IAASB) and the International Ethics Standards Board for Accountants (IESBA) on judging firms’ environmental claims [38].

COP28, hosted by the UAE in 2023, faced criticism for “greenwashing” over its host country website (previously available at https://cop28.com), which featured a “low carbon” toggle marketed as a sustainability feature. However, investigations revealed that this feature only hid images visually, though they were still downloaded [39]. Furthermore, the “low carbon” version was not the default setting, rendering the largest version of the website for new visitors, and the layout failed to adapt to hidden images, leaving empty spaces, giving users the impression that this mode was “lifeless” [40], incomplete, or not fully functional, and “underwhelming experience” [41]. The website received an F rating from Website Carbon, and each visit was estimated to produce 1.7g CO₂e [40]. Additional research found the low carbon version did not reduce device power consumption; in fact, it slightly increased it on laptops [42]. COP28 also announced the ITU’s Green Digital Action initiative, “where tech companies and governments committed to advancing climate action through digital technology” [43]. This juxtaposition highlights a tension between the ambitions of such initiatives and the reality of digital environmental impact.

Our prior research on the management policies of United Nations websites shows that, in the absence of a system-wide web sustainability policy, the United Nations remains vulnerable to reputational risks and accusations of “greenwashing” [33], ultimately fuelling “climate fatigue”, and “COP fatigue” [8,37]. The ECBI has explicitly warned that failure by the UNFCCC to adopt robust sustainability measures “would have major negative consequences for the multilateral climate regime. [10]”. This underscores the urgent need for comprehensive policies to ensure the COP’s web presence aligns with its climate action objectives.

4.1. Methodology overview

The primary contribution of this study is a methodological framework that connects research questions (RQs) with measurable variables, data sources, and analytic tools. Table 1 summarises this framework.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Overview of research framework.

https://doi.org/10.1371/journal.pclm.0000767.t001

4.2. Data collection and processing script

To assess the environmental impact of COP host country websites, we developed a Python utility to reconstruct the original total page size with high fidelity. The utility (available at https://doi.org/10.5281/zenodo.15295396) can be used by other researchers to replicate our longitudinal analysis of COP host country websites or apply similar methods to other websites.

The Python script is dependent upon Playwright, a cross-browser automation library, to replay Wayback Machine snapshots within a headless instance of the Chromium browser, capturing all network requests and logging every resource (HTML, CSS, JavaScript, images, video, fonts, etc.) along with their type and size in kilobytes (KB). This enables identification of the primary contributors to page weight and supports estimation of the associated carbon footprint using the Website Carbon API (available at https://api.websitecarbon.com) (see Section 4.5). When the script completes analysing the dataset, it exports a CSV file with results for all the webpages, including total and average page size (KB), estimated emissions (grams CO₂e), Website Carbon ratings (A+ to F), and a full breakdown of page composition by size and percentage. The script also generates a chart showing emission trends across websites.

4.3. Addressing wayback machine limitations

The development of this method was prompted by the fact that simply submitting Wayback Machine snapshot page addresses (URLs) to existing website analysis tools is not feasible and does not yield accurate results. We observed that webpages retrieved from the Wayback Machine often take a long time to load, leading to timeout errors and incomplete resource capture when analysed with conventional tools. Even when an archived page loads successfully, distortions introduced during archiving and replay—the processes that reconstruct and display a webpage as it appeared at the time of capture—can compromise accuracy, and it is often unclear which resources, if any, are missing from the page [50].

The first of these distortions to page size is the Wayback Machine’s reliance upon a method of replay called “integrated replay” [51], which injects additional resources, such as the navigational toolbar and scripts for emulating older technologies like Flash via Ruffle, directly into the HTML document of archived pages [52]. These modifications inflate the apparent size of the page, making it difficult to measure the original content accurately.

To mitigate this, we use the Wayback Machine API to replay snapshots with the navigation toolbar hidden (if_ version) [53]. By appending “if_” to the snapshot URL datetime. For example, using the Wayback Machine API, the COP29 homepage can be accessed at https://web.archive.org/web/20241111035138if_/cop29.az/en/home with the toolbar hidden. This preserves original resources while removing some of the extra elements introduced by integrated replay with the toolbar active. However, even with the toolbar hidden, integrated replay still adds resources from the address https://web-static.archive.org. Our Python script identifies and excludes resources from this address from total page size calculations, ensuring an accurate reconstruction of the original page weight.

Furthermore, when a page is archived by the Wayback Machine, it makes modifications to the original document to preserve the page. To do this, the Wayback Machine saves the original resources and then applies two kinds of adjustments. First, it makes archival linkage modifications, which alter original resource paths (URI-Rs), so they point to the archive rather than the live web, ensuring that content loads from the archived copy [51]. Second, it makes replay-preserving modifications, which adjust HTML elements and attributes to maintain page functionality [51]. These collective changes increase the size of the HTML document, since URI-Rs pathways and other attributes are lengthened, and the additional characters directly add bytes to the HTML document.

The API also allows retrieval of an unmodified version by appending “id_” to the snapshot URL datetime (id_ version) [54]. For example, the same unmodified version of the COP29 homepage can be accessed using the Wayback Machine API at https://web.archive.org/web/20241111035138id_/cop29.az/en/home. Although this version is unmodified, it introduces issues of temporal coherence, where resources may be pulled from both the live web and the archived web [50,55]. This often excludes content from CDNs, which may fail to load or be substituted with modern versions. In practice, the COP29 homepage retrieved with the id_ version fails to render some images due to temporal coherence, reducing visual completeness and leading to inaccurate measurements of page weight.

As a result, our method relies on the strengths of both approaches: retrieving page resources using the if_ version, which replays the snapshot with the toolbar hidden and preserves as many original resources as possible and retrieving the unmodified HTML document using the id_ version, which avoids modifications to the HTML. This approach allows us to distinguish between archived capture bytes retained from the original snapshot and replay-transformed bytes added during the replay process, providing a clearer measure of the page’s original size and composition. Furthermore, we accounted for extra bytes from Wayback Machine comments, which provide metadata about the snapshot, such as the date and time it was captured, adding roughly 1,500 bytes to HTML and 800 bytes to style sheets and scripts.

To benchmark our method, we calculated a final page size using a blank webpage (https://webpagetest.org/blank.html) retrieved from the Wayback Machine id_ version and compared our results against the widely recognised Website Carbon tool (available at https://websitecarbon.com), finding a margin of error within 8%. Even among existing tools, such as Beacon (available at https://digitalbeacon.co) and Website Carbon, absolute values can differ substantially (around 24.5%), but all consistently indicate the same relative trends. To account for additional bytes introduced by replay-preserving modifications, we applied a conservative 8.5% reduction. This reduction factor is intentionally conservative, based on our benchmarking, to ensure that our estimates do not overstate the true page size, reinforcing the reliability and cautious nature of our approach.

Beyond the limitations of accurately measuring archival webpages, the Wayback Machine does not provide information on whether websites were hosted using renewable energy, restricting our ability to assess the environmental sustainability of the hosting infrastructure for sites that are no longer active. In the absence of this data, we have assumed that all archived sites in this study are powered by non-renewable energy sources. For the few COP host country websites that remain active, we assessed their hosting sustainability using The Green Web Foundation’s Green Web Check (available at https://thegreenwebfoundation.org/green-web-check) to determine whether they are powered by renewable energy.

While not without limitations, this heuristic approach provides a practical way to use the Wayback Machine for retrospectively evaluating the environmental impact of websites. As “web archives [are] no longer limited to the novelty of exploring” and are increasingly used in broader research [49], our approach enables this type of analysis for the first time while accounting for the limitations of the Wayback Machine. It facilitates estimation of changes in page weight over time and highlights relative growth trends rather than absolute measurements.

4.4. Quantifying webpage energy and emissions

To convert kilobytes (KB) into grams of CO₂e, we used the Website Carbon API (available at https://api.websitecarbon.com), which is based on the Sustainable Web Design Model (SWDM) (available at https://sustainablewebdesign.org/estimating-digital-emissions), an open-source framework for estimating digital emissions, developed through a multi-year collaboration among industry leaders in SWD, including Wholegrain Digital, Mightybytes, Footsprint, EcoPing, and The Green Web Foundation [56]. The CO₂e results also graded in the API using Website Carbon’s rating system (see Table 2), which assigns a letter grade A+ to F based on the estimated CO₂e results [57]. While other methods exist and emerging research continues to improve approaches for estimating website carbon emissions, for this study we use the SWDM as implemented by Website Carbon, applying their letter score because it provides a simple, easily interpretable metric that clearly communicates website growth and environmental impact, particularly for non-technical readers.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Website carbon’s rating scale [57].

https://doi.org/10.1371/journal.pclm.0000767.t002

The SWDM used in our method has inherent limitations, as with any estimation approach. Direct conversion of bytes to emissions does not account for CPU/GPU load, since not all bytes are equal in their environmental impact: script execution, animations, and multimedia content generally require more computational power and energy than plain text. Furthermore, it does not consider user-side factors such as time spent on page, screen brightness, or browser choice, and it cannot capture dynamic features like lazy loading, conditional script execution, or progressive hydration [24]. Additionally, the model does not account for contextual changes in computational and network efficiency over time, which is particularly relevant in this longitudinal study (see Section 5).

While methodological rigour is paramount in environmental research, models like the SWDM remain valuable. Despite their limitations, they can help contextualise the ecological impacts of web technologies and reconnect users with the material consequences of digital infrastructure. In this way, they serve as useful narrative instruments, fostering greater awareness of the impact of websites, even if they sometimes fall short of comprehensive measurement.

4.5. Dataset

The URLs for each host country website vary between COPs, typically reflecting the host country’s name, event branding, or using a format such as “COP29”. There is no public database of COP host country websites: URLs were sourced from related literature and press. COP host country websites were retrieved using the Wayback Machine, regardless of the website’s activity status. The Wayback Machine was selected as it is the first large-scale public internet archive, operating since 1996 and providing the most comprehensive temporal coverage [58]. It also easy to use and can be systematically queried to construct datasets for the analysis of websites’ changing environmental impact.

The earliest available website snapshot from each COP’s start date was selected, or the closest alternative if none was available, excluding those marked as green in the Wayback Machine interface, which indicate redirects (3xx). The list of snapshot dates analysed were recorded and are listed in the dataset. Websites were then evaluated at peak operational usage, as some webpages may be later changed: the COP28 host country website was updated following scrutiny of its “low carbon” mode [40,41].

Our dataset includes 10 webpages per COP host country website: the homepage (the root URL) and nine supplementary webpages. These supplementary pages were selected based on their prominence, typically appearing in the primary navigation section of the homepage while excluding those related to login or account management. If the navigation menu webpages were exhausted, we collected URLs directly linked from the homepage in order of appearance. The choice of ten pages reflects both methodological practicality and patterns observed in recent web sustainability research. Sampling across ten pages ensures that we capture variability by including media-rich, high-traffic entry points such as the homepage alongside lighter, content-specific subpages, while maintaining a manageable and representative dataset consistent with common website navigation structures.

4.6. Missing data

Wayback Machine itself was not created until 1996, so no snapshots for COP1 (1995) and COP2 (1996) have been captured [58]. We also found no evidence they ever existed, as corroborated via email by the UNFCCC Webmaster [59]. In creating our dataset, we observed an increase in available snapshots over time (the Wayback Machine initiated its first crawl in 1996, with the option for users to suggest pages for preservation only introduced in 2001 [58]).

As previously discussed, technological limitations further affect the fidelity of archived webpages and the completeness of our dataset. Older COP host country websites that relied on now-obsolete technologies like Adobe Flash no longer render correctly when viewing the unmodified versions of webpages using the Wayback Machine API, which now uses Ruffle, a Rust-based emulator, to display older Flash-based content, which can introduce additional page size overhead [52]. In addition, differences between contemporary browsers and those used at the time of the original publication can affect how pages are replayed, including variations in script execution, rendering behaviour, and layout, further influencing the completeness of snapshots during the archiving process and replay process.

For COP3 (1997) and COP4 (1998), the host country websites used frames for their navigation, an outdated web design technique where the consistent navigation menu remains on the homepage while the main content loads within an iframe. While we were able to extract the content displayed within the iframe for each page in the menu, this method resulted in missing bytes from the homepage’s navigational menu. Consequently, our dataset for COP3 and COP4 lacks the full structural context of the original webpages, limiting the accuracy of size and efficiency assessments.

6.1. Webpage size results

Fig 1 presents the average, maximum, and minimum estimated emissions per pageview for COP host country websites. While average emissions provide a general indication of growth, they can obscure the variability between different webpages within the same website. Including maximum and minimum values illustrates the range of emissions and highlights cases where pages (often media-heavy or script-intensive) contribute disproportionately to the site’s overall footprint. This approach ensures that both central trends and outliers are represented in our analysis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Average-Max-Min chart showing changes in the mean grams of CO₂e emissions per pageview of COP host country websites (1997–2025).

https://doi.org/10.1371/journal.pclm.0000767.g001

Fig 1 reveals a substantial increase in the size of COP host country websites from 1995 to 2024. In the earlier COPs (COP3–COP14, 1997–2008), average emissions remained relatively low, with an overall mean of 0.0166g CO₂e. However, from COP15 (2009) onwards, emissions began to rise more noticeably. COP17 (2011) stands out as a major anomaly, with average emissions surging to 1.917g CO₂e, representing a 2,700% increase from COP15 (2009), due to a video file in the footer of the website. While this edge case does not reflect a steady trend, the broader shift towards heavier websites became clearer in later COPs. COP18 (2012) marked a more sustained increase, and by COP25 (2019), average emissions had peaked at 2.419g CO₂e, with a maximum of 5.031g CO₂e. Overall, average emissions from COP3 (1997) to COP29 (2024) increased by 13,778%, while the maximum emissions rose by 29,494% from COP3 (1997) to COP25 (2025).

Alongside the increase in average webpage size, the variation between individual pages has also expanded significantly. At COP3 (1997), the largest recorded webpage was calculated to create an estimated 0.036g CO₂e, whereas at COP25 (2019), the most extreme case reached 5.031g CO₂e, a 13,875% increase. However, this comparison must be interpreted with caution, as COP25’s (2019) largest page was heavily media-based (see Table 3) and structured as a single-page application (SPA), with some individual webpage links dynamically updating content on the same webpage without fully reloading. The SPA is divided by the number of subpages within the navigation, this result provides a more accurate representation. This makes it an outlier rather than a representative benchmark.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Results for the mean emissions for COP host country websites, including the mean webpage size (KB), CO₂e (g per pageview), the mean Website Carbon rating, and whether the still active website is hosted using renewable energy.

https://doi.org/10.1371/journal.pclm.0000767.t003

6.2. Webpage composition results

The results on webpage composition (Table 4) highlight clear shifts in content distribution across COP websites. Over time, the rise of diverse content types has substantially increased overall page weight. As shown in Fig 1, this growth stems from a heavier reliance on media-rich and data-intensive infrastructure, including larger page sizes, high-resolution media, extensive scripting, and resource-heavy third-party integrations. Consequently, the digital footprint of COP host country websites has expanded considerably.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Composition of COP host country websites (1995–2025).

https://doi.org/10.1371/journal.pclm.0000767.t004

To better visualise these trends, we present two separate graphs: one for the early COPs (COP3–COP14, 1997–2008) (Fig 2) and another for COP15 (2009) onwards (Fig 3). This split provides clearer detail for the earlier years, where total page sizes were relatively small, while accommodating the larger and more variable sizes of later COP websites. Notably, although COP14 (2008) and COP15 (2009) have similar mean page sizes (COP14 [2008]: 54.74 KB; COP15 [2009]: 86.74 KB; see Table 3), they appear very different across the two graphs (Figs 2 and 3), underscoring the substantial growth in page size from COP15 (2009) onwards.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Early COP Host Websites (COP3–COP14, 1997–2008): Page Size and Resource Composition.

https://doi.org/10.1371/journal.pclm.0000767.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Later COP Host Websites (COP15–COP30, 2009–2025): Growth in Page Size and Resource Composition.

https://doi.org/10.1371/journal.pclm.0000767.g003

Early COP websites (COP3–COP14, 1997–2008) seen in Fig 2, were relatively simple, primarily composed of HTML and low-resolution images, with minimal or no use of style sheets and scripts. For example, COP3 (1997) had 14.03 KB of HTML and 21.38 KB of images, with no recorded use of external scripts or style sheets (see Table 4). This pattern remained largely consistent until COP14 (2008), when more complex elements, such as style sheets and scripts, began appearing more frequently (see Fig 3).

A significant shift occurred from COP15 (2009) onwards, with a substantial increase in style sheets, scripts, and multimedia content. Pages from COP18 (2012) onward commonly have 20–50% of their weight in scripts. For instance, COP18 (2012) pages average ~45% scripts, COP21 (2015) ~45%, COP28 (2023) ~43%, COP29 (2024) ~45%. Similarly, image content grew: COP25/26/27 (2019, 2021, 2022) each have ~50–70% of weight in images. This mirrors industry trends: as bandwidth, accessibility, user expectation changes and more grew, sites added richer media (high-res images/video) and complex JavaScript frameworks (React, Angular, etc.).

COP17 (2011) stands out as an outlier, with a dramatic spike in total webpage size primarily due to a single embedded video (7383.9 KB). While script usage (276.51 KB) was also high, the exceptional size increase at COP17 (2011) was not indicative of a consistent upward trend. In contrast, later COPs, such as COP21 (2015), saw a broader rise in style sheets and scripts (1448.22 KB and 1891.4 KB, respectively). While these developments improved interactivity and visual richness, they also contributed to the increasing environmental impact of COP host country websites.

6.3. Homepage size results

Many secondary pages within our dataset were predominantly text-based, which significantly skewed the average page size downward. In contrast, homepages, often serving as the primary entry point for visitors, tended to be more media-rich, incorporating images and videos, thereby increasing webpage size and associated emissions. This pattern is clearly reflected in Table 4, where the peaks in the chart align with the results for homepage sizes displayed in Fig 4.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. The changes in mean grams of CO₂e emissions per pageview of COP host country homepages (1997–2025).

https://doi.org/10.1371/journal.pclm.0000767.g004

While mean testing facilitated the identification of structural consistencies across webpage composition (primarily in style sheets and scripts), it did not fully account for the outsized impact of homepage emissions on total website carbon footprints. Given that homepages generally receive the highest traffic volumes, their disproportionate contribution to site-wide impact warranted a distinct analytical approach. By isolating homepage emissions (see Fig 4) from the broader dataset, we aimed to achieve a more nuanced understanding of both the uniformity of COP websites in their structural elements and the real-world environmental implications of their most frequently accessed pages. The percentage increase in mean CO₂e emissions of COP host country homepages from 1997 to 2024 (COP3 0.036 kg to COP29 2.897 kg) is 7,947%, and the largest distinction in webpage size is between COP10 (2004: 0.009 kg) and COP29 (2024: 2.897 kg), with a difference of 32,089%.

6.4. Hosting assessment results

For the limited number of active COP host country websites as of the study date, we assessed whether they are hosted on renewable energy-powered infrastructure. While use of renewable-powered hosting has recently gained attention, it is important to acknowledge that such infrastructure was not widely available or actively advocated in earlier decades. For example, initiatives like the Green Grid, one of the first industry efforts to promote energy-efficient data centres only emerged around 2007 [61]. Although renewable energy sources were prevalent before this, their integration into digital infrastructure was limited.

Using The Green Web Foundation’s Green Web Check (available at https://www.thegreenwebfoundation.org/green-web-check) and Website Carbon (available at https://www.websitecarbon.com) and existing analysis we found that only four of the active/recently active websites were hosted on servers that utilise renewable energy (COP26 [2021], COP27 [2022], COP28 [2023], COP29 [2024]) [41,44,45]. However, none of these websites transparently communicated this information, indicating a gap in sustainability awareness and reporting. Given that the choice of hosting provider can reduce a website’s carbon footprint by up to 9%, integrating and communicating the use of green hosting solutions should be a priority for future COP websites [62]. Notably, COP30 (2025) was not hosted on verified renewable energy infrastructure [63]. This is particularly important as it the COP30 conference marks the midpoint between the adoption of the Paris Agreement in 2015 and the 2030 Agenda.

7.1. Increasing complexity of COP host country websites

COP websites have become significantly more complex over time, reflecting broader web trends. In 2014, the global median page size was ~ 1.2 MB; by 2024, it had grown to ~2.65 MB [47]. COP websites frequently exceed this, with many between 3–10 MB. A sharp jump occurred after COP17 (2011), where the average page size rose to ~7.4 MB and remained high in subsequent years. For example, COP25 (2019) averaged ~9.4 MB, and COP28 (2023) ~6.6 MB. Given that the average CO₂e emission per pageview globally is ~ 0.36 g (according to Website Carbon) [64], many recent COP pages emit 10 × that amount. This is echoed in Website Carbon ratings: COP3–15 (1997–2009) scored A + , but from COP17 (2011) onwards, most sites fall into the D–F range. In essence, modern COP websites are substantially more complex, heavier, and increasingly less environmentally efficient than their predecessors.

In earlier COPs (COP3–COP14, 1997–2008), websites primarily consisted of basic HTML with small images and minimal style sheets or scripts, often linking directly to PDFs rather than embedding large amounts of content within web pages (an approach that may have been more efficient in some cases). Additionally, website navigation structures have evolved significantly. Earlier COP host websites typically featured a left-side navigation menu, allowing for a more granular listing of pages. In contrast, modern mobile-first designs often consolidate multiple sections into a single page, where previously, the content may have been spread across two or three distinct pages. While this approach enhances accessibility on mobile devices by simplifying navigational menus, it can also contribute to information being less discoverable and contribute to the overall material downloaded.

7.2. Expansion of multimedia and scripts

From COP15 (2009) onwards, the use of scripts, and multimedia elements expanded significantly. For instance, COP17 (2011) saw a dramatic increase in webpage size due to a high volume of scripts (276.51 KB) and an embedded MP4 video file adding 7,383.9 KB to the total webpage size (see Fig 2). Similar trends were observed in COP21 (2015) and COP25 (2019), where style sheets and scripts became dominant contributors to webpage size, leading to a notable rise in equivalent emissions per pageview.

Furthermore, the use of rich media increased sharply. This aligns with observations from the Web Almanac, which notes that modern page bloat is largely driven by larger media. Our COP data reflects this clearly: for example, COP25’s (2019) images alone total ~7,089 KB (see Table 4), well above the ~ 1 MB typical image weight reported in the Almanac [47]. Supporting this trend is the idea that, as past limitations in network speed and device performance have fallen away, websites have become increasingly resource-intensive, driven by high-quality visuals and complex JavaScript frameworks such as React and Angular. This reflects Jevon’s Paradox, where greater efficiency or capacity leads to increased overall consumption rather than a reduction [20].

7.3. Estimated emissions from website traffic

A key limitation of this study is the lack of access to website analytics, which prevents precise estimations of total carbon emissions from website traffic. To approximate impact, we relied on in-session participant data from the UNFCCC, assuming each delegate accessed only the homepage of the host country website (see Fig 5). This provides a baseline for understanding the minimum emissions generated, though actual figures are likely much higher when considering additional visitors. Since COP30 has not yet taken place at the time of this study (it is scheduled to start on 10 November 2025, with the COP30 Heads of State Summit beginning on 6 November), participant numbers are not yet available [65]. Therefore, it is excluded from this analysis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Estimated CO₂e emissions from COP host homepage views based on UNFCCC’s in-session participant data COP3, (1997) to COP29 (2024).

https://doi.org/10.1371/journal.pclm.0000767.g005

Table 5 shows that estimated homepage emissions rose alongside the number of in-session participants across COP events. At COP3 (1997), with 3,969 participants, homepage emissions were just 0.14 kg CO₂e—roughly the amount a mature tree absorbs in two days [66]. In contrast, COP29 (2024), despite having the second-highest participant count of 40,335, generated an estimated 116.85 kg CO₂e from homepage visits alone. This represents an 83,400% increase from COP3 (1997), illustrating how larger page sizes dramatically amplify CO₂ emissions as site traffic grows, creating an exponential rise in the digital footprint of COP host country websites. While we are cautious about relying on estimates or metaphors, and prioritise reporting grounded in empirically derived data, indicative figures can support interpretation of scale. For example, based on standard CO₂-sequestration rates, it would take approximately five to ten mature trees a year to absorb the 116.85 kg CO₂e generated by homepage visits during COP29 (2024) [66].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Estimated CO₂e emissions from COP host country website homepage views based on UNFCCC’s in-session participant data, COP3 (1997) to COP29 (2024).

https://doi.org/10.1371/journal.pclm.0000767.t005

Crucially, these figures only account for registered in-session participants and do not include additional visitors accessing the site independently, or visits to other pages beyond the homepage. For instance, COP28 (2023) was estimated to have attracted over 400,000 total visitors a difference just over 320,000 from those recorded by the UNFCCC, meaning actual emissions could be significantly higher [9,67]. If the same scaling applied, homepage visits to the COP28 (2023) host website alone could have generated approximately 649.32 kg CO₂e, an amount that would take an estimated 26–65 mature trees a year to absorb [66].

7.4. The digital footprint of COPs

Table 5 highlights the need to reconsider not just the organisation of the physical event but also the growing digital footprint of COPs and its environmental impact. As the conference expands, rising participation increases physical resource consumption through attendance while also driving higher emissions from the associated digital infrastructure, reinforcing the interdependence between its logistical and online dimensions. These findings align with the ECBI’s call to scale down the conference, not only in its physical presence but also through a more sustainable digital strategy [10]. Unlike other proposed interventions by the ECBI that are harder to unwind due to the ratcheting effect of “summit nostalgia [10],” reducing emissions from COP host country websites presents a practical, low-disruption opportunity to lower the summit’s overall footprint independent from the conference’s organisation itself.

7.5. The trade-off between digital and print materials

One argument in favour of COP host websites’ growth is that they may offset emissions by reducing reliance on printed materials and promoting dematerialisation. “Paper-lean” and “paper-light” policies put in place by United Nations meeting organisers throughout the past decade have been incorporated into UNFCCC sustainability guidelines with “good results” [15]. However, the extent to which COP websites have effectively replaced printed materials remains unclear. As our research has shown, “digitalisation’s promise of dematerialisation has not yet materialised [1].” Digitalisation is often framed as a greener alternative, reinforcing a false divide between the physical and digital, whereas they are deeply interconnected, obscuring the natural and abiotic resources required by computing for their operation [23],“the cloud is not floating; it’s grounded in enormous data centres, deep-sea cables, and largely fossil-fuelled power grids” [68].

The Digital Economy Report 2024 explicitly acknowledges this issue, stating that they had “failed to make the connection between physical and digital; instead, they are one and the same [4].” What this reveals is that the trade-off between digital and print is not simply a question of substitution or efficiency. While digital systems may reduce some emissions associated with printing and distribution, they introduce their own environmental costs in ways that are often hidden or externalised. While prior research has compared the environmental impacts of print materials, and digital formats through the life cycle assessments [69], analysis falls outside the scope of our current study. Nonetheless, it is important to recognise this broader body of work, which further reinforces the need for more nuanced and systemic evaluations of digital infrastructure and its material consequences. Recognising this entanglement challenges the perception that digitalisation inherently reduces environmental impact and calls for more systemic thinking about both formats’ infrastructural and ecological footprints.

8.1. Enforcement of a page weight budget

Our first recommendation is the enforcement of a “page weight budget” [30], requiring all webpages to achieve at least a B rating (≤ 0.341g CO₂e) on the Website Carbon Calculator (available at https://websitecarbon.com), balancing ambition with feasibility. Given the difficulty of achieving an A+ rating for media-heavy websites, relying solely on Website Carbon rating categories may discourage participation, as lower grades could be perceived negatively. Standardising CO₂e disclosure per page provides a more transparent and objective measure of environmental impact without the stigma of a letter grade, ensuring that host countries remain engaged in sustainability efforts while maintaining accountability. This can be calculated using The Green Web Foundation’s CO2.js Playground (available at https://observablehq.com/@greenweb/co2-js-playground). Finally, tools such as CO2.js (available at https://thegreenwebfoundation.org/co2-js) or Website Carbon’s API (available at https://api.websitecarbon.com) should be used to monitor and display equivalent emissions and total page size (KB).

8.2. Adherence to existing SWD frameworks

To achieve the aforementioned we recommend following existing SWD frameworks, including the W3C Web Sustainability Guidelines (WSG) (available at https://w3c.github.io/sustainableweb-wsg) and Sustainable Web Design Principles (available at https://sustainablewebdesign.org/guidelines). These frameworks provide methodologies to reduce resource consumption, eliminate redundancy, and improve performance efficiency.

8.3. Hosting using servers powered by renewable energy

COP host country websites are to be hosted using sources with the lowest possible carbon intensity (ideally generated by wind or solar rather than from non-renewable sources). This can be verified through tools like The Green Web Foundation’s Green Web Check (available at https://thegreenwebfoundation.org/green-web-check) and its impact in reducing the impact of webpages can be assessed using The Green Web Foundation’s CO2.js Playground (available at https://observablehq.com/@greenweb/co2-js-playground). Where feasible, host countries should prioritise domestic hosting providers that meet sustainability standards, reducing data transfer-related emissions This information should be stated publicly, for example in the website footer or in the colophon (see Section 8.5).

8.4. Ensure that navigation is well-structured

To optimise accessibility and minimise the environmental impact of host country websites, site architecture should be streamlined. While we did not conduct a formal user navigation analysis, improving navigation structure remains a widely recommended intervention, as highlighted in W3C WSG guidelines. Navigation should be intuitive and well-structured, incorporating search features that enable users to find information efficiently. According to the W3C WSG, ensuring clear navigation and wayfinding is an advisory technique to meet the W3C’s WSG success Criterion 2.8: “Ensure that navigation and wayfinding are well-structured [29].” Although no specific limit is set on the number of key links in the main navigation, best practices recommend keeping a website’s sitemap concise and easy to navigate. Poor navigation can negatively affect user experience, leading to longer website sessions and unnecessary page visits. These behavioural factors indirectly increase emissions and underscore the importance of efficient information architecture alongside technical sustainability measures such as managing page weight.

8.5. Publishing a colophon

A colophon—traditionally a publishing note detailing production information—must be included in the website footer and potentially linked via the CO₂e disclosure to contextualise emissions data. This colophon should document web sustainability measures, development methodologies, and key tools used, ensuring transparency. Any additional sustainability measures beyond those outlined in this guidance should also be explicitly stated to prevent ambiguity or unverified claims. Aligning with UNFCCC COP procurement policies, this approach enhances accountability and facilitates knowledge-sharing within sustainability and web development communities.

Furthermore, including a simple colophon, or adopting more formal methods such as The Green Web Foundation’s Carbon.txt or [70], for EU-based COPs, aligning with the Corporate Sustainability Reporting Directive (CSRD) [71], can help establish a more standardised and transparent approach to sustainability reporting. Historically, sustainability reporting was not prioritised, as evident in archived websites. Our Wayback Machine-based method addresses this gap, but had robust reporting policies been in place, simpler assessment methods would have sufficed. Implementing such policies now would not only reduce the need for retrospective analysis but also enable more efficient evaluation of sustainable web practices in the future.

8.6. Post-event reporting: Website analytics and environmental impact

Finally, post-event reporting for COPs must integrate website analytics, including total page weight, emissions, and user interactions. Our analysis relied on proxy methods and archival data to estimate the impact of COP host country websites, but this approach is limited by gaps, inconsistencies, and modifications introduced during the archiving process (see Section 4.5).

Our analysis of in-session participant data and homepage views (see Table 5 and Fig 4) highlights the significant digital impact of COP websites. Given the ECBI’s recommendation to reduce in-session participants in favour of online engagement [72], it is critical to quantify the extent of this shift and its environmental implications. However, without precise website traffic data, we are left estimating emissions based on participant numbers alone, an approach that fails to account for the potentially vast number of additional online visitors. For instance, while COP28 (2023) had around 70,000 in-session participants, estimates suggest it attracted over 400,000 visitors in total [67]. If homepage visits significantly exceed in-session participant numbers, the actual emissions from host country website could be far greater than our calculations suggest.

This gap in reporting raises a crucial issue: without accurate analytics, efforts to reduce website emissions by optimising page size may be rendered ineffective if traffic levels are far higher than assumed. We recommend adopting web sustainability focused analytics tools such CarbonClicks (available at https://carbonclicks.io) or Cabin (available at https://withcabin.com) to calculate real-world engagement. As with in-session participant records, website site traffic must be measured and published to provide an accurate basis for impact assessment, optimisation strategies, and, where necessary, offsetting residual emissions.