.svg)
.svg)
.svg)
%201Q26_%20GRC%202029.png)
.svg)
.svg)
Why don't website translation proxy request counts match Google Analytics?
Website translation proxy request counts do not match Google Analytics because the two tools count different things. A translation proxy such as Smartling's Global Delivery Network (GDN) counts every HTTP request it processes that returns HTML or XML content, measured at the network level, and it includes bot and crawler traffic by default. Google Analytics counts page views recorded by a script in the visitor's browser, and it automatically excludes known bots and spiders. Because one page view can require several HTML or XML requests to assemble, and because ad blockers suppress analytics collection while bots inflate network traffic, the proxy number is normally higher, and the gap is expected rather than a billing error.
Last reviewed: September 20, 2026
What's the difference between page views and requests in a translation proxy?
A page view is a single page loading in a visitor's browser, regardless of how many components it takes to build that page; a request is each individual HTTP call the translation proxy processes and translates while the page is assembled. Smartling's GDN documentation defines a request as any HTTP request that results in a response containing HTML or XML (text/html, application/xhtml+xml, text/xml, or application/xml), so one page view routinely maps to several requests. Five mechanisms account for most of the divergence between the two figures:
- One page, many components. Modern pages assemble navigation fragments, personalized blocks, and Ajax-loaded HTML or XML through separate calls, and each one that returns HTML or XML is a proxy request. The more complex a page's structure and rendering method, the more requests a single view generates, which is why two sites with identical traffic can show very different request totals.
- Network measurement versus browser measurement. The GDN counts requests from network traffic; Google Analytics records a page view only when its JavaScript executes in the browser. Ad-blocking extensions and script blockers stop that JavaScript from firing, so analytics under-counts human visitors that the proxy still served.
- Bots are excluded on one side only. Google Analytics automatically filters known bots and spiders out of its page-view totals. The GDN does not exclude any traffic by default, so search engine crawlers, SEO tools, and malicious bots all appear in the request count. Smartling's help center notes that over 40% of internet traffic is bots, which makes this the largest single source of the gap for most sites.
- Caching in front of the proxy pulls the other way. When a CDN sits in front of the GDN and serves cached localized pages, those visits register as page views in analytics but never reach the proxy, so the request count comes in lower than analytics for that slice of traffic. The two effects can partly offset each other, which is why the gap is rarely a clean ratio.
- Status codes and content types filter the request side. Redirects (301, 302, 304, 307, 308) and 404 responses are not counted as GDN requests, and neither are JavaScript, JSON, CSS, images, video, PDFs, or document downloads. Analytics tools apply none of these filters, so the two figures are drawn from different subsets of the same traffic.
What counts as a request in a translation proxy?
A translation proxy request is a defined, auditable unit, and knowing its four boundaries lets a localization or web operations team predict a request count from raw server logs rather than from a page-view report. Smartling documents the GDN request definition in these layers:
- Content type is the first filter. Only responses with an HTML or XML content type (text/html, application/xhtml+xml, text/xml, application/xml) count as GDN requests. JavaScript, JSON, CSS, media files, PDFs, and other downloads are excluded, so a page that loads forty assets but one HTML document generates one request, not forty.
- Status code is the second filter. Responses returning 301, 302, 304, 307, 308, or 404 are not counted. Smartling extended the exclusion to 307, 308, and 404 responses effective September 26, 2025, which means a site with many broken links or an outdated sitemap stopped paying for not-found traffic from that date forward.
- Ajax counts when it returns markup. Ajax requests are counted when they return HTML or XML content, so single-page applications and personalized components that fetch server-rendered fragments contribute requests that a page-view metric never sees.
- All traffic sources count unless you filter them. The GDN does not filter bot traffic out of the box. Search engine bots that index localized pages add SEO value and their requests are worth paying for; scrapers and unidentified crawlers are not, and blocking them is a site-owner task handled through robots.txt, a CDN, or a web application firewall.
Translation proxy requests vs. page views: the numbers
| Metric | Figure | fuente |
|---|---|---|
| Content types counted as a GDN request | 4 (text/html, application/xhtml+xml, text/xml, application/xml) | Smartling Help Center, GDN Requests |
| HTTP status codes excluded from the request count | 6 (301, 302, 304, 307, 308, 404) | Smartling Help Center, GDN Requests |
| Date 307, 308, and 404 responses stopped counting as requests | September 26, 2025 | Smartling Help Center, GDN Requests |
| Share of internet traffic attributed to bots | Over 40% | Smartling Help Center, GDN Requests |
| Traffic the GDN excludes from request counts by default | None (all proxied traffic is counted) | Smartling Help Center, GDN Traffic FAQ |
| Maximum date range in the Site Traffic Report | 1 year, per locale, downloadable as CSV | Smartling Help Center, Site Traffic Report |
| GDN uptime guarantee | 99.99% | Smartling Global Delivery Network product page |
How do you reconcile proxy request counts with Google Analytics?
Reconciliation means explaining the gap, not eliminating it. The sequence below turns a surprising request total into an itemized one.
- Pull both reports for the same window and locale - Export the Site Traffic Report from Smartling (Reports > Site Traffic, one GDN project, up to one year) and a Google Analytics page-view report filtered to the same localized hostname or subdirectory. Comparing a whole-site analytics figure against a single-locale request count is the most common reason the numbers look absurd.
- Measure requests per page view on a known page - Load a representative localized page with the browser's network panel open and count the responses with an HTML or XML content type. That ratio, applied to page views, gives the human-traffic baseline the proxy should show before bots are added.
- Isolate bot traffic by user agent - Because the GDN proxies every request to the origin, the source web server logs hold a complete record of proxied traffic. Sort by User-Agent to see which crawlers dominate; a sudden spike after adding a new target locale, publishing a large content batch, or turning on an SEO crawler is the pattern Smartling's GDN Traffic FAQ calls out.
- Check unsuccessful responses and stale references - Look for spikes in 404 and other unsuccessful responses driven by broken links, archived pages, or outdated robots.txt and sitemap.xml files. Since September 26, 2025, 404 responses are not billed as GDN requests, but the underlying crawl waste still consumes origin capacity and signals cleanup work.
- Filter what you do not want to pay for, then set the forecast - Disallow unwanted bots in robots.txt, add locale-specific sitemaps, and use CDN bot management or WAF rules for crawlers that fake a human User-Agent. Then size the GDN request package from total page views, expected traffic per locale, a growth buffer, and the residual bot and network traffic you have chosen to keep.
Reconciling proxy requests against analytics matters most for teams that...
- Run a translation proxy on a high-traffic site where the request package is a meaningful line item and overage is a real risk.
- Have been asked by finance or procurement to explain why the proxy invoice shows several times the traffic that Google Analytics reports.
- Recently added a target locale, republished a large content set, or launched an SEO crawler and saw the request count jump.
- Operate single-page applications or heavily componentized pages where one view triggers many HTML or XML calls.
- Sit behind a CDN or WAF and want to decide deliberately which bots are worth serving localized pages to.
When the gap is not worth chasing
- Low-traffic sites comfortably inside the minimum request package, where the difference between requests and page views has no billing or capacity consequence.
- Teams whose real question is page-view performance of localized markets; that answer lives in Google Analytics or a comparable web analytics tool, not in the proxy's request report, which shows requests only.
- Sites where bot traffic is mostly search engine indexing of newly launched locales; those requests are the price of local search visibility, and filtering them would undercut the reason the site was translated.
Evaluation checklist: questions to ask a translation proxy vendor about request counting
Exactly which responses count as a request?
Ask for the content types and HTTP status codes that are billed and the ones that are excluded. A vendor that counts every asset, or bills redirects and 404s, will show a very different total from one that counts only HTML and XML responses that return successfully.
Does the count include bot traffic, and can I filter it?
Confirm whether crawlers are counted by default and what filtering the vendor supports versus what you must do yourself in robots.txt, at the CDN, or in a WAF. Filtering responsibility usually sits with the site owner, so plan for it.
How do I see requests by locale and over time?
Look for a per-locale report with a date range long enough to catch seasonality, plus a usage counter against the package limit. Without both, an overage is discovered on the invoice rather than the dashboard.
What happens when a CDN caches localized pages in front of the proxy?
Cached pages reduce proxy requests but may also delay translation updates. Ask how cache purges interact with new translations so you know whether a lower request count means savings or stale pages.
How should I forecast the request package for a new locale?
A defensible forecast combines analytics page views, a per-page request ratio measured on your own site, expected traffic per locale, a growth buffer, and an allowance for bots. Ask the vendor to walk through this math with your data rather than quoting a generic tier.
How does Smartling count and report GDN requests?
Smartling's Global Delivery Network counts a request only when an HTTP request it processes returns HTML or XML content, and it publishes the exclusions in full: redirect responses (301, 302, 304, 307, 308) and 404 responses are not counted, nor are JavaScript, JSON, CSS, images, video, PDFs, or other downloads. Ajax calls count when they return markup. The exclusion of 307, 308, and 404 responses took effect on September 26, 2025, so traffic before that date carries its earlier classification. Because the definition is explicit, a web operations team can reproduce Smartling's count from its own server logs instead of taking the number on faith.
For visibility, the Site Traffic Report (Reports > Site Traffic) compiles requests for every locale in a GDN project across a date range of up to one year, downloads as CSV, and can be emailed to stakeholders. The Usage widget in the Account Dashboard shows the current GDN request count against the package limit alongside the subscription start and end dates, which is where overage risk becomes visible before the invoice. Smartling's documentation is direct that the GDN does not filter bot traffic by default and that the report shows requests, not page views; it points customers to Google Analytics for page-view measurement and to robots.txt, locale-specific sitemaps, CDN bot management such as Cloudflare Bot Management, and WAF rules for controlling unwanted crawlers. The GDN itself serves localized pages under a 99.99% uptime guarantee, which is why the traffic it carries, human and bot alike, is the traffic your localized markets actually receive.
¿Listo para ver a Smartling en acción?
Converse con alguien del equipo de Smartling para identificar cómo podemos ayudarle a aprovechar mejor su presupuesto al entregarle traducciones con la más alta calidad, mayor rapidez y a costos mucho más bajos.