How Iszdown tracks outages

Iszdown is an independent website for checking service availability. We combine user-submitted outage reports with estimated down-report activity to help you understand whether other people may be experiencing problems.

Sources, shown separately

Outage reports are reports submitted to Iszdown, plus identified external report feeds when available. Repeat submissions from the same service and IP address are limited within each ten-minute reporting window. Tweets, website checks, and autocomplete suggestions do not count as submitted outage reports.

Estimated down reports use Google Trends searches that express problem intent, such as “ChatGPT down” or “ChatGPT not working.” These are relative US activity scores from 0 to 100. A value of 100 is the peak within that query and time window; it does not mean 100 people submitted reports. Scores from different services or separately fetched windows are not absolute volumes and should not be added together.

Official status, website checks, and recovery

Public official status feeds are collected every two minutes for supported providers. Distinct products use explicitly mapped components; missing product coverage is unknown. Provider announcements can lag or omit individual customer issues. Incident context from infrastructure providers does not establish a dependency or root cause.

Our website-check queue runs each minute, targeting five-minute checks for the top twenty services and hourly checks for others with a public website address. Timeouts, DNS, TLS, and HTTP errors are recorded separately. Authentication and bot restrictions (401, 403, 429) are inconclusive and excluded from the success rate. Website checks measure HTTP response headers from one US server; they do not test full page load, sign-in, or application functions. Success percentages describe collected samples, not continuous uptime. Gaps stay unknown. Measurements are retained for seven days.

Slowdown flags need three recent successful samples and five earlier baseline samples, a doubling in median response time, and at least 500 ms extra latency. This cannot identify server load as the cause. Recovery distinguishes provider resolution from three subsequent successful website checks. Optional regional diagnostics use Globalping community probes, reuse results for fifteen minutes, and have a shared sixty-test hourly limit.

Autocomplete and emerging services

One shared collector makes at most five autocomplete requests per minute, using US English settings without signed-in personalization. It targets neutral service-name prefixes and rotates alphabet discovery queries. The top five services have five-minute refresh targets, the next fifteen twenty-minute targets, and other services hourly or daily targets. Queue capacity and provider backoff can delay collection. Suggestions reflect search patterns, can lag incidents, and are not counts of people or reports. A first observation establishes a baseline. New problem phrases need another recent query with matching problem intent before bringing forward a Trends check; global provider backoff still applies. Snapshots and seven days of bounded history support comparisons.

What the 24-hour graph shows

Charts cover a rolling 24-hour window, grouped into ten-minute intervals. Report counts are summed; Google Trends samples are averaged within each interval. Missing Trends samples stay blank. Partial samples are excluded. Each successful Trends response replaces that service’s previous full window so independently scaled snapshots are not spliced together.

Updates and missing data

Popular services and services attracting new attention receive priority. The collector checks for eligible work every thirty seconds. The highest-ranked services have a five-minute refresh target, followed by fifteen-minute, hourly, and daily targets for less popular services. Provider limits, sparse search volume, failures, and backoff can delay a refresh. The last successful check is shown on each page.

A report increase can suggest possible problems. A surge in estimated activity can indicate growing concern. Neither proves that every user is affected. If recent data is unavailable, the page says so. No unusual activity and no submitted reports do not guarantee that a service works for everyone.

Country-specific estimates for top services

The top five services, ranked by configured popularity and recent page activity, receive additional country-filtered Google Trends queries for the United Kingdom, Canada, Australia, India, Germany, Brazil, and Japan. US estimates remain on the main service pages. The top-five group is reevaluated every fifteen minutes. Regional queries have an hourly refresh target and share the existing collection budget; roughly one of every three eligible polling slots is reserved for countries outside the US. Provider throttling pauses both queues together. New country pages stay out of the sitemap and are marked noindex until usable country-specific data exists.

Country pages currently use English text and reference local time zones. German, Portuguese, and Japanese problem-intent phrases are added to English queries in those countries. Different keyword sets and separately normalized 0–100 scales cannot establish cross-country report counts or rank countries by affected users. Global official incidents, submitted reports, and measurements from our US server are not relabeled as local observations. Missing country data is not replaced with US data. Previously selected services can retain dated country observations, with collection marked paused.

New services and distinct products

Outage-related keywords seen in separate observation windows can create new service pages automatically. Known spelling variants are normalized, while distinct products keep distinct pages. New published pages enter the service directory and sitemap automatically. A new page may initially have too little data to determine current status.

Citing Iszdown

Link to the specific service’s canonical page, include the time you accessed it, and identify whether you are discussing recorded outage reports or estimated activity. Use the displayed observation time when available. Recheck the page before describing a service’s current condition: status can change quickly.

For example: “Iszdown’s ChatGPT page showed increased estimated down-report activity when accessed at [your access time].” Do not describe this estimate as an official outage announcement or as Downdetector’s submitted-report counts.

We are not affiliated with the services we monitor, Google, or Downdetector. Product names belong to their respective owners. For service-specific confirmation, also consult the provider’s official status information.

Google Trends data guidance · AI discovery and citation guide