> ## Documentation Index
> Fetch the complete documentation index at: https://docs.open.cx/llms.txt
> Use this file to discover all available pages before exploring further.

# Help center analytics

> Measure help center readership, article and search performance, recorded content activity and translation gaps. Compare periods and inspect contributing activity.

Open a help center and choose **Analytics** beside **Settings** in the editor. You can also open **Analytics** from its settings sidebar. You need permission to read that help center; access to other reports is not required.

> [!NOTE]
>
> This guide describes the Overview, Articles, Searches, Content activity and Translation gaps
> reporting update. Reader data starts with the updated visitor tracking; earlier reader activity is
> not imported. **First observed data**, available from **About this data**, shows where recorded
> reader history begins. Content activity uses its own recorded editorial history.

## Overview

The opening summary highlights the change in recorded page views. The summary shows the recorded total without a growth claim when current history is incomplete or the comparison count is unavailable or zero.

**Page views**, **Unique visitors** and **Article views** give the period totals. **Article engagement** shows a two-color donut for helpful and not-helpful votes, with helpfulness in the center. Exact counts, comparison changes and average visible reading time remain beside it. Hover a colored segment to see its vote label and count. An empty ring means no votes; a measured 0% means votes were recorded but none were helpful. Select **Page views** or **Unique visitors** to open a drawer with the daily or language breakdown. Closing the drawer returns you to the same place in Overview. Select **Article views** or an engagement metric to open the contributing article table.

In **Readership over time**, switch between **Page views** and **Unique visitors**. The solid line shows the selected period; the dashed line shows your chosen comparison when daily comparison history is available. Both date ranges and durations appear above the chart. Periods align from day one, and the tooltip shows each point's actual date. Different-length periods retain their own lengths; neither line is stretched or converted to a daily average. Dates without recorded activity appear as gaps, and a shorter period ends at its own last day. With **No comparison**, only the selected period appears.

* **Page views** count page displays, including article and non-article pages. Reloading or returning to a page creates another display. Repeated delivery of the same display does not add views.
* **Unique visitors** count anonymous browsers in this help center over the whole selected period. They do not identify people: another browser, device or cleared browser data may count separately.
* **Article views** count displays of article pages. **Unique readers** count browsers that displayed those articles.

**Most read articles** previews up to six articles with views in the selected period, ordered by article views. It uses the same dates, time zone and language as Overview. Select an article for its detailed report, or choose **All articles** for the full sortable table. Article and topic selections from other reports do not narrow this preview.

**Readership by language** ranks languages by recorded page views. The breakdown drawers show 10 rows per page. Use **Previous** and **Next** to move through it. Pagination changes the visible rows, not the period totals or downloads. Daily and language breakdowns calculate their own unique counts. The same browser may appear in several rows, so adding the rows does not give the whole-period unique total.

## Article performance

Choose **Articles** to sort article performance, inspect a current topic and its subtopics, or open an article's detailed daily and language breakdowns. The table starts with 25 articles per page, ordered by views. Select a column heading to change the ordering. On smaller screens, use **Sort by** and the direction arrow above the article list. All six metrics share a single header row; the helpfulness bar shows the share of votes marked helpful.

The individual article report separates readership and reader feedback into labeled metric sections. Open **About article metrics** for the definitions and reading-time measurement limits. Daily and language tables retain all six metrics and show 10 rows per page.

Topic filters use the article's **current placement in the recorded language**. Moving an article changes which topic includes its historical activity. Deleted articles remain in the unfiltered article list and their article detail, but their editor link is unavailable. Current-topic filters follow the current topic tree and omit deleted articles.

### Visible reading time

Reading time counts time while the article tab is visible, without an idle cutoff. It does not establish whether someone was actively reading. The average uses measured article views, including a measured zero; a view with no measurement is **Unavailable**.

Time is recorded periodically and when the reader hides or leaves the article. Hidden time is excluded. An abrupt browser termination can lose the final interval.

### Helpfulness

Readers can vote **Helpful** or **Not helpful** once per article view. A later article view can vote again. Repeated delivery of a vote does not add another vote.

**Helpfulness** is helpful votes divided by all votes. **No votes** is different from 0%: 0% means votes were recorded and none were helpful.

## Search performance

Choose **Searches** to see how often readers search, how many results they see and whether they open an article. Each completed, displayed result set counts as a search, including autocomplete. Failed, cancelled or superseded requests do not count as searches or zero-result searches. Repeating a query in a later search counts again.

* **Average displayed results** averages the result count shown for each recorded search.
* **Zero-result rate** is searches displaying no results divided by all matching searches.
* **Click rate** is searches with at least one result click divided by all matching searches. Opening several articles from one search counts once toward this rate.
* **Result clicks** counts article opens from matching searches. Repeated delivery of an open does not add another click.

**Search activity over time** leads with daily stacked bars: searches with at least one click and searches without clicks. Their combined height is the daily search count. Zero-result counts remain separate, so the chart does not assume that result counts and click events are mutually exclusive.

Filter by a literal part of a search term, search surface or zero-result status. Type in **Search term contains** to update the report automatically after a short pause. Clearing the field removes only the text filter; the selected search surface and displayed-results filter stay active. Terms group regardless of capitalization and surrounding spaces; accents, punctuation and spaces inside the term remain significant. Select a term with searches in the reporting period for its exact contributing searches. A term with activity only in the comparison period is labeled **Only in the comparison period**; choose that date range to inspect its searches. Select a metric to inspect matching searches, searches with clicks, zero-result searches or **Articles opened from search**. These views preserve the applicable date, language and search filters. Choosing **Searches** in the main report tabs opens the top of the report and retains your selected search view. Selecting a metric or search term takes you to its matching rows.

If **Articles opened from search** is empty while **Displayed results** is set to **Zero results**, the empty state explains the active filter. Choose **Show all searches** there to include searches that displayed articles, keeping your other filters.

The **Recorded searches** view starts with the search query, followed by its date, time, language, surface and result counts. Click-rate bars show the proportion of searches with an article opened.

Search activity belongs to the time the results were displayed. A later article click remains attributed to that originating search, so older periods can update as delayed clicks arrive. The opened-article table shows click counts and how many searches produced them. It does not reconstruct every displayed result or measure how often a specific article appeared. Deleted articles retain their recorded clicks.

An empty search population shows unavailable averages and rates, rather than 0%. A recorded search with no clicks has a measured click rate of 0%.

## Needs attention

In **Overview**, **Needs attention** highlights articles and search terms that meet these rules:

| Finding | Rule |
| - | - |
| Low helpfulness | Below 70% helpful, with at least 100 views and 10 votes. |
| Declining helpfulness | A drop of at least 10 percentage points against your selected comparison period, with at least 10 votes in each period. |
| Searches with no results | At least 10 searches for the term, with 50% or more returning no results. |
| Searches with few article opens | At least 10 searches for the term, with fewer than 20% opening an article. |

Each finding shows the counts and rule that triggered it. Select an article to open its report, or a search term to see its recorded searches. Findings follow the selected dates and language. Declines appear only when comparison is enabled and that period's history is available. A percentage-point drop compares helpfulness directly: 80% to 70% is a 10-point drop.

The preview shows up to three articles ranked by not-helpful votes and three search terms ranked by frequency. The Overview download includes every matching finding. No findings can mean there is not enough recorded activity; it does not certify the content's quality. These rules do not change, publish or translate content automatically.

## Traffic sources

**Overview → Traffic sources** shows the original referring domains that brought readers into the help center, ranked by page views. Each row also shows unique visitors. The entry source stays with page views during internal navigation, including language changes. A new external entry replaces the source. If the browser prevents local storage, attribution continues through navigation within the current page but can be lost on a full reload.

Only the hostname is retained, including any subdomain. URL paths, query strings and fragments are discarded. Referrers do not establish a person's identity or prove whether a human or an automated agent visited.

* **Direct or unknown** means no usable referrer was available, including browsers that suppress it.
* **Source not collected** means that activity came from before source tracking or an older reader. It is not counted as direct traffic.

Use **Previous** and **Next** to see more sources. Unique visitors are counted separately for each source and cannot be added to reconstruct the overall total. Historical source information is not reconstructed.

Choose **AI referrals** beside **All sources** to see recorded readership attributed to recognized AI services. Each row keeps the original domain visible alongside the service name, page views and unique visitors. The view follows the selected dates and language. The same labels also appear in **All sources**.

Recognition covers dedicated domains for ChatGPT, Claude, Perplexity, Gemini, Microsoft Copilot and Grok. General search, company and social domains are not assumed to be AI referrals. If an AI tool or browser omits the referrer, that activity remains **Direct or unknown**; no matching rows does not mean nobody arrived from an AI tool. This view does not measure automated crawlers, citations in AI answers or your OpenCX agents using articles.

Overview CSV and Excel downloads include **all** traffic sources, with an `ai_referral_provider` column for recognized services, regardless of the selected source view. Existing recorded domains can receive a label without changing their underlying events or counts; missing historical referrers cannot be recovered.

## Dates, filters and comparisons

Reports start with the last 30 days. Open **Date range** to choose a rolling preset, a calendar period or **Custom range**. The **Date range** and **Compare with** controls sit beside **Export report**, above the report tabs, as compact buttons showing each period’s dates. Open a picker to see the period duration. Time zone and **Filters** sit beside the report tabs. These controls wrap into stacked rows on smaller screens. Open **Compare with** to choose **Previous period**, **Custom range**, or **No comparison**. Open the **Filters** popover to choose an optional language. Article reports also include **Current topic** and metric conditions in this popover. Select an active filter chip to clear that filter, or choose **Clear filters** to restore the default filters while keeping the selected calendar dates. The date range can cover up to one year.

In **Articles → Filters → Article metrics**, choose **Add condition**. Select a metric, a condition (**Greater than**, **At least**, **Equal to**, **At most**, or **Less than**) and a value. Choose **Apply filters** when ready. You can filter views, unique readers, visible reading time, helpful votes, not-helpful votes and helpfulness. Enter reading time in seconds and helpfulness as a percentage from 0 to 100.

Every condition must match the article's values in the main selected period. For example, combine **Unique readers greater than 100** with **Not helpful at least 5** to find widely read articles that need attention. Comparison values remain visible for matching articles; conditions do not filter the comparison period independently. Missing reading-time measurements and articles without votes do not match numeric conditions for those metrics, including conditions for zero.

Conditions apply to the full article list before pagination and to its CSV/Excel download. Remove an individual condition using its active filter chip, or use **Clear filters** to clear them all. Opening an article shows its complete daily and language breakdowns for the selected dates.

Open an article for its individual report, with separate daily trend cards for article views, unique readers and visible reading time, followed by the reader-feedback donut. Hover a trend for its date and value. Reading-time charts leave gaps for unmeasured days. Blue bars in the article table compare views within the current page. Article filters do not change the site-wide Overview totals. Translation gaps have their own visible target-language, topic and status filters, with no historical period.

**Previous period** is selected by default. It uses the same filters over an equal-length period immediately before the selected one. Its dates update when you change the main date range or time zone. For example, October 14–November 12 compares with September 14–October 13.

Choose **Custom range** to select your own comparison dates, then select **Apply comparison**. Both periods can cover up to one year and can have different lengths; the picker shows both durations. Percentage changes compare the reported totals or rates directly, without converting totals to daily averages. Custom dates stay fixed when you change the main period. Changing the time zone preserves the calendar dates you selected for both periods.

Choose **No comparison** to view the main period on its own. If a daylight-saving change shifts automatic comparison boundaries away from midnight, the label includes times. **New** means the comparison count was zero and the selected-period count is positive; unavailable comparisons show a dash. Exports include the actual dates for both periods.

If a selected period extends before the first observed data, its totals cover recorded activity only. Earlier history is unavailable, rather than measured zero. Comparisons are withheld when the comparison period is not fully covered. Critical coverage and unavailable-comparison notices remain visible. Open **About this data** beside the Analytics heading for recorded history, period comparisons and unique-visitor definitions. Close it with the close button or Escape. A loading error is shown separately and never presented as an empty report. When you change filters, the loaded report stays visible with **Updating report… Previous results stay visible.** until the new results arrive. Filters remain available; exports and actions on the previous rows pause during the update. Opening a new search or content breakdown loads its rows without replacing the surrounding report. The optional Most read articles preview can show its own loading or retry state while Overview remains available.

## Content activity

Choose **Content activity** to inspect distinct articles with recorded creation, update and publication actions. Repeated updates to one article count as one updated article for the selected period. The **Recorded actions** table retains each action separately.

**Recorded activity over time** shows daily creations, updates and publications as grouped bars above the detailed rows. Actions sit side by side because one article can contribute to several actions; they are not stacked into a combined total.

Explicit edits, saved versions and accepted article changes count as updates. Autosaved drafts do not. Publishing an edited draft can record both an update and a publication. Automatic translation changes may not appear: this report covers recorded editorial actions, rather than reconstructing every change.

Filter by date, time zone, recorded action or language. Select a contributor to filter by author and open their recorded actions. Selecting a metric opens the matching action list. Contributor and recorded-action tables show 25 rows per page. **Daily recorded activity** shows 10 rows per page, with **Previous** and **Next** controls. Downloads include all matching rows.

Contributors can be people, named API keys or **System**. Deleted keys and unknown users keep their recorded type. Daily and contributor counts are calculated independently: several people can update the same article, so adding their counts can exceed the whole-period distinct total.

A publication across all languages is one recorded action with **Language not recorded (all-language action)**. It appears under **All languages**. A specific-language filter matches only actions that explicitly recorded that language; it does not infer past languages from the article's current translations.

Open **About recorded article activity** for editorial metric details and the earliest recorded action. **First recorded activity** describes editorial history independently of reader tracking. Periods before that point have unavailable history. Comparisons use your chosen period with the same filters, and are withheld when that period begins before recorded history. Deleted articles remain in the action history; their editor links are unavailable.

## Translation gaps

Choose **Translation gaps** for a **Current snapshot** of published source articles needing translations. This report has no date filter or period comparison. Your reporting dates, language, time zone and comparison choice are preserved when you return to another tab.

An article qualifies when its own source-language copy is published. An unpublished edit does not hide an older published copy. Targets are your help center's primary and enabled languages, excluding each article's own source language. Removed languages and deleted articles are excluded.

* **Missing translation** means no saved version exists in the target language. An autosaved draft alone still belongs here.
* **Saved, unpublished** means a saved target-language version exists, but none is published.
* A target language with a published copy has no gap, even if a newer unpublished edit exists. The report does not measure whether translations are up to date.

The **Translation gap breakdown** shows a compact pie chart for missing and unpublished versions among matching gaps, with both counts and the number of source articles affected. These are article-language pairs, not a translation-completion percentage. Beside it, **Gaps by language** ranks target languages by total gaps, with each bar split into missing and saved-unpublished versions. Hover for exact counts, or select a language to filter the matching rows. Both charts use all matching gaps across every table page. Use the language chart’s page controls when more than six languages have gaps.

Each row is one article and target-language pair. The source-article total counts each affected article once; the two status counts count pairs. Filter by target language, status or current topic and its subtopics. Select a status metric to open its matching rows.

Topic filters follow the target language's current article placement and topic tree. An article with no placement in that target language remains visible under **All topics**, but does not match a topic filter. Selecting an article opens that target language in the editor. Viewing the report never translates or publishes anything.

## Download a report

Open **Export report**, then choose **Export CSV** or **Export Excel**. The download follows the report, filters, sort order and drill-down currently open, including an individual article, contributing searches, recorded actions or translation status. It includes **all matching rows**, even when you are viewing a later page.

A download supports up to **50,000 result rows**. If more rows match, narrow the date range or other filters and try again. No partial file is returned. Translation gaps remain a current snapshot, with no date range or comparison.

Excel files contain **Summary** and **Results** sheets, plus **Daily** or **Languages** breakdown sheets where that report provides them. Overview also includes **Traffic sources**, **Articles needing attention** and **Searches needing attention**. The Summary records the selected period, time zone, filters and available collection coverage, or the snapshot time for translation gaps. Whole-period unique and distinct totals are calculated independently of result and breakdown rows; do not add row values to reconstruct those totals.

CSV contains the same sections in one table. Its `record_type` column identifies `summary`, `result`, `daily` and `language` rows. Overview adds `traffic_source`, `attention_article` and `attention_search`. Source rows distinguish `referral`, `direct_or_unknown` and `not_collected` in `source_status`. Columns that do not apply to a section are blank. Customer-entered text that could be interpreted as a spreadsheet formula is prefixed with an apostrophe in CSV; Excel keeps that text as literal text.

Dates and timestamps include their time zone or use unambiguous calendar dates. Visible reading durations are in **milliseconds**. Rate columns end in `_ratio` and contain values from 0 to 1: **0.5 means 50%**. Missing measurements and unavailable comparisons are blank; a measured zero remains numeric zero. Downloads do not fill in unavailable history.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.