Is Your Team Using AI? What Admin Dashboards Actually Show

A review of documentation for seven AI admin consoles — from OpenAI, GitHub, Anthropic, Microsoft, Google and Cursor — catalogued 85 named metrics and found that all seven count seats and active users, but no two define an active user the same way and none measures whether the work was correct. Six of the seven consoles report spend or credits, with Google Workspace's Gemini reports the exception, while GitHub's median minutes to merge is the only cycle-time measure in the census. No console reports defect rate, rework, reverted commits or review burden, and every figure is vendor-computed from vendor telemetry with no third-party audit.

Asked whether the AI spend is working, most executives open the admin dashboard of whichever product they pay for and read out an active-user number. This page is about what that number is, and is not. We read the documentation for seven admin consoles, from OpenAI, GitHub, Anthropic, Microsoft, Google and Cursor, and catalogued 85 named metrics. All seven count seats and active users. No two define an active user the same way. None measures whether the work was correct. This post is for the executive or IT lead choosing which numbers to put in front of a board. It sets out what every console counts, what only some count, and what none count; a table of twelve words that mean different things in different consoles; and the two new measures that reached dashboards in the week before this page's date. It does not argue that usage is the wrong question; our post on completed work as evidence https://www.digitalapplied.com/blog/ai-usage-completed-work-evidence-guide makes that case and we do not repeat it here. Every number in every console is vendor-computed from vendor telemetry; nothing in this census has been audited by a third party. This blog runs on Anthropic's models, and Anthropic's consoles are read by the same rule as the others. 1. 01Seven consoles count active users; six definitions are in play.An explicit prompt, a sent message, token consumption, a deliberate feature interaction, being shown a response, or using any AI feature. Put two vendors' active-user counts on one chart and you are comparing different events. 2. 02Six of seven count spend; four classify the work; one reports a cycle time.Google Workspace's Gemini reports have no cost metric. OpenAI, Microsoft and Cursor classify what was done; Google Workspace goes only as far as which app and feature was used. GitHub's median minutes to merge is the only speed measure in the census. 3. 03No console measures quality.Nothing reports defect rate, rework, reverted commits or review burden. The nearest things are review findings, which the vendors themselves say to check with reviewers, and an estimated 'assisted hours' figure with no published formula. 4. 04The vendors that attribute code also publish where it breaks.Anthropic calls its figures a deliberate underestimate. Cursor says formatters invalidate its diff signatures and that its code attribution has not reached background agents or the CLI. GitHub says developers can switch off the IDE telemetry the breakdowns rely on. 01 — The findingWhat every dashboard counts, and none do Count seats and active users Every console has both. Not one defines 'active' the same way as its neighbour, and one uses a 30-day month while the rest use 28. Classify what the work was OpenAI samples messages into tasks; Microsoft labels chat intents; Cursor classifies agent sessions on-device; Google Workspace reports per feature. GitHub, Anthropic and Google Cloud do not. Measure whether it was right No defect rate, rework, reverts or review burden anywhere in the census. Volume, spend and adoption are what the consoles know. Three more counts sit between those. Six of seven report spend or credits; Google Workspace's Gemini reports are the exception, and their answer to heavy use is a licence upgrade rather than a cost figure. Three of seven measure agent actions rather than messages: GitHub's cloud-agent and third-party-agent fields, Anthropic's Cowork dispatch turns and actions, and Cursor's cloud-agent pull requests and bot automation level; Microsoft counts agent responses and sessions but not what the agent changed. And exactly one console reports a cycle time: GitHub's median minutes to merge, with Copilot-authored and Copilot-reviewed variants. Every other console stops at how much happened, not how fast. One console asks the humans. OpenAI's impact survey is the only built-in instrument that collects a self-reported outcome, and OpenAI's own page says it is directional and not a causal measure of return. Microsoft's dashboard can ingest survey data, but from a separate product and only above a licence threshold. Messages shows the number of messages associated with a category or task. A message count doesn't tell you how much work the team finished.OpenAI, ChatGPT enterprise usage-insights documentation, read September 22, 2026 02 — The censusTable 1: coverage by console One row per console, four questions: does it count seats and active users, does it count spend, does it classify the work, and does it measure what agents did. "Partly" means the console has a weaker form of the measure than the column asks for. The 85 underlying metrics, with their exact field names, windows and export paths, are in the research file this page is built from and summarised in the methodology. | Vendor documentation read September 22, 2026, dated by each vendor's changelog or release notes; see the methodology for the two vendors whose pages carry no date. | | | | |---|---|---|---| | Console | Seats and active users | Spend | Task classification | |---|---|---|---| | OpenAI ChatGPT Admin Console | Yes: seats purchased, enabled, active, pending; power users | Yes: credits, cost in USD, tokens | Yes: a sampled task classifier not exportable | | GitHub Copilot usage metrics | Yes: daily, weekly, 28-day active users; adoption phase | Partly: AI credits used, 'not invoicing totals' | No | | Anthropic Console and Enterprise analytics | Yes: seats assigned, daily to monthly active users, adoption rate | Yes: estimated cost per user per model; tokens | No: counts and breadth, never topic | | Microsoft 365 Copilot usage report and dashboards | Yes: enabled, active, returning, habitual and power users | Yes: Copilot credits for agents | Yes: chat prompt intents ask, catch up, draft | | Google Workspace Gemini reports | Yes: active users, eligible licences, usage bands | No | Partly: usage per app and per feature | | Google Cloud Gemini Code Assist metrics | Yes: hourly to 28-day active users shown a response | No: tokens and calls only | No | | Cursor team analytics | Yes: daily active users by surface | Yes: spend in cents per user; per-user limits | Yes: conversation insights by category, work type, complexity | | Agent measures, same sources and date. "Agent" means a background or cloud agent acting on code or files, not a chat assistant. | | |---|---| | Console | Agent actions and outputs | |---|---| | OpenAI ChatGPT Admin Console | Partly: Codex share of merged commits; code-review findings | | GitHub Copilot usage metrics | Yes: cloud agent PRs merged, third-party agent sessions, CLI customisations | | Anthropic Console and Enterprise analytics | Yes: Cowork dispatch turns and actions; Claude Code commits and PRs | | Microsoft 365 Copilot usage report and dashboards | Partly: agent responses, sessions, users; not what agents changed | | Google Workspace Gemini reports | No | | Google Cloud Gemini Code Assist metrics | No: IDE interactions only | | Cursor team analytics | Yes: cloud agent PRs opened and merged; bot automation level | 03 — The trapTable 2: same word, different thing This is the table the post exists for. Every row is a word that appears on more than one dashboard and means something different on each. The definitions are the vendors' own, condensed; the last column is why a board deck built from two consoles will not reconcile. | Definitions condensed from each vendor's documentation, read September 22, 2026. | | | | |---|---|---|---| | Word | Some vendors mean | Others mean | Why it matters | |---|---|---|---| | Active user | GitHub: sent an explicit prompt. OpenAI: sent at least one message. Anthropic: consumed tokens. | Microsoft: deliberately used a feature; passive AI does not count. Google Cloud: was shown a response. Cursor: used any AI feature. | Google Cloud's count can exceed everyone else's for identical behaviour; Microsoft's is the strictest. | | Monthly | GitHub: trailing 28 days. Microsoft and Google: 28-day periods. | Anthropic: trailing 30 days. | A 28-versus-30-day window shifts the denominator by about 7% before anything changes. | | Active versus engaged inside GitHub | Active: used Copilot that day. | Engaged: used a feature on at least two distinct days in 28. | Since September 17 GitHub has two bars in one product; reporting one as the other misstates the hurdle. | | Adoption phase inside GitHub | Dashboard: every user active across the trailing 28 days. | API: users active on that record's day. | GitHub's own docs say the dashboard and API cohort counts do not match. | | Message | OpenAI: ChatGPT messages plus Codex turns, combined. | Anthropic: chat messages only; Claude Code sessions counted separately. | One number spans a chat product and a coding agent; the other does not. | | Chat | OpenAI: ChatGPT chats plus Codex threads started that day. | Cursor: one surface among composer, agent and command requests. | A fused counter versus one of five. | | Session | Anthropic: a Claude Code session, which essentially never spans a UTC day. | Microsoft: continuous interaction with an agent; a new session after 30 minutes idle. | An idle timeout manufactures sessions from one long piece of work. | | Acceptance rate | GitHub: acceptances over generation events. Google Cloud: accepted over suggested. | Anthropic: accepted edit-tool proposals. Cursor: two rates, applies and tab completions. | Four denominators. Cursor publishes two because its surfaces differ. | | Lines of code | GitHub: suggested versus added in the editor. | Anthropic and Cursor: accepted in-session versus survived into a merged commit. | Three tiers exist; only Anthropic and Cursor report the third. | | Credits | OpenAI: credit-priced usage, with USD cost reported separately. | GitHub: consumption analysis, not invoicing. Microsoft: agent credits in Microsoft 365. | Three unrelated currencies, none of which is an invoice. | | Power user | OpenAI: top 20% of senders, 75+ messages a week, 3+ tools a month. | Microsoft: 15+ actions a week in 9 of 12 weeks. Google: top 10% with 20+ uses in 28 days. | Three thresholds, three windows, three units. | | Task or use case | OpenAI: a classifier over a sample of messages; not exportable. | Cursor: on-device classification of agent sessions; exportable. Microsoft: chat prompt intents only. | 'What is the team using AI for?' has three incompatible answers. | Anthropic's Claude Code documentation says its contribution metrics "are deliberately conservative and represent an underestimate", counting only lines and pull requests with high confidence of Claude Code's involvement. Cursor's says diff signatures "may be invalidated if automated code formatting is modifying lines" and that code tracking has not reached its background agents or CLI. GitHub's notes that developers can switch off the IDE telemetry the breakdown arrays rely on, so server-side-only users inflate active-user totals above the breakdowns. Read these before the percentage. 04 — This weekThe two measures that arrived this week Two consoles changed in the days before this page's date, and both changes replace a volume count with a judgement. On September 16, OpenAI published a walkthrough of its Admin Console analytics https://openai.com/index/how-to-connect-ai-usage-to-business-value/ : a Usage view of active users, credits and tokens across ChatGPT Work and Codex; an Insights view whose task classifier groups a sample of messages into use cases and tasks, with the share of credits by model, reasoning and speed setting, a plugin leaderboard and a skills view; and an Outcomes view with the Codex share of merged commits and lines of code. The post's worked example of return on investment is, in its own words, hypothetical, and we do not print its figures. Two limits matter: the classifier works on a sample, and OpenAI's help centre says the task insights cannot be exported. On September 17, GitHub added feature engagement https://github.blog/changelog/2026-09-17-copilot-impact-dashboard-now-shows-feature-engagement to the Copilot impact dashboard and its usage-metrics API. A user counts as engaged with a feature only when they used it on at least two distinct days in the 28-day window, across code completion, agent edits, passive and active code review, the cloud agent, the CLI and the app. GitHub's documentation adds two cautions: the per-feature counts overlap and must not be summed, and a user who chats with Copilot every day but never produces a completion or an agent edit is not in the adoption cohort at all. The same day GitHub put agentic CLI customisations into the API, with top-five counts for skills, custom agents, MCP servers, slash commands and plugins, and the customer-defined names hidden. How to read them: OpenAI's classifier answers what the work was; GitHub's engagement bar answers who actually adopted a feature rather than who touched it once. Both are more useful than a message count and both are fenced by their own vendors. Neither says whether the work was any good, which is the question our post on Anthropic's oversight metrics https://www.digitalapplied.com/blog/anthropic-agent-oversight-metrics-coverage-latency-escalation approaches from the supervision side. 05 — The deckWhat to report to the board Every item below exists as a named metric in at least one console, so it can be sourced rather than estimated. The rule that makes the list honest is the last one. 1. Seats assigned against seats active in the last 28 days, with one window across vendors and the window stated. 2. Adoption rate with its denominator printed. Anthropic ships it as a field; everyone else needs the division done by hand. 3. Returning users. Microsoft’s “returning users”, active in the prior and the current 28-day period, is the only named retention metric in the census; rebuild the equivalent elsewhere from per-user daily rows. 4. Depth, not only breadth: GitHub’s two-days-in-28 engagement bar or Microsoft’s habitual and power bands, with the bar named. 5. Spend per active user, available from Cursor, Anthropic, OpenAI and, with its caveat, GitHub. Not available from Google Workspace. 6. AI-attributed share of merged code with the attribution method beside it. Never the percentage alone. The harness matters as much as the model, as our post on harness cost https://www.digitalapplied.com/blog/coding-agent-harness-cost-harnesstax-same-model-5x showed, and our census of what companies disclose about their own AI share https://www.digitalapplied.com/blog/ai-share-of-own-work-tech-companies-disclosed-figures shows how far definitions drift. 7. One cycle-time number. GitHub ships median minutes to merge. If you are not on GitHub, this row comes from your own version control, not the AI vendor. 8. One thing the dashboard cannot tell you, named explicitly: whether the work was correct. No console in this census measures it, and a board that hears that sentence once will ask the right question next quarter. If you want the deck built, our analytics service https://www.digitalapplied.com/services/analytics starts from these eight rows. 06 — How we built itMethodology A census of what seven admin consoles document about themselves. Not a test of any console, not a ranking, and not a recommendation of one vendor over another. - Population - Seven consoles: OpenAI's ChatGPT Admin Console and Workspace analytics; GitHub's Copilot usage metrics and impact dashboard; Anthropic's Console analytics and Claude Enterprise analytics; Microsoft's Microsoft 365 Copilot usage report, Copilot Dashboard and Agent Dashboard; Google's Workspace Gemini reports and Cloud Gemini Code Assist metrics; Cursor's team analytics and admin API. Google has at least two consoles in different products with no shared metric, so it appears twice. - Inclusion rule - A metric needs a name and a definition in the vendor's own documentation. Eighty-five such metrics were catalogued with their window, scope and export path; this page prints the coverage summary and the twelve-word comparison, not the full list. Cursor is included as an AI developer-tool vendor whose documentation the blog's link policy allows. - Dating - GitHub fields are dated by changelog entry general availability February 27, 2026; agent-app fields August 7; engagement and CLI fields September 17 . Anthropic pages are bounded by the platform release notes, whose newest entry was September 18. Microsoft pages carry document dates from January 2025 for the Graph usage-report reference to September 8, 2026 for the Viva metric reference. Google Cloud's metrics page was last updated September 18, and the Workspace report set dates from a February 17 update. OpenAI console features are dated by its enterprise release notes March 12, July 16, July 23, August 13 and September 11 and the September 16 post. Two sources carry no date at all: OpenAI's Admin API reference and every Cursor documentation page; their rows state the design as read and claim no date. - What was excluded - Seat prices no pricing page was read . OpenAI's return-on-investment example, which its page calls hypothetical. Admin analytics for Google's Gemini Enterprise agent platform not located . A customer testimonial in OpenAI's post. Anthropic's contribution metrics under zero data retention and its Claude Code activity on Amazon Bedrock, which the documentation says are not reported. - Corrections to common versions - GitHub's "monthly" is 28 days and Anthropic's is 30. GitHub's dashboard and API adoption cohorts do not match, by GitHub's own statement. Anthropic's guide and its summaries endpoint define an active user differently; this page uses the field-level definition, token consumption. Microsoft's "Copilot assisted hours" has no published formula. - As-of date - This page belongs to the September 19, 2026 batch; its sources were collected on September 22, 2026. Every dated source is on or before September 19, 2026; undated pages are stated as designs, not dated facts. - Known limitations - Every metric is vendor-computed from vendor telemetry with no third-party verification. This blog runs on Anthropic's models; Anthropic's consoles are catalogued by the same rule as the other five vendors'. - Refresh - Extended when a vendor adds a metric that changes a row in the coverage table or the twelve-word table, or when any console ships a quality measure. 07 — Next stepThe dashboard tells you who used it, not what it did for you Print the definition next to every number before it reaches a board Take the eight-row list above, pull each figure from the console that has it, and write the vendor's definition and window in the same cell. Where two vendors' numbers sit side by side, check them against the twelve-word table first. Then add the sentence no console can supply, that nothing here measures whether the work was right, and ask what evidence would.