1. ed.cafe
  2. AI safety guides
  3. Workforce Literacy
Workforce Literacy Includes: Pre-Registered Dual-Denominator Finish Ledger For Boston learners, counselors & educators

The Finish Ledger: Per-Cohort Completion of the Google AI Professional Certificate, Counted Two Ways

The ledger has zero rows today because no ed.cafe cohort has run. This is its method, published first, so every future completion figure can be checked.

Margot Reyes The evidence synthesizer · ed.cafe editorial persona · Researched & verified by ed.cafe editors
Published · 6 min read (1,345 words)
ed.cafe guide banner: The Finish Ledger: Counted Two Honest Ways by Margot Reyes
Status as of October 2026: US federal & state rules (Massachusetts highlighted where distinct). Primary sources & sample sizes linked inline.

Status on 15 October 2026: zero rows. No ed.cafe cohort has run, so the Finish Ledger reports nothing yet, and it will not show estimates or pilot numbers in the meantime. The public record is just as thin. Coursera shows 1,632,553 learners enrolled in the Google AI Professional Certificate and no completion rate. Massachusetts has published sign-ups, not completions. This page publishes the counting method first, before any results exist: two denominators, two kinds of completion never merged, cost per certificate, and a column for misses.

The headline number is an enrollment count

The largest public number for this certificate is 1,632,553. That is the enrollment count on the Coursera program page on 15 October 2026. It is cumulative and global. It has no numerator.

Massachusetts made the certificate free for residents until 31 December 2027, according to the Massachusetts AI Hub. The only public count of state participation is MassLive's analysis of more than 38,300 sign-ups through late June 2026, as summarized by WindowsForum. That summary notes the analysis does not establish completion.

Enrollment tells you demand. It does not tell you whether a support model works.

Why the numbers in circulation do not answer the question

Two kinds of figure fill the gap, and neither is a completion rate for this certificate.

Graduate outcome surveys. Google's May 2025 impact report reports "over 70% of our U.S. graduates" with a positive career outcome, from a 2024 graduate survey. The denominator is graduates who responded. Non-finishers are outside it by design.

Older cohort figures for other certificates. The nearest supported-cohort figure is for the Google IT Support Professional Certificate. Jobs for the Future reported in 2021 that 1,325 of 4,497 enrolled community-college students completed all five courses. That is 29.5%. Different certificate, different learners, different era.

So a funder today cannot answer a basic question. Of the adults who start the Google AI Professional Certificate inside a supported group, what share finish?

The dimension that matters: where the starting line is drawn

The same JFF cohort produces two honest rates. Against all 4,497 enrolled: 29.5%. Against the 2,902 who completed at least one course: 45.7%. A 16.2-point gap, from one choice of denominator.

The Finish Ledger removes that choice by printing both, every time:

  • Day-1 enrollees: every learner on the cohort roster on the first day, whether or not they attend.
  • Session-1 starters: every learner who attended the first session.

A provider that reports only the second number can make a weak cohort look strong. Printing the first beside it closes that door.

The two completions, never merged

Completion A, the certificate. The Google AI Professional Certificate shows as earned in Coursera by the report date. This is the figure funders ask for.

Completion B, the work sample. A one-page task from the learner's field, scored on ed.cafe's published 12-point rubric: four criteria scored 0 to 3, with a pass at 8 of 12 and no zeros. The rubric explainer sets out the scoring.

A learner can earn A without B, or B without A. Adding them together would hide exactly what a funder or employer needs to see.

The ledger, as it stands

Copy the template. Every row will be filled from aggregate cohort records only. No learner-level data leaves the room.

Pre-Registered Dual-Denominator Finish Ledger
COHORT  HOST TYPE  WINDOW  DAY-1  SESSION-1  CERTS  A/DAY-1  A/S-1  SAMPLES  B PASS  COST/CERT  MISSES
#1      -          -       -      -          -      -        -      -        -       -          -

Status, 15 Oct 2026: Circle #1 forming. No cohort has run. No figures exist.

Column definitions:

  • CERTS: certificates earned by the report date (Completion A).
  • A/DAY-1 and A/S-1: CERTS divided by each denominator, with the counts shown, never the percentage alone.
  • SAMPLES and B PASS: work samples submitted and the number scoring 8 or more with no zeros.
  • COST/CERT: the partner fee divided by CERTS.
  • MISSES: where non-finishers stopped (by course), and what will change next cohort.

Cost per certificate, down to the dollar

Cost per certificate is the partner's flat fee divided by verified certificates. It is the number a budget officer can compare with any other use of the money.

Illustrative example: a program pays $10,000 for a cohort and 20 learners earn the certificate. Cost per certificate is $500. If 10 earn it, the cost is $1,000. The fee did not change. The completion count did.

The current ed.cafe Finish Line fee is $7,500 for up to 30 learners, as listed on the Staff Room page. The ledger will divide that real fee by real certificates. Nothing more.

Competing explanations the ledger must expose

A completion rate is only as honest as the conditions printed next to it. Four confounders matter most for this certificate:

  • Selection. A host site that brings its most motivated learners will post higher rates. ed.cafe's house rules bar sponsors from picking learners; host type is printed so readers can compare like with like.
  • Prior progress. Many learners arrive having started alone. The ledger will record how many had any course already complete on day one, because a cohort of half-finished learners is a different base rate.
  • Report timing. Some learners finish after the last session. The ledger reports Completion A at the report date and again at a 30-day follow-up, labeled separately.
  • Small numbers. A cohort of 30 swings by 3.3 points per person. One cohort is an anecdote. Comparisons wait until at least three cohorts exist.

Adding a partner's own counts

The ledger is built so that other programs running this certificate can sit beside ed.cafe's rows. A career center, library or college that runs its own group can send the same aggregate columns: day-1 count, session-1 count, certificates by report date and, if it scores work samples, passes. Rows are labeled by host type, never by organization, unless the partner asks to be named in a signed agreement, which matches the site's house rule on naming partners.

The point is comparison on equal terms. A row with only one denominator cannot be added. A program that sends a weak cohort with both denominators is exactly what the ledger needs, because a benchmark built only from successes would repeat the survivor math it was designed to remove.

Method, limits and what would falsify it

The method has not yet been independently reviewed. Until a university or evaluation-firm reviewer signs off, treat it as ed.cafe's proposal, open to challenge. Any change to a definition will be logged with the date, and earlier rows will not be silently recalculated.

What would make the ledger misleading:

  • A day-1 roster defined after early dropouts (the denominator moves late).
  • Certificates counted from learner self-report instead of Coursera.
  • Missed cohorts left off the page.

What would undercut ed.cafe's own thesis: supported cohorts that finish at rates no better than the 29.5% all-enrolled benchmark from the JFF network. If that happens, the ledger will say so in the misses column. The site's house rules already commit to publishing misses.

Questions to put to any completion claim

Use these on ed.cafe's future numbers as much as anyone else's:

  1. What is the count on the bottom of the fraction, and when was that roster fixed?
  2. Is completion verified in the platform, or self-reported?
  3. Are certificates and any skill measure reported separately?
  4. Is there a cost per completer, using the actual fee?
  5. Where are the cohorts that missed?

For why these questions matter across every figure you will see, read why certificate completion rates run from 3% to 85%, and if you run a group of learners, the Staff Room page shows how to plan a cohort that reports this way.

Sources

Free for Massachusetts learners

Finish Google's AI certificate with 30 neighbors in Boston.

Circle #1 is forming now (closes December 1, 2026; zero cohorts have run yet, and every completion count will be published two ways). Educators can run a Finish Line cohort in the Staff Room; L&D leaders can inspect license waste in the Council.