Google's core updates do not announce what changed. The industry reverse-engineers the impact by watching which sites rose and which fell. The right unit of analysis is a stable cohort measured before and after each rollout, with the underlying signals tagged at the profile level. That is what this dataset gives us.
How the Cohort Was Measured Across Updates
The same 1,184 client profiles tracked weekly on a 5x5 grid for 14 months (January 2025 to March 2026). For each core update we measured the per-profile rank delta from the week before rollout start to the week after rollout completion. We then segmented profiles into winners (lifted 2+ positions), losers (dropped 2+ positions), and stable (within 1 position). For each segment we ran a profile-level audit of seven review signals to see which ones predicted membership in each group.
1,184 profiles. 14 months. 4 core updates. 7 review signals tagged per profile. Winner / loser / stable segmentation per update.
March 2025 Core Update
The first update in our cohort window. Volatility was moderate. The signal that predicted winner status most strongly was response rate. Profiles with high owner-response rates were 2.4x more likely to be winners than losers.
- Cohort winners: 31% of profiles
- Cohort losers: 22%
- Stable: 47%
- Strongest predictor of winner status: owner response rate >= 80% within 72 hours
- Strongest predictor of loser status: review recency > 90 days at update start
August 2025 Core Update
The most volatile update in the window. Winners and losers spiked at the expense of stable profiles. The strongest predictive signal shifted from response rate to review velocity. Profiles in the top quartile of trailing-90-day velocity were 3.1x more likely to be winners.
- Cohort winners: 38%
- Cohort losers: 29%
- Stable: 33%
- Strongest predictor of winner status: trailing-90-day velocity in top quartile
- Strongest predictor of loser status: more than 5% of recent reviews flagged as inauthentic by our internal scoring
November 2025 Core Update
Lower volatility. The update read as a refinement of August's logic rather than a new direction. The new dimension that emerged: review-text quality. Profiles with reviews containing specific service detail (location, time, named staff) outperformed profiles with shorter generic reviews of the same star average.
- Cohort winners: 26%
- Cohort losers: 19%
- Stable: 55%
- Strongest predictor of winner status: median review length > 28 words and at least 30% containing specific service detail
- Strongest predictor of loser status: median review length < 12 words across the trailing 90 days
Three review signals predicted winner status across all four updates: response rate, velocity, and recency. The next update will continue to favour the harder path.
March 2026 Core Update
The most recent update in the window. The defining shift was AI-content detection. Profiles whose reviews showed templated phrasing patterns (likely AI-generated) lost ground rapidly. The signal that mattered most for winners was a low share of inauthentic-flagged content combined with a steady velocity.
- Cohort winners: 29%
- Cohort losers: 33%
- Stable: 38%
- Strongest predictor of winner status: combined low inauthentic share (<3%) and trailing-90-day velocity in top half
- Strongest predictor of loser status: inauthentic-flagged share > 10% of trailing 90 days
Signals That Worked Across All Four Updates
Three signals predicted winner status across every update in the window. They are the closest thing the dataset has to update-proof variables.
- Owner response rate of 80%+ within 72 hours: positive predictor in 4 of 4 updates
- Trailing-90-day review velocity in the top half of the cohort: positive predictor in 4 of 4 updates
- Recency: most recent review within the last 30 days: positive predictor in 4 of 4 updates
Signals That Stopped Working
Two signals lost their predictive power across the window. Both are signals the SEO industry still references as if they were stable variables.
- Pure review-volume comparisons (without velocity): predictive in March 2025, neutral by November 2025
- Star rating in isolation (without volume context): predictive in March 2025, weakened in August, near-zero in March 2026
What Four Updates Tell Us About the Direction
Google's local algorithm has moved consistently toward signals that are harder to game. Velocity, recency, response rate, and review text quality require sustained operational effort. Volume and stars are easier to inflate and have correspondingly lost predictive weight.
The directional bet for the next 12 months is straightforward. Build a profile that looks operationally maintained: weekly new reviews, owner responses inside 72 hours, customer-attached photos, and reviews that contain real service detail. Avoid every shortcut that produces volume without quality. The next update will continue to favour the harder path.
