According to Discover Monitor data, on September 16 there were sharp changes in the news and publishers appearing in Discover feeds. This could indicate an algorithm update, although there is no official confirmation from Google yet.
In the data comparison between the second half of August and the first half of September, visibility weighted by position became more concentrated and the LATAM share rose. But with new data, starting on September 16, 2026, we ran a short comparison between the same weekdays of the previous week, September 10–11, and September 17–18. In that later window, part of the concentration observed in the first half of the month reverses: top 1 −1.825 pp, top 5 −6.020 pp, top 10 −6.015 pp, and HHI from 0.04077 to 0.03172.
- The takeaway from the first half of the month: Compared with August, the top 5 went from 29.241% to 34.726% (+5.485 pp), the HHI from 0.02954 to 0.03816, and LATAM from 9.505% to 11.073% (+1.568 pp).
- The cut after September 16 changes the distribution. Between September 10–11 and 17–18, LATAM drops from 11.073% to 6.179% (−3.281 pp), which not only reverses the increased weight of LATAM publishers, but falls to far lower visibility levels.
- Major publishers disappear from Google Discover several publishers with significant visibility that had been growing strongly in recent months disappear completely from Google Discover starting on September 16 (we will not name them for obvious reasons, since as of today, September 21, they have not recovered any visibility in Discover, zero appearances). Although we have direct confirmation from some of these publishers, we cannot confirm that this is a ban without access to more data or official confirmation from Google. It could be due to other reasons such as technical issues, a temporary Google glitch (indexing problems were reported at several publishers in the preceding days), or a tightening of the quality guidelines for Google Discover.
- Major changes in the relative visibility of TOP publishers. For example, www.abc.es loses −2.572 pp, www.cronista.com falls −3.471 pp, and www.elconfidencial.com −1.925 pp, while www.elmundo.es rises +1.200 pp.
- The signal is descriptive and provisional. only two days have passed as of the date of this analysis, and the bootstrap is descriptive, not a significance test.
1. Cumulative reading: August versus September
Increased concentration among high-authority publishers.
- The top 1 went from 8.099% to 11.661% (+3.562 pp)
- The top 5 from 29.241% to 34.726% (+5.485 pp)
- The top 10 from 45.666% to 50.026% (+4.360 pp)
- The HHI from 0.02954 to 0.03816.
- LATAM also increased from 9.505% to 11.073% (+1.568 pp).
These results describe the observed visibility data; they do not measure traffic, clicks, or preferences.
2. September 18 update: 10–11 versus 17–18
The new comparison examines September 10–11 (the Thursday and Friday of the previous week) against September 17–18. In the later pattern, the top 1 share goes from 12.003% to 10.177% (−1.825 pp), the top 5 from 36.473% to 30.453% (−6.020 pp), and the top 10 from 51.910% to 45.895% (−6.015 pp). The HHI falls from 0.04077 to 0.03172.
This is a partial reversal of the concentration pattern seen in the first half of the month, not a contradiction of the monthly comparison yet, because the sample is still small.
3. LATAM and positions in the later cut
The LATAM share drops from 9.460% (august) to 6.179% (−3.281 pp) (september last days). The same direction appears across all three sensitivity checks: deduplicated exact URL with best position, comparison of the full days September 10 versus 17, and a common time window up to 16:48.
4. Sites, rotation, and consistency
The main movements include www.cronista.com (−3.471 pp), www.abc.es (−2.572 pp), and www.elconfidencial.com (−1.925 pp). On the other hand, www.elmundo.es advances +1.200 pp. We have 137 new sites in the second window and 75 missing ones. Of 37 movements of at least 0.25 pp, 36 keep their direction across the concentration sensitivity checks. The Argentine outlet www.lanacion.com.ar is the exception: its −0.332 pp decline does not hold a stable direction across all controls.
5. Method, scope, and what it does not allow us to conclude
Each appearance receives a DCG weight: 1/log2(position+1). Shares are normalized within each daily block before averaging over the period. An entity absent on a valid day counts as zero.
The data is consistent with a distribution change after September 16, but it does not prove an update or an algorithmic origin. Editorial output, the news agenda, and the composition of available URLs may also play a role.
