On September 21, we published a preliminary mid-month analysis due to the possible Discover algorithm update and penalties affecting several news outlets.
Once September had ended, we conducted a new analysis covering the full month. September yielded two findings that should be kept separate. In the monthly results, okdiario.com moved into first place, and the group of the five sites with the highest daily visibility gained ground compared with August. Within the month itself, however, the sample recorded a broad redistribution of relative visibility starting on the 17th, following a day—the 16th—that we identify as the possible date of the changes.
The analysis compares August and September 2026 using a methodology based on observed appearances and position. Thus, the monthly summary describes the full month available, while the before-and-after comparison around the 16th seeks to identify when some movements were concentrated; it does not replace the monthly result.
Monthly results: okdiario.com leads as the top five become more concentrated
Here are the largest decreases and increases in visibility share in September 2026:
In the comparison of daily averages, okdiario.com rose from 5.059% to 8.171% in weighted visibility share, an increase of 3.112 percentage points (pp). It climbed from second to first place. ABC also made clear gains, rising from 3.796% to 5.375% (+1.579 pp) and reaching third place. Threads went from 1.860% to 2.756% (+0.896 pp), moving up from 13th to 8th place.
The gains were not universal: X.com lost 2.027 pp, falling from 3.279% to 1.252%, and dropped from eighth to twentieth place. El Español fell from 7.370% to 5.989% (−1.381 pp), although it remained in second place. Mundo Deportivo declined from 2.325% to 1.456% (−0.869 pp). Cronista recorded a monthly decrease of 0.818 pp, from 3.102% to 2.284%.
| Site | August | September | Change |
|---|---|---|---|
| okdiario.com | 5.059% | 8.171% | +3.112 pp |
| ABC | 3.796% | 5.375% | +1.579 pp |
| Threads | 1.860% | 2.756% | +0.896 pp |
| X.com | 3.279% | 1.252% | −2.027 pp |
| El Español | 7.370% | 5.989% | −1.381 pp |
| Mundo Deportivo | 2.325% | 1.456% | −0.869 pp |
The average daily share of the five most visible sites in each block rose from 29.323% to 31.538% (+2.215 pp). This is a daily concentration measure: each day has its own top five, and the valid dates are then averaged. The list of sites in the top 20 changed little (18 of the 20 remained), although 27 sites moved by at least 0.25 pp in absolute terms.
As a monthly sensitivity check, the 27 material changes retained their direction after deduplicating repeated URLs. In addition, 169 sites that appeared in August did not appear in September, although none had been in August's top 20 or reached a 0.25% share. This monthly zero is not equivalent to Cronista's near-continuous absence during the period after the 16th described below.
What happened after the possible September 16 update?
To review the changes within the month, the valid days from the 1st–15th were compared with those from the 17th–30th, and the 16th is treated separately as a transition. In that comparison, Cronista went from 4.554% before the cutoff to 0.022% after it (−4.532 pp). It had no appearances in the sample from the 17th through the 29th and reappeared with five records on the 30th. This describes an almost complete absence during that observed period, not a confirmed exclusion from Discover.
El Español (7.535% to 4.494%, −3.041 pp) and okdiario.com (9.630% to 6.683%, −2.947 pp) also declined. Clarín went from 1.686% to 0.625% (−1.061 pp). In the opposite direction, Threads rose from 1.618% to 3.746% (+2.128 pp), and YouTube from 2.483% to 4.288% (+1.805 pp). ABC, Infobae, and La Nación ended the post-cutoff period with shares above their earlier averages: +0.597 pp, +0.447 pp, and +0.291 pp, respectively.
| Site | Valid days 1–15 | Days 17–30 | Change |
|---|---|---|---|
| Cronista | 4.554% | 0.022% | −4.532 pp |
| El Español | 7.535% | 4.494% | −3.041 pp |
| okdiario.com | 9.630% | 6.683% | −2.947 pp |
| Clarín | 1.686% | 0.625% | −1.061 pp |
| Threads | 1.618% | 3.746% | +2.128 pp |
| YouTube | 2.483% | 4.288% | +1.805 pp |
| ABC | 5.064% | 5.661% | +0.597 pp |
| Infobae | 1.867% | 2.314% | +0.447 pp |
| La Nación | 1.577% | 1.869% | +0.291 pp |
The largest change between two consecutive days occurred from the 16th to the 17th: the total distance between daily shares was 29.14%. This metric, called total variation (TV), is equal to half the sum of the absolute differences in share across all sites. It does not measure traffic loss and is not a test of statistical significance. In addition, there were other large jumps: from the 29th to the 30th (26.73%) and from the 13th to the 14th (25.53%). Therefore, the 16th–17th stands out in the record, but it was not the month's only pronounced change.
The post-cutoff averages also do not mean there was an identical decline every day. The share of the five leading sites fell from 34.47% before the cutoff to 28.67% afterward, and the aggregated LATAM share went from 11.12% to 6.85%. However, both averages combine days with different values: LATAM's share, for example, rebounded to 11.88% on the 30th. X.com recovered after a one-off decline between the 16th and 17th. Tiempo.com did not appear on the 17th or 18th, but appeared again on the 19th and had later records; there is no evidence of a sustained disappearance in September.
What the signal does and does not allow us to conclude
There are indications that some of the decline in concentration and LATAM share persisted beyond the first few days after the cutoff. In the first and second weeks afterward, the top-five average share was 29.73% and 27.60%, respectively, while the LATAM share was 6.78% and 6.93%. A descriptive comparison matching days by weekday composition also retains the difference. This supports the view that the movement was not limited to a single day, but it does not identify its cause. Furthermore, the deterioration of okdiario.com and El Español was already underway before the 16th in some daily indicators: not all of the change began on that date.
The hourly capture cadence also changed between the 16th and 18th. Although the calculation normalizes appearances within each block and gives valid days equal weight when averaging them, these differences call for caution when interpreting the results.
Share is relative: if some sites gain weight within the sample, others may lose share even if their absolute volume has not fallen. The analysis does not include Google Search Console clicks or impressions, so it cannot establish that an outlet lost that traffic, received a penalty, or was removed from the platform. A site's absence from this capture also does not prove that it was absent from Discover overall. The differences could reflect algorithm changes, but they could also reflect the content mix, personalization, or characteristics of the capture process itself.
Changes in LATAM and sections
In the monthly results, the aggregate share classified as LATAM fell from 9.446% to 9.078% (−0.368 pp). Within that classification, Cronista declined by 0.818 pp and Clarín by 0.524 pp, while Infobae increased by 0.486 pp and La Nación by 0.327 pp.
The section labels recorded by the source also changed: “españa” (Spain) rose by 2.827 pp, “curiosidades” (Curiosities) by 2.356 pp, and “deportes” (Sports) fell by 0.628 pp.
Was there a Google update on September 16?
No announcement of a Discover or core systems update beginning on September 16 was found in the official channels reviewed through October 5, 2026. Google announced a spam update on September 24, which could explain some of the reversals seen on September 30. Given the dates, that announcement does not explain the first change observed between the 16th and 17th. The absence of a located announcement does not prove that there were no internal or unannounced changes, nor does it rule out other explanations. The Discover Monitor data sample detects sharp visibility movements around the cutoff, but there is no official confirmation.
How to interpret the figures
The calculation assigns each appearance a position weight of 1/log2(position+1). Within each daily block, these weights are normalized to obtain shares; the blocks are then averaged across the day and the valid dates, giving each day equal weight. This means a day with more extraction rows does not dominate the average simply because it has more records. A total of 43,820 news items in August and 42,629 in September were used to conduct the analysis. A site with no appearances in a captured and valid block contributes zero for that block.
Share changes are expressed in percentage points: going from 5% to 8% means +3 pp, not +3%. The methodology was checked against a variant that removes exact duplicate URLs within each date and block, retaining the best position. Separately, the 25 material changes in the within-month diagnosis also retained their direction after deduplication.
September combined renewed monthly leadership with a marked internal redistribution around the second half of the month. The signal is clear enough to warrant monitoring.
