Skip to content
Back to research

Nozzle Research

Google Plays Dumb

By Derek PerkinsOctober 20, 2026Data through October 21, 20266 min read

Google has been playing dumb with rank trackers: it serves a normal page, the full query in its title, with results about one word of it. For “easy strawberry jam”, its AI Overview still gives a jam recipe, over results about the TV series Easy. It came and went from March 2026, and from September 16 it reached up to two fifths of SERPs a day, on page 1 or page 2.

This is a companion to Poisoned Google SERPs, which counts a different fault on the same SERPs: the query answered, with the keyword’s usual sites swapped out. The words on the page tell the two apart.

SERPs analyzed
248M
ignored part of the query on page 1 or 2
13.9%
ignored part of the query at the peak, September 28
39.9%
of these pages say “Showing results for”
0.0%

Key findings

  1. One word, not the querySince September 16, page 1 or page 2 ignored part of the query on 13.9% of the SERPs judged on both pages, against 0.06% before, and on 39.9% at the peak.
  2. Page 2 went firstFrom September 17 to 29, a fifth to two fifths of page 2s a day ignored part of the query, nearly all under a normal page 1.
  3. The page plays alongThe title still shows the full query on 99.8% of these pages, and “People also search for” follows the one word.
  4. It came and went from MarchBursts of up to 0.3% of SERPs a week from March; in late September, 25.0%, seventy times July’s peak.
  5. A one-day crash in your rankingsA tracked site misses the top 10 on 70% of these pages; 86% of the time the next SERP is normal again.

The query sent, as Google answered it

Six of the study’s keywords, as Google answered them. The struck words are gone from the results; each opens its page below.

What Google answers instead

Each pair is the same keyword, device and target location, one to three days apart. Pick one, then drag the divider across the page. Google’s own panels, the AI Overview, the Ether price chart and the map, still answer the query sent. The organic results answer one word.

Jump to
Google results page for “easy strawberry jam”, normal SERP on Oct 1, Desktop, United States
Google results page for “easy strawberry jam”, words ignored SERP on Sep 30, Desktop, United States
Normal · Oct 1Words ignored · Sep 30

Drag the divider, or click anywhere on the page to move it there. Scroll inside the frame to move down both SERPs together.

Which word survives

Most of these pages dropped one word of the query, and on half of them one word is all the results still carry. It can be any word: the first, as in “easy strawberry jam”, one in between, as in “valentine gifts for kids”, or the last, as in “eth price”.

How many words of the query Google drops

  • One word66.2%
  • Two words21.6%
  • Three or more12.3%
  • Of 497,362 pages ignoring part of the query since Sep 16.

How many words are left on most of the results

51.5%one word left
  • One word left51.5%
  • Two or more left25.1%
  • None on most results23.4%
  • Of 497,362 pages ignoring part of the query since Sep 16.

Download the data (6 rows, one per chart and part; gzipped CSV). It holds every number in the two charts, with no account or keyword in it.

Chart table
ChartPartPagesOfShare
Words droppedOne word329,198497,36266.2%
Words droppedTwo words107,204497,36221.6%
Words droppedThree or more60,960497,36212.3%
Words leftOne word left256,196497,36251.5%
Words leftTwo or more left124,706497,36225.1%
Words leftNone on most results116,460497,36223.4%
What each column means
chart
dropped (how many words of the query a page dropped) or left (how many words most of its results still carry)
item
The part of the chart: one, two or three_or_more words dropped; one, two_or_more or none left
pages
Page-1 SERPs ignoring part of the query since September 16 that the part counts
total
Every page-1 SERP ignoring part of the query since September 16: the pages both charts count

When it hit

SerpApi’s first report of the fault came on May 4, 2026. In Nozzle’s data it starts in March, in short bursts.

Pages with one word of the query left, week by week

  • Page 1 or 2
  • Page 1
  • Page 2
0%0.5%1%1.5%Mar 2026AprMayJunJulAugSepOctMay 4SerpApi’s first reportPage 1 or 2 off the top, to 25.0%Page 2 off the top, to 24.6%

Each week’s share of judged page 1s and page 2s with one word of the query left, and in the thick line the share of SERPs judged on both pages with it on either, each keyword judged against its own previous four weeks. The axis stops at 1.5%, so the bursts can be read: page 2 and page 1 or 2 leave the chart in mid-September, and the labels give their highest weeks.

Chart table
Week ofPage 1 or 2Page 1Page 2
March 9, 20260.06%0.01%0.05%
March 16, 20260.05%0.01%0.04%
March 23, 20260.07%0.02%0.05%
March 30, 20260.11%0.03%0.08%
April 6, 20260.06%0.01%0.05%
April 13, 20260.05%0.01%0.04%
April 20, 20260.04%0.01%0.03%
April 27, 20260.04%0.02%0.03%
May 4, 20260.03%0.01%0.02%
May 11, 20260.08%0.03%0.05%
May 18, 20260.10%0.03%0.07%
May 25, 20260.08%0.02%0.05%
June 1, 20260.17%0.07%0.10%
June 8, 20260.16%0.07%0.09%
June 15, 20260.14%0.07%0.07%
June 22, 20260.07%0.02%0.04%
June 29, 20260.05%0.01%0.03%
July 6, 20260.04%0.01%0.03%
July 13, 20260.15%0.02%0.13%
July 20, 20260.35%0.05%0.30%
July 27, 20260.07%0.01%0.06%
August 3, 20260.03%0.01%0.03%
August 10, 20260.04%0.01%0.03%
August 17, 20260.03%0.01%0.02%
August 24, 20260.04%0.02%0.02%
August 31, 20260.03%0.01%0.02%
September 7, 20260.03%0.01%0.02%
September 14, 20266.7%0.55%6.5%
September 21, 202620.0%1.2%19.2%
September 28, 202625.0%0.84%24.6%
October 5, 20266.2%1.3%4.9%
October 12, 20261.5%0.65%0.66%
October 19, 20261.5%0.80%0.48%

Download the data (33 rows, one per week; gzipped CSV). It holds every number in the chart as shares, with no account or keyword in it.

What each column means
week
The Monday that starts the week
page1_pct
Share of the week’s judged page 1s, in percent, with one word of the query left: a word the keyword’s results usually carry is gone from them, and exactly one word of the query is on most of them, each keyword judged against its own previous four weeks
page2_pct
The same share of the week’s judged page 2s
either_pct
Of the week’s SERPs judged on both pages, the share with one word of the query left on page 1 or page 2

Before September it came in bursts that peaked at 0.3% of SERPs a week, on page 1 or 2: late March in continental Europe, June mostly on US desktop, and the week of July 20 in every market. In the week of September 28 it was 25.0%, seventy times the July peak.

From September 16 it hit page 2 first, nearly always under a page 1 that answered the query. Page 1 followed on September 20. This chart follows both pages of the same desktop SERPs; the thick line is page 1 or 2.

Page 1 or 2 ignores part of the query

0.06% to 13.9%

SERPs ignoring part of the query on either page: baseline, then since Sep 16.

Page 1 ignores part of the query

0.02% to 2.2%

SERPs whose page 1 ignored a word of the query: baseline, then since Sep 16.

Page 2 ignores part of the query

0.05% to 12.6%

The same SERPs, with a word of the query ignored on page 2: baseline, then since Sep 16.

Under a page 1 that answers the query

93.6%

Of the SERPs ignoring part of the query on page 2 since Sep 16, the share whose page 1 answered the query sent.

SERPs ignoring part of the query, per day, by page

Page 1 or 2 ignores part of the queryPage 1 ignores part of the queryPage 2 ignores part of the query
BaselineSep 16Page 2 first affectedSep 20Page 1 first affected0%12.5%25%37.5%50%Aug 20Aug 27Sep 3Sep 10Sep 17Sep 24Oct 1Oct 8Oct 157.6%5.1%4.3%

Thick line: SERPs ignoring part of the query on page 1 or page 2. Shaded: the baseline, Aug 20 to Sep 9. Vertical dashes: Sep 16, when page 2 first ignored part of the query, and Sep 20, when page 1 followed. Hollow points: thin days, with fewer than 25,377 SERPs (a fifth of a usual day), where a few SERPs move the line a lot; a break is a day with fewer than 100.

Daily table
DaySERPsPage 1 or 2 words ignoredPage 1 words ignoredPage 2 words ignoredBothPage 1 one wordPage 2 one word
Aug 20210,6750.06%0.01%0.04%0.0%0.01%0.01%
Aug 2112,6760.03%0.0%0.03%0.0%0.0%0.01%
Aug 225,2600.06%0.04%0.02%0.0%0.02%0.02%
Aug 2332,2310.02%0.01%0.01%0.0%0.01%0.0%
Aug 24345,4270.08%0.02%0.06%0.00%0.01%0.02%
Aug 25416,9250.08%0.02%0.07%0.00%0.01%0.02%
Aug 26361,3620.05%0.01%0.04%0.0%0.01%0.01%
Aug 27124,5540.05%0.01%0.04%0.0%0.01%0.01%
Aug 28434,3560.06%0.01%0.04%0.00%0.01%0.01%
Aug 29480,8990.06%0.02%0.04%0.00%0.01%0.01%
Aug 30409,1220.07%0.02%0.05%0.00%0.01%0.01%
Aug 31513,5840.05%0.01%0.04%0.0%0.01%0.01%
Sep 1409,3890.05%0.02%0.04%0.00%0.01%0.01%
Sep 2511,8330.05%0.02%0.04%0.00%0.01%0.01%
Sep 3376,2680.07%0.02%0.06%0.00%0.01%0.02%
Sep 41,4860.13%0.0%0.13%0.0%0.0%0.07%
Sep 51,4140.0%0.0%0.0%0.0%0.0%0.0%
Sep 61,3880.22%0.07%0.14%0.0%0.07%0.14%
Sep 71,4160.0%0.0%0.0%0.0%0.0%0.0%
Sep 81,4350.0%0.0%0.0%0.0%0.0%0.0%
Sep 98,2310.18%0.01%0.17%0.0%0.01%0.01%
Sep 10378,7700.12%0.04%0.09%0.00%0.02%0.02%
Sep 110––––––
Sep 129–––0.0%––
Sep 133910.0%0.0%0.0%0.0%0.0%0.0%
Sep 147,4190.18%0.09%0.08%0.0%0.09%0.04%
Sep 158,5400.26%0.11%0.15%0.0%0.09%0.06%
Sep 1673,5695.0%0.03%5.0%0.00%0.01%4.6%
Sep 1744,84122.7%0.04%22.7%0.02%0.03%20.4%
Sep 1895,99623.4%0.03%23.4%0.01%0.02%21.6%
Sep 19139,25420.4%0.04%20.4%0.02%0.02%17.9%
Sep 20137,68922.1%0.09%22.1%0.05%0.06%19.7%
Sep 2164,09123.8%0.07%23.7%0.04%0.04%20.8%
Sep 22206,44029.1%0.19%29.0%0.11%0.12%26.2%
Sep 23144,44228.9%0.44%28.7%0.23%0.20%24.8%
Sep 24114,69830.6%0.58%30.3%0.31%0.25%26.8%
Sep 25147,49329.0%1.1%28.4%0.59%0.41%24.6%
Sep 26278,75031.7%1.7%30.9%0.94%0.59%27.1%
Sep 27189,10636.1%2.6%35.1%1.5%0.84%29.8%
Sep 2856,28439.9%1.5%39.3%0.91%0.60%36.5%
Sep 29142,38228.5%2.6%27.2%1.4%0.81%21.9%
Sep 30134,9715.7%2.7%3.9%0.87%0.95%1.3%
Oct 1114,9534.6%2.1%3.2%0.67%0.75%1.1%
Oct 2136,0172.8%1.4%1.8%0.42%0.53%0.74%
Oct 3107,6056.8%3.8%4.3%1.2%1.3%1.4%
Oct 4122,5046.3%3.4%4.0%1.0%1.2%1.5%
Oct 5104,0427.4%4.1%4.7%1.4%1.4%1.5%
Oct 680,9106.0%3.5%3.6%1.0%1.2%1.2%
Oct 7158,1167.3%4.1%4.7%1.4%1.3%1.5%
Oct 8156,8826.5%3.9%3.9%1.3%1.3%1.3%
Oct 9123,9807.4%4.9%4.2%1.7%1.7%1.4%
Oct 10138,2136.7%4.5%3.7%1.5%1.6%1.2%
Oct 11122,3877.5%5.0%4.2%1.7%1.7%1.4%
Oct 12152,7104.9%3.1%2.8%0.99%1.2%0.97%
Oct 13117,8146.9%4.6%4.0%1.7%1.5%1.3%
Oct 14168,2141.2%0.66%0.69%0.19%0.26%0.24%
Oct 15147,9630.89%0.48%0.55%0.14%0.21%0.19%
Oct 16158,2441.4%0.81%0.74%0.18%0.36%0.26%
Oct 17126,8853.2%2.1%1.8%0.66%0.70%0.60%
Oct 18174,6303.0%2.0%1.6%0.61%0.74%0.55%
Oct 19103,0275.1%3.4%2.9%1.1%1.2%0.96%
Oct 20111,6944.1%2.7%2.2%0.86%0.91%0.72%
Oct 2165,4367.6%5.1%4.3%1.8%1.6%1.4%

Download the data (63 rows, one per day; gzipped CSV). It holds every number in the chart and the tiles, with no account or keyword in it.

What each column means
date
Day the SERP was scheduled, UTC
judged
Desktop SERPs that day with five or more organic results on each page, a query of two or more words and a keyword with five or more baseline SERPs
page1_other_query
Of those, sERPs whose page 1 ignores part of the query: a word of the query that at least 60% of the keyword’s baseline page-1 results carried is on at most 20% of them
page2_other_query
The same for page 2
both_other_query
SERPs ignoring part of the query on both pages; page1_other_query + page2_other_query − both_other_query is the page 1 or 2 line
page1_one_word
Of the SERPs ignoring part of the query on page 1, those whose results carry exactly one word of the query, on 70% of them or more
page2_one_word
The same for page 2

Page 1 alone covers every SERP of the study: 0.7 to 3.8% a day since September 20, against 0.02% before. The rest of the study counts page 1. Pick a country or a device.

Ignores part of the query

0.02% to 1.8%

SERPs with a word of the query dropped from their results: baseline, then every SERP since Sep 16.

One word left

51.5%

Of the SERPs ignoring part of the query since Sep 16, the share whose results carry exactly one word of it.

Worst day

3.8%

Oct 8, the highest daily share since Sep 16 on a day with enough SERPs.

Normal on the next SERP

85.8%

Of the SERPs ignoring part of the query since Sep 16, the share whose keyword’s next SERP, within three days, answered the query.

Share of SERPs ignoring part of the query, per day

BaselineSep 16Page 2 first affectedSep 20Page 1 first affected0%1.25%2.5%3.75%5%Aug 20Aug 27Sep 3Sep 10Sep 17Sep 24Oct 1Oct 8Oct 152.9%

Shaded: the baseline, Aug 20 to Sep 9. Dashed: Sep 16, when page 2 first ignored part of the query and the period the tiles compare begins, and Sep 20, when page 1 followed. Hollow points: thin days, with fewer than 191,540 judged SERPs in this slice (a fifth of its usual day), where a few SERPs move the line a lot.

Daily table
DayAll SERPsJudged SERPs
Aug 200.03%903,685
Aug 210.01%1,097,353
Aug 220.02%900,104
Aug 230.01%1,060,631
Aug 240.03%957,702
Aug 250.02%1,052,966
Aug 260.02%1,293,109
Aug 270.01%1,132,347
Aug 280.02%1,056,705
Aug 290.03%1,152,569
Aug 300.03%1,002,382
Aug 310.02%1,205,684
Sep 10.03%912,220
Sep 20.03%1,198,440
Sep 30.03%950,897
Sep 40.01%1,157,679
Sep 50.01%942,298
Sep 60.01%1,103,389
Sep 70.02%995,133
Sep 80.02%1,085,699
Sep 90.02%979,684
Sep 100.04%1,082,600
Sep 110.02%953,121
Sep 120.03%1,057,942
Sep 130.03%854,781
Sep 140.14%690,424
Sep 150.04%282,867
Sep 160.04%179,089
Sep 170.06%103,157
Sep 180.07%224,010
Sep 190.06%353,696
Sep 202.5%576,353
Sep 213.0%292,133
Sep 221.0%531,341
Sep 232.3%417,703
Sep 243.3%354,437
Sep 251.1%390,581
Sep 261.6%616,536
Sep 272.3%429,394
Sep 281.5%135,630
Sep 292.7%330,440
Sep 303.0%1,181,853
Oct 11.2%962,055
Oct 20.75%1,145,805
Oct 31.7%919,690
Oct 41.4%1,074,493
Oct 52.6%968,472
Oct 62.5%1,063,421
Oct 72.2%964,383
Oct 83.8%1,063,071
Oct 92.0%956,004
Oct 102.0%1,052,944
Oct 112.1%901,962
Oct 121.5%1,102,492
Oct 131.9%889,327
Oct 141.2%1,091,955
Oct 151.2%885,307
Oct 160.74%1,080,945
Oct 171.3%869,433
Oct 181.4%1,001,524
Oct 192.3%919,950
Oct 201.7%1,023,214
Oct 212.9%901,950

Download the data (1,890 rows, one per day, country and device; gzipped CSV). It holds every number in the chart and the tiles, with no account or keyword in it.

What each column means
date
Day the SERP was scheduled, UTC
country
Target country, ISO 3166 code; other is every country under 1% of SERPs, combined
device
desktop, iphone or android
judged
SERPs with five or more organic results, a query of two or more words, and a keyword with five or more baseline SERPs
other_query
Of those, SERPs ignoring part of the query: a word of the query that at least 60% of the keyword’s baseline results carried, in their titles, snippets or bolded text, is on at most 20% of them
one_word
Of those, SERPs whose results carry exactly one word of the query, on 70% of them or more
followed_up
SERPs ignoring part of the query followed by another judged SERP of the keyword within three days
back
Of those, SERPs whose next SERP answered the query sent

The device barely matters.

What else plays dumb

The organic results are not the only part that plays dumb. On the normal SERP for “trailer lock reviews”, “People also ask” asks “What is the most effective trailer lock?”. On the page answered as “reviews”, it asks “Can I trust product reviews?”.

People also ask

Share of its questions

About the full query

Normal SERP49.5%
Words ignored29.5%

Only about the one word left

Normal SERP27.3%
Words ignored45.9%

436,111 questions on 109,588 normal SERPs, and 421,165 on the pages with one word left that followed them. The rest have neither: 23.1% and 24.6%.

People also search for

Share of its suggested searches

About the full query

Normal SERP69.5%
Words ignored18.7%

Only about the one word left

Normal SERP10.7%
Words ignored57.5%

826,224 suggested searches on 113,143 normal SERPs, and 827,809 on the pages with one word left that followed them. The rest have neither: 19.8% and 23.9%.

Download the data (4 rows, two per part; gzipped CSV). It holds every number in the chart as counts of items, with no account or keyword in it.

What each column means
part
The part of the page: People also ask, its questions, or People also search for, its suggested searches
serp
normal, the keyword’s previous SERP, 48 hours or less before, that answered the query and was not poisoned; or other_query, the page-1 SERP after it with one word left, since September 16
pairs
Pairs of the two SERPs that both show the part
items
The distinct questions or searches the part showed on those SERPs, every pair together
query_items
Items about the full query: with a word the organic results dropped on the page ignoring part of the query
one_word_items
Items only about the one word left: with the word left on that page and none of the dropped ones
neither_items
Items with neither

“People also search for” follows the one word most: “eth price” goes from suggesting “ETH price prediction” to “Price HVAC”. Google’s own panels mostly hold: shopping, the image pack, the knowledge panel and the map keep nearly as much of the keyword’s previous SERP as normal pages do, while the organic results keep 18.5% of its sites.

What each part of the page keeps from the keyword’s previous SERP

  • Baseline, Aug 20 to Sep 9
  • Pages ignoring part of the query
0%25%50%75%100%

Keeps less on pages ignoring part of the query

  • Organic results
    90.0% → 18.5%
  • People also search for
    89.2% → 26.4%
  • People also ask
    87.1% → 25.0%
  • AI Overview sources
    72.0% → 52.4%
  • Discussions and forums
    92.5% → 75.7%
  • Local pack
    87.1% → 71.3%
  • Video
    81.5% → 69.4%
  • Knowledge panel
    87.2% → 77.2%
  • Image pack
    90.6% → 81.9%
  • Shopping
    91.7% → 88.6%

Keeps more on pages ignoring part of the query

  • Ads
    74.5% → 88.4%

Of the sites, questions, searches or businesses a part showed on the keyword’s previous SERP that answered the query, 48 hours or less before, the share it shows again, where both SERPs show that part. Each keyword’s pages ignoring part of the query against its own baseline, each keyword weighted by its count of them.

Part table
PartKept from the previous SERPShown on
BaselineWords ignoredNormal since Sep 16BaselineWords ignoredNormal since Sep 16
Organic results90.0%18.5%74.8%100.0%100.0%100.0%
People also search for89.2%26.4%78.1%81.1%61.7%76.5%
People also ask87.1%25.0%68.4%76.8%63.5%73.3%
AI Overview sources72.0%52.4%66.5%55.8%62.0%64.4%
Discussions and forums92.5%75.7%90.3%18.4%17.3%18.7%
Local pack87.1%71.3%69.6%28.1%30.9%29.7%
Video81.5%69.4%80.1%19.5%24.1%20.3%
Knowledge panel87.2%77.2%80.1%5.4%6.9%7.3%
Image pack90.6%81.9%83.6%27.0%31.4%28.3%
Shopping91.7%88.6%94.2%9.6%9.6%10.0%
Ads74.5%88.4%80.9%9.0%9.8%7.1%

Download the data (11 rows, one per part of the page; gzipped CSV). It holds every number in the chart and the table, with no account or keyword in it.

What each column means
part
The part of the page
kept_baseline_pct
Of the sites, questions, searches or businesses the part showed on the keyword’s previous SERP that answered the query, the share it shows again, where both SERPs show the part: baseline SERPs, August 20 to September 9
kept_other_query_pct
The same for pages ignoring part of the query since September 16, 2026, each against the keyword’s previous SERP that answered the query
kept_normal_pct
The same for the same keywords’ SERPs since September 16, 2026 that answer the query
shown_baseline_pct
Share of baseline SERPs showing the part
shown_other_query_pct
Share of pages ignoring part of the query showing the part
shown_normal_pct
Share of the same keywords’ SERPs since September 16, 2026 that answer the query showing the part

Countries, searches and industries

Two thirds of the study’s datasets had at least one page ignoring part of the query. Belgium ran highest, Japan lowest.

Share of SERPs ignoring part of the query since Sep 16, by target country

World
Europe
  • Under 1%
  • 1–1.5%
  • 1.5–2%
  • 2–3%
  • 3% or more
  • Not enough data

A country needs 10 datasets and 10,000 judged SERPs to be coloured. Hong Kong and Singapore are too small to draw at this scale and show as a dot.

Country table
CountryIgnores part of the queryJudged SERPs
Belgium4.8%316,062
Thailand4.0%20,370
New Zealand2.7%178,521
Norway2.6%34,458
Austria2.5%76,304
Mexico2.5%210,463
Portugal2.4%197,077
Switzerland2.3%66,949
Argentina2.2%55,333
Bulgaria2.2%24,195
United States2.1%16,765,307
Netherlands2.0%271,887
Romania1.9%24,479
Poland1.8%66,950
Sweden1.8%48,955
Australia1.7%475,621
Israel1.7%23,909
Saudi Arabia1.7%44,327
South Africa1.7%65,579
Spain1.7%742,946
Brazil1.6%126,906
Singapore1.6%65,500
United Arab Emirates1.6%48,455
Chile1.5%30,359
Taiwan1.5%13,800
Canada1.4%1,975,943
Colombia1.4%16,423
Germany1.4%555,326
India1.4%225,060
South Korea1.4%14,174
Finland1.3%32,032
Indonesia1.3%20,281
Italy1.3%254,563
Türkiye1.3%39,191
Ukraine1.3%16,106
France1.2%596,558
Greece1.2%42,087
Ireland1.2%89,263
Malaysia1.2%17,703
Peru1.0%22,124
Hong Kong0.9%18,147
United Kingdom0.8%2,590,408
Vietnam0.8%19,064
Japan0.6%173,089

Download the data (44 rows, one per country; gzipped CSV). It holds every number on the map, with no account or keyword in it.

What each column means
country
Target country, ISO 3166 code
judged
SERPs with five or more organic results, a query of two or more words, and a keyword with five or more baseline SERPs, since September 16, 2026
other_query
Of those, SERPs ignoring part of the query: a word that at least 60% of the keyword’s baseline results carried is on at most 20% of them
  • Longer queries more. 0.9% of two-word queries, 3.2% of five words or more.
  • The long tail more. 2.1% under 100 searches a month, 0.5% at 100,000 or more.
  • “Near me” and postal codes less. 1.4% and 1.1%, against 1.8% overall.

Each ridge is one industry’s US page 1s, day by day; the grey ridge is every US SERP.

Share of US SERPs ignoring part of the query per day, by industry

BaselineSep 16Page 2 first affectedSep 20Page 1 first affectedEvents and entertainmentReal estateAll US SERPsTravel and hospitalitySoftware and technologyTelecom and internetTransportationJobs and careersHealth and medicalHome servicesLegalFinance and bankingPetsRestaurants and foodRetail and shoppingEducationInsuranceWeddingsAutomotiveBeauty and fashionHome and gardenSince Sep 163.0%2.6%2.1%2.1%1.9%1.9%1.9%1.8%1.7%1.7%1.7%1.6%1.6%1.6%1.6%1.5%1.5%1.5%1.2%1.1%1.1%Aug 20Aug 27Sep 3Sep 10Sep 17Sep 24Oct 1Oct 8Oct 15

Every ridge is on one scale: the tallest, events and entertainment SERPs on Sep 29, is 7.1%. Grey: every judged US SERP. Shaded: the baseline. Dashed: Sep 16, when page 2 first ignored part of the query, and Sep 20, when page 1 followed. The numbers are the share since Sep 16, or on the day you point at.

Industry table
IndustrySince Sep 16Worst dayJudged SERPs
Events and entertainment3.0%7.1%, Sep 2988,235
Real estate2.6%5.9%, Sep 294,843,927
All US SERPs2.1%4.0%, Oct 816,765,307
Travel and hospitality2.1%4.1%, Oct 8225,819
Software and technology1.9%4.4%, Oct 8257,666
Telecom and internet1.9%4.0%, Oct 81,568,874
Transportation1.9%4.5%, Oct 8292,471
Jobs and careers1.8%4.0%, Sep 29277,528
Health and medical1.7%3.9%, Sep 24233,238
Home services1.7%4.1%, Sep 24202,909
Legal1.7%5.1%, Sep 2448,538
Finance and banking1.6%4.2%, Sep 24273,934
Pets1.6%4.6%, Sep 2933,364
Restaurants and food1.6%4.4%, Sep 24248,162
Retail and shopping1.6%3.6%, Sep 212,193,597
Education1.5%4.5%, Sep 2453,248
Insurance1.5%6.0%, Sep 2187,467
Weddings1.5%5.1%, Sep 2828,026
Automotive1.2%4.4%, Oct 6536,182
Beauty and fashion1.1%3.2%, Oct 894,018
Home and garden1.1%3.5%, Sep 20599,730

Download the data (1,323 rows, one per day and industry; gzipped CSV). It holds every number in the chart, with no account or keyword in it.

What each column means
date
Day the SERP was scheduled, UTC
industry
The keyword’s industry, or Unclassified
judged
US SERPs that day with five or more organic results, a query of two or more words and a keyword with five or more baseline SERPs
other_query
Of those, SERPs ignoring part of the query: a word that at least 60% of the keyword’s baseline results carried is on at most 20% of them

What it does to rankings

Where a project’s own site is one of the keyword’s usual sites, it was out of the top 10 on 69.8% of these pages, against 2.0% in the baseline. Then it passes: nine spells in ten last one SERP, and the keyword’s next SERP within three days answered the query again 85.8% of the time. A site that “lost” a keyword for a day most likely had its query answered for one word.

How to spot it

  • The title says the full query. On 99.8% of these pages, with no “Showing results for”.
  • Every result is about one word. A ranking change moves sites. It doesn’t change what the results are about.
  • It wears poisoning’s marks. On desktop keywords that usually run to page 8, pagination stops by page 4 on 43.0% of these pages, against 0.2% normally, and one domain fills three spots or more on 32.2%, against 10.0%.
  • More than half also fail the poisoning test. 54.0% lose most of the keyword’s usual sites too. The words decide which fault it is.
  • It is older. It came and went from March (When it hit), and SerpApi has tracked it since May.

Method

  • SERPs. A representative sample of the SERPs Nozzle tracks for its customers, the poisoned study’s datasets: 47 countries, desktop, iPhone and Android, from March 9 to October 21, 2026. A smaller sample of them back to October 2025 never passes 0.07% of page 2s a week before March.
  • Judged. A query of two words or more besides stop words, five organic results or more, and five earlier SERPs of the keyword to compare with. Only judged SERPs are counted: 253 million in all, 248 million since March 9, 54.0 million from August 20 and 27.0 million since September 16.
  • The test. A word is usual when 60% of the keyword’s results carry it, in titles, snippets or Google’s bolding. A page ignores part of the query when a usual word is on 20% of its results or fewer, and has one word left when exactly one query word is on 70% or more. The history compares each keyword with its previous four weeks; every other chart with its SERPs from August 20 to September 9.
  • Page 1 or 2. A SERP counts once when either page ignores part of the query, of the SERPs judged on both pages: 4.7 million desktop SERPs since September 16.
  • Questions and suggestions. Compared with the keyword’s normal SERP before, 48 hours or less earlier. One counts as about the full query when it has a word the results dropped.
  • Examples. Picked because they show plainly which words went. Each renders Nozzle’s stored page-1 HTML in Chrome with Google’s scripts blocked. The sites are Google’s results, not Nozzle customers.