PrismCrawl
← Back to blog

Published

Google Plays Dumb: When Results Answer Only One Word of Your Query

Search Google for "easy strawberry jam" and you expect jam recipes. On some requests in autumn 2026, the page title still showed the full query, the AI Overview still gave a jam recipe, and the organic results were about a TV series called Easy.

Rank tracker Nozzle calls this pattern Google playing dumb. The page looks normal and returns HTTP 200. Most of the organic results answer one word of the query and ignore the rest.

It is a companion to poisoned SERPs, where Google swaps a keyword's usual sites for others. The two faults often appear together, but they are measured differently, and the partial query version is easier to detect from a single page.

What a partial query results page looks like

Nozzle's published examples show the pattern clearly:

  • "easy strawberry jam": five of six usual sites on page one were gone, and the results followed the word "easy".
  • "trailer lock reviews": five of seven usual sites were gone, and the results followed "reviews".
  • "white label local seo": all nine usual sites were gone, and the results followed "local".

In each case the results are about something, just not what was asked. A reader can spot the problem at a glance. A parser can't, because every field it reads is present and well formed.

How Nozzle measures it

Nozzle's method works on the words of the query rather than on domains:

  1. A query word counts as "usual" for a keyword when it appears in the titles, snippets, or bolded text of at least 60% of that keyword's results.
  2. A page ignores part of the query when a usual word appears on 20% of its results or fewer.
  3. A page has one word left when exactly one query word appears on 70% or more of its results.

The baseline is August 20 to September 9, 2026. Nozzle's study covers 52 countries and desktop, iPhone, and Android results from March 9 to November 3, 2026.

How often it happens

These figures come from Nozzle's study as updated November 3, 2026. Nozzle is a rank tracking vendor and measures its own customers' keywords, so treat the numbers as one large sample.

  • Onset. Small bursts appeared from March 2026. The current wave started on page two on September 16, 2026, and reached page one on September 20.
  • Rate. Since September 16, 13.1% of SERPs judged on both pages had page one or page two affected, compared with 0.06% at baseline. The daily peak was 39.9% on September 28.
  • Page one. Counting page one alone, the rate is 2.6% against a 0.02% baseline. Page two is hit much harder, at 11.4%.
  • Words dropped. On 70.4% of affected pages, Google dropped one word of the query.
  • Rankings. When a tracked site normally ranked for the keyword, it was outside the top 10 on 65.9% of affected pages, compared with 1.9% at baseline.
  • Duration. Three spells in four lasted a single SERP. The next SERP within three days answered the full query 61.8% of the time.
  • Query shape. Longer and rarer queries are hit more often. Nozzle reports 1.3% of two-word queries affected against 4.8% of queries with five or more words, and 3.1% of keywords under 100 monthly searches against 0.6% of keywords with 100,000 or more.

The overlap with poisoning is large. Nozzle found that 56.6% of partial query pages also lost most of the keyword's usual sites, which means they would count as poisoned too.

Others have seen it

SerpApi's public roadmap has an open issue, first reported on May 4, 2026, about Google searches "intermittently receiving irrelevant results." Nozzle cites it as the earliest public report of the fault. Later comments on the issue describe poisoned results across Google Search and Google Shopping. On October 26, 2026, a SerpApi team member wrote that changes had reduced the frequency, but "the issue is still present."

Google has not commented publicly.

How to detect partial query results

This fault has one advantage over poisoning: you can often catch it from the page alone. A normal results page for a longer query mentions most of the query's meaningful words across its titles and snippets. A partial query page leans heavily on one of them.

The check below counts how many results mention each query word and flags pages where one word dominates while another almost disappears. It uses the same thresholds as Nozzle's method.

import re

STOPWORDS = {"a", "an", "and", "the", "for", "of", "to", "in", "on", "with", "how", "what", "is"}


def word_coverage(query: str, results: list[dict]) -> dict[str, float]:
    """Share of results whose title or snippet mentions each query word."""
    words = [w for w in re.findall(r"[a-z0-9]+", query.lower()) if w not in STOPWORDS]
    texts = [f"{r.get('title', '')} {r.get('snippet', '')}".lower() for r in results]
    if not texts:
        return {}
    return {w: sum(w in t for t in texts) / len(texts) for w in words}


def looks_partial(query: str, results: list[dict]) -> bool:
    coverage = word_coverage(query, results)
    if len(coverage) < 2:
        return False  # a one-word query can't lose a word
    shares = sorted(coverage.values())
    return shares[-1] >= 0.7 and shares[0] <= 0.2

A few refinements make it more reliable in production:

  • Compare with the keyword's history. Some words rarely appear in results even on a good page, such as a brand name the results spell differently. Nozzle only checks words that are usual for the keyword. If you store history, use the same rule.
  • Handle plurals and variants. "Reviews" and "review" should count as one word. A simple stemmer helps.
  • Fetch again before reporting. Most spells last one SERP, so a second request a few minutes later usually settles it.

This check costs almost nothing to run: one pass over the titles and snippets you already parsed. It pairs well with the history comparison in our poisoned SERP guide, which catches swaps that still mention the query's words.

Why this matters for SERP data

A partial query page breaks rank tracking the same way a poisoned page does. Tracked sites vanish, unrelated sites appear, and the report shows a ranking loss that never happened for real searchers. It also hurts anyone using search results to ground an AI answer, because the retrieved sources are confidently off topic.

As with every silent scraper failure, the request succeeds and nothing in the HTML marks the problem. The defense is a correctness check on the content, plus a retry when the check fails.

Frequently asked questions

What does it mean when Google "plays dumb"?

It describes a Google results page that looks normal but whose organic results answer only one word of the query. For example, a search for "easy strawberry jam" returned results about a TV series called Easy. Rank tracker Nozzle described and measured the pattern in October 2026.

How is this different from a poisoned SERP?

A poisoned SERP is defined by domains: most of the keyword's usual sites are missing. A partial query page is defined by words: most results ignore part of the query. The two often overlap. Nozzle found that 56.6% of partial query pages also lost most of the keyword's usual sites.

How common are partial query results?

In Nozzle's data, 13.1% of SERPs judged on both pages had page one or page two affected since September 16, 2026, against a 0.06% baseline. Page one alone was affected 2.6% of the time. Page two is hit much more often.

Can I detect a partial query page without historical data?

Often, yes. Count how many results mention each meaningful word of the query. If one word appears on 70% or more of the results while another appears on 20% or fewer, the page probably ignored part of the query. Historical data makes the check more reliable.

Has Google confirmed this behavior?

No. As of November 2026, Google has not commented publicly. The evidence comes from rank trackers and SERP API providers analyzing their own collected results.

Related posts