Which searches show Google AI Overviews more often?
Which searches show an AI summary on Google? Google calls these summaries AI Overviews. When planning content, track whether a summary appears and whether it cites your website as separate outcomes. In this article accepted at a peer-reviewed conference, researchers found Overviews in 64.7% of question-form searches and 9.5% of other searches within a US trending-query sample. DijitalPi commentary: Track customer questions separately from other searches when deciding content priorities. However, the study does not establish that turning a headline into a question secures a citation. User behaviour, traffic loss and sales effects were not measured, and the rates cannot be transferred directly to Turkish searches or Türkiye.
ACADEMIC WORK REVIEWED
Measuring Google AI Overviews: Activation, Source Quality, Claim Fidelity, and Publisher Impact
ResearchersHaofei Xu, Umar Iqbal, Jacob M. Montgomery
arXiv / ACM IMC 2026 · 1 October 2026 · v2, accepted at a peer-reviewed conference · Peer-reviewed article · Source language: English
English review: 15 October 2026 · The source date appears in the citation above.
Editorial review draft
The source has been checked; the review still carries its original editorial status. DijitalPi did not conduct a customer experiment for this paper.
LET'S READ THE RESEARCH TOGETHER
We first explain the researchers' question, method and findings. We then discuss how to interpret the results, clearly separating DijitalPi's commentary from the source.
01 / WHAT DID THE RESEARCHERS WANT TO UNDERSTAND?
Which searches produce Google AI Overviews, which sources do they select, and how faithfully do they represent those sources?
The researchers examined Overview activation, source selection, agreement between claims and sources, and advertising on cited pages together. This allowed them to assess both how often summaries appeared and which publishers and source material they relied on.
Explaining the context · DijitalPi commentary
DijitalPi commentary: Content planning involves separate decisions: which customer questions to answer, which searches to monitor, and which business outcomes to associate with visibility. This research offers observations relevant to the first two decisions. Establishing whether a citation brings visits or customers requires additional behavioural data.
02 / HOW WAS THE RESEARCH CONDUCTED?
How did they test the question?
Haofei Xu, Umar Iqbal and Jacob M. Montgomery of Washington University in St. Louis collected queries from 19 categories on the US Google Trends dashboard over 40 days, from 13 March to 21 April 2026. From a single AWS us-east-1 location in Northern Virginia, they ran 55,393 query executions covering 32,608 unique queries, using a clean, cookie-free browser profile for each search. They observed 7,583 AI Overviews. This was an observational study, without random assignment or an experiment estimating causal effects.
Understanding the method · DijitalPi commentary
The researchers first recorded whether each query produced an Overview. A query counted as question-form if its first word matched one of 15 specified interrogatives. This classification applied to the search query, not the headline of a content page. They compared Overview sources with organic results on the first page of the same search. The researchers assessed website credibility using a score called PC1. It ranges from 0 to 1 and combines several credibility indicators. They excluded websites without a score from this comparison.
The researchers split Overview text into claims that could be checked individually. They used Grok 4.1 Fast Reasoning to compare each claim with its cited source. Claims with explicit or inferred support counted as consistent. Conflicting sources, direct contradiction and absence of support were separate inconsistency categories. Claim extraction was checked against human assessment on 100 Overviews, and verification on 100 verdicts. Advertising measurement counted embedded advertising frames matching EasyList filters in source-page code. It did not measure user clicks or publisher revenue.
03 / RESEARCH FINDINGS
Question-form queries produced Overviews more often, and cited sources did not fully match first-page results.
The researchers found Overviews in 13.7% of all query executions, 64.7% of question-form queries and 9.5% of other queries. Of the cited source domains, 29.8% were absent from the first page of the corresponding search. Among sources with PC1 scores, mean credibility was 0.732 for Overview sources and 0.611 for all first-page results. However, the first 1 or first 3 organic results scored more highly than Overview sources. A second credibility measure produced a small difference in the opposite direction that was not statistically significant. Of 98,020 checked claims, 88.97% were classified as consistent with their sources and 11.03% as inconsistent. 6.98% were not addressed by the sources, and 2.66% directly contradicted a source. The advertising method also detected visible ads on 30,994 of 61,212 cited URLs (50.63%). That measurement does not establish traffic or revenue loss.
Overview activation for question-form and other queries
64.7% / 9.5%These are rates for two query groups in the US trending-query sample, not the effect of changing a content headline.
Cited source domains absent from the first page
29.8%Overview source selection did not fully overlap with first-page rankings for the same search.
DijitalPi commentary: When grouping searches by question form, record Overview presence, citation of your own page and faithful representation of that page separately. Repeat observations on different days: in the study, Overview activation agreed across repeated searches in 95.92% of comparisons, but only 0.24% of comparisons where both searches produced an Overview had exactly the same cited URL set. A single screenshot should not be treated as evidence of lasting visibility.
The 11.03% classified as inconsistent is not a direct misinformation rate. Uncrawled social content and paywalls can inflate the number of claims lacking detected support. Under the authors’ assumption that treats the most claims as supported, residual inconsistency falls to roughly 5.3%. This is an assumption-based calculation, not a definitive corrected rate. The advertising figure is also a lower bound because uncrawled social and video platforms were treated as carrying no ads.
05 / CONCLUSION AND OPEN QUESTIONS
What did we learn, and what do we still not know?
The reviewed arXiv v2 text is a peer-reviewed conference article accepted at ACM IMC 2026. Its findings indicate that query structure, source selection and faithful representation should be assessed separately. However, the single US location, 40-day trending-query sample and 51.4% share of sports queries limit generalisation; the study supplies no directly transferable rates for Turkish searches or Türkiye. User behaviour, personalisation and traffic loss were not measured. Credibility comparisons cover only scored domains, and human validation does not eliminate sampling or evaluator bias. The work was partially supported by NSF grant CNS-2543517. All three authors are university researchers. No company affiliation is stated, and no conflict-of-interest declaration was found in the source.
Our open question at DijitalPi: which groups of Turkish customer questions regularly produce Overviews, and does being cited in those Overviews occur alongside observable behaviour on your website?
We created these scenarios to make the topic concrete. They are not cases from the paper or measured client results.
EXAMPLE 01
E-commerce: product names and care questions
An illustrative furniture retailer is deciding how much content work to allocate to product names and care questions.
Will publishing answers to care questions get us cited in an AI Overview?
Open the recommendation for example 01+
First observe whether Overviews appear and whether your retailer is cited as separate outcomes.
Group product-name searches separately from question-form care searches. On different days, record Overview presence and cited sources using consistent location and browser conditions. Track your retailer’s first-page ranking separately. The research examined query structure. It did not show that changing a content headline into a question increases citation.
EXAMPLE 02
Tourism: a citation without a first-page ranking
An illustrative tourism business finds its local guide cited in an AI Overview for a search where it does not appear on the first page.
How should we assess this visibility if we are absent from the first page?
Open the recommendation for example 02+
Record Overview citations separately from organic rankings.
Note the page linked from the Overview alongside the first-page results for that search. Check whether the citation persists on different days. Monitor site visits and booking enquiries separately rather than treating a citation as either outcome. The paper’s 29.8% concerns cited domains absent from the first page, not an expected level of visits or bookings for your business.
EXAMPLE 03
B2B software: a cited feature explanation
An illustrative business software company sees an AI Overview describe a feature differently while linking to its own help page.
Does a link to our page mean the explanation represents it accurately?
Open the recommendation for example 03+
Check the citation and the product claims separately.
Match each product claim in the Overview to the relevant wording on the linked page. Record claims absent from the source separately from claims that directly contradict it. Where another cited source is inaccessible, mark the assessment as uncertain. The paper’s inconsistency rate does not mean that every such claim is false.
TRY IT WITH YOUR TEAM
Build your query monitoring list.
Separate question-form searches from other customer searches, then plan to record Overview presence, citations and agreement with source text on different days. Measure visits and business outcomes separately; this is a proposed pilot that DijitalPi has not conducted.
As search results change, review your search and traffic data to decide which pages need improvement. SEOMACHINE helps turn those data into content priorities; visibility in AI answers needs a separate check. Explore SEOMACHINE ↗
The academic work belongs to the researchers named above. This page contains DijitalPi's explanatory review and original business examples; it is not a full translation of the paper.
Cookie preferences