We ran the same 19 auto insurance searches in all 50 states and 6 Canadian provinces. Whether an AI Overview appears turns out to be almost entirely decided by one thing, and it is not the topic.
Most published work on AI Overviews reports a single presence rate for a category. We wanted to know whether that number means anything, so we held the search set constant and varied only the geography: the same 19 auto insurance searches, geo-targeted to each of 50 states and 6 provinces, in one capture window in July 2026.
The category-level number is 76.6%. It is also close to useless, because the searches inside it are not behaving as one population.
There is no middle. Four intent groups sit between 90% and 97%. Two sit under 13%. Nothing lands in between.
The near-me result is the sharpest. Two searches, "car insurance near me" and "auto insurance agents near me", run across 56 geographies, produced 112 captures. An AI Overview appeared in exactly one of them, in one geography. Every single one returned a Local Pack.
The finding we did not expect is that "car insurance" and "auto insurance", with no modifier at all, sit with the near-me group rather than with everything else. Overview presence of 12.5%, Local Pack presence of 98.2%.
Google is reading an unmodified head term as a local intent, not an informational one. For a shopper typing two words, it returns a map of nearby agencies rather than a written explanation. That is a reasonable interpretation of an ambiguous query, and it has a large consequence: the two highest-volume searches in the entire category are not being intercepted by the answer layer at all.
| Search | Intent | AI Overview | Local Pack | Sources |
|---|---|---|---|---|
| does car insurance cover rental cars | informational | 98.2% | 0.0% | 6.6 |
| what does full coverage car insurance cover | informational | 98.2% | 0.0% | 7.4 |
| how much car insurance do i need | informational | 96.4% | 0.0% | 5.5 |
| what is minimum car insurance coverage | informational | 96.4% | 0.0% | 5.7 |
| how do car insurance deductibles work | informational | 94.6% | 0.0% | 6.1 |
| car insurance comparison | comparison | 100.0% | 0.0% | 8.0 |
| cheapest car insurance | comparison | 98.2% | 7.1% | 5.8 |
| best auto insurance | comparison | 96.4% | 26.8% | 7.5 |
| best car insurance companies | comparison | 92.9% | 87.5% | 4.7 |
| how much is car insurance | cost | 98.2% | 0.0% | 9.8 |
| car insurance rates | cost | 96.4% | 0.0% | 6.7 |
| average car insurance cost | cost | 92.9% | 0.0% | 4.6 |
| car insurance quotes | quote | 96.4% | 0.0% | 7.4 |
| auto insurance quotes | quote | 89.3% | 0.0% | 7.1 |
| cheap car insurance quotes | quote | 83.9% | 8.9% | 5.4 |
| car insurance | head | 17.9% | 98.2% | 0.5 |
| auto insurance | head | 7.1% | 98.2% | 0.6 |
| auto insurance agents near me | near me | 1.8% | 100.0% | 0.1 |
| car insurance near me | near me | 0.0% | 100.0% | 0.0 |
Each row is 56 captures, one per geography. Auto insurance, July 2026.
One search sits on the boundary. "Best car insurance companies" returned an Overview 92.9% of the time and a Local Pack 87.5% of the time, the only search in the set where both features appear together in most geographies. It is the hardest search here to be an organic result on.
Presence ranged from 63.2% in Alberta, British Columbia and Colorado to 84.2% in Maine, New Jersey, North Dakota and Wyoming, with a standard deviation of 5.7 points. Meaningful, but an order of magnitude smaller than the intent effect, which spans 96 points.
Canada ran lower than the United States overall, 69.3% against 77.5%, and its answers cited fewer sources, 4.8 against 5.3. British Columbia was among the lowest measured anywhere, which is worth noting in a province where auto insurance is publicly administered.
Across 5,571 citations, carriers supplied 34.7%. Comparison and quote sites supplied 25.6%, publishers 17.0%, and forums including Reddit and YouTube 12.0%. State regulators supplied 3.9%.
The smallest group is the most interesting. Personal injury law firms supplied 1.7% of all sources, 95 citations, largely on searches about minimum coverage requirements. They are competing for space in insurance answers and almost no insurance marketer is watching them.
If you sell insurance locally, the searches that produce phone calls are the ones the answer layer has left alone. Near-me and head terms are still a Local Pack fight, and a Local Pack fight is winnable by a small business in a way that a national content fight is not.
If you are a carrier, the reverse. Comparison, cost and coverage-explainer searches now return an answer above your result more than 95% of the time. Writing a better explainer page will not recover that click. It may still be worth writing, but as citation material rather than as a landing page, and those are different jobs with different success measures.
Sample — 19 auto insurance searches, geo-targeted by location code to each of 50 US states and 6 Canadian provinces, captured July 2026. 1,064 keyword-geography pairs.
Selection — searches were chosen by intent quota, not by search volume or cost per click. A value-ranked set would have contained almost no informational or cost terms, since those carry little CPC, and the finding would have been invisible.
Weighting — presence rates here are unweighted. One search, one vote. Weighting by volume would bias the result toward commercial terms and distort the intent comparison.
Limits — this is a single capture window, not a three-capture median. A repeat run of one geography hours apart moved presence by about 5 points, so treat differences smaller than that as noise. Intent classification is a judgment call. Coverage is Google AI Overviews only.
Next — the same 19 searches rerun monthly. Home and commercial lines to follow.
In this study, 76.6% of 1,064 auto insurance captures returned one. That single figure is misleading on its own: presence ranged from 0.9% on near-me searches to 96.9% on comparison searches.
Almost never in this data. One of 112 near-me captures returned an Overview. All 112 returned a Local Pack.
Only slightly. State-level presence ranged from 63.2% to 84.2%, a standard deviation of 5.7 points, against an intent effect spanning 96 points.
Monthly, using the same fixed search set, and this page is updated in place rather than republished. The dateline shows the capture window in use.
We run this analysis on your query set, your states and your lines, and send you the composition table with your competitors on it.