Google's AI Overview appeared in only 18 of 90 local searches in a new test by GBP SEO Tools, and when it did name businesses, it mostly just repeated the map pack. The test ran 15 service trades, each in its own US city, four times in the middle of the night and twice more at midday, all on mobile. The map pack showed on all 90 searches. The AI answer showed on 18, or 20%, and not one of the 15 searches got an AI answer in all six rounds.
The headline finding is how unsteady the AI answer is. Seven of the 15 searches never got one at all. The other eight got one sometimes: personal injury lawyer in Phoenix and accountant in Portland in four rounds of six, plumber in Austin in one. The same search, minutes apart, could show a full list of local plumbers, then no AI answer, then a two-business answer. A single check of a keyword tells you almost nothing about whether a business usually appears in the AI answer.
The midday rounds confirmed the pattern: the AI answer showed on 6 of 30 midday searches, the same 20% as at night. What changed was the map pack itself: only 2 of the 15 searches had the same three businesses at night and at midday, as closed businesses dropped out overnight. The AI answer moved with the pack, naming the night businesses at night and the midday businesses at midday.
Twelve of the AI answers named local businesses, and in ten of them the businesses were exactly the ones in the map pack, with the answer sitting below the pack rather than above it. For electrician in Charlotte, the AI named the pack's three businesses and described them as local providers with high customer satisfaction ratings, then asked what kind of job the searcher needed. Plumber in Austin and personal injury lawyer in Phoenix got the same treatment. In most cases, the business that wins the AI answer is the business that already won the map.
Only twice in 90 searches did the AI build its own list. For chiropractor in San Diego, the AI Overview sat above the map pack with a block titled Top-Rated San Diego Chiropractors: four business cards with ratings and Call, Directions and Website buttons. Two of the four were the map pack's businesses, and the other two were not in the pack at all, but both had more reviews than either pack business, 399 and 517 against 138 and 89. For personal injury lawyer in Phoenix at midday, the AI ran a long answer with a list of popular firms, three from the pack and one outside it, with addresses but no buttons.
Five AI answers sat above the map pack and named no local business at all. For accountant in Portland, in four rounds out of six, the AI explained what an accountant is. Physical therapist in Philadelphia got a definition too, citing clinics in Tulsa. These answers push the map pack down the screen without naming anyone, and they mostly hit trades whose name is also a job title. There is nothing to win inside them, so the work is making the listing worth the scroll once the customer gets past it.
This test sits against the bigger worry from earlier in the month. On November 10, Search Engine Roundtable reported that Google is testing a stripped-down local pack inside AI Overviews, with no call, directions or website buttons, no photos and no reviews, spotted by local search expert Joy Hawkins. In the GBP SEO Tools test, the closest thing to that version appeared once, and it still kept its buttons. It is still a test, Google has not announced it, and most searchers do not see it.
For local SEOs, the practical message is calmer than the headlines. Winning the map pack is still the job, because the AI answer copies it ten times out of twelve. Keep hours current, list the main services, and reply to reviews so the recent ones describe what the business does, because the AI summary leans on exactly those profile details. When Google's AI does build its own list, a deep review history counts: the two businesses it added in San Diego had far more reviews than the pack businesses they joined.
For link builders and agencies, the test is a warning about how to report AI visibility to clients. A client who sees named by the AI on Monday and not named on Tuesday will think something broke, when nothing did. The right metric is a share over time, such as named in 3 of the last 8 checks, reported next to the map pack position, not a yes-or-no screenshot. One check proves an AI answer can appear for a keyword. It cannot prove how often, or whether a business is usually in it.
GBP SEO Tools plans to rerun the same 15 searches in December and add question-style searches such as best plumber in Austin, which trigger AI answers more often. Three signs would show the AI answer starting to replace the map: the AI appearing on the same search in most rounds, AI lists adding businesses from outside the pack more often, and AI local lists with no Call or Website buttons. Until then, the map pack is still the list that matters.
Sources: GBP SEO Tools, November 16, 2026.
Related reading on this site: November Core Update Collides With Black Friday, and a Clean Read Will Not Come Until December, Core Update Volatility Started Days Before Google Announced It, November Search Console Click Spikes May Be Bots, SEJ Reports, SEJ Pulse: Core Update, ChatGPT Business Hub and AI Mode's New Model, AI Overview Citations: What 15 Studies Agree On, and Don't, and Black Friday SEO Starts Eight Weeks Out, and November Is Already Late.
Linkbaits helps you create content that earns citations and backlinks. Start with the toolkit and build your first linkable asset this week.
Start free trial