Local AI Overviews and map packs recommended different businesses in a manual study
A 480-search study offers a reason to measure both modules. Understand its sample limits and build a branch-level audit.
A manual study published on Semrush’s blog on November 17, 2026, found that AI Overviews increasingly appeared on high-intent local searches within its sample. Author Chris Shirlow recorded 480 searches across 12 U.S. markets. An AI Overview appeared on 292 of those searches, or about 61%. In the final group, it appeared on 79 of 80 searches. Original study
The study gives local businesses a reason to inspect both AI recommendations and the familiar map pack. Its final group was captured on one day, and the author explicitly allows for a test or an incomplete rollout. A business should therefore use the findings to design its own observations while keeping the study’s geographic and temporal boundaries visible.
Read the sample before using the headline figure
The 480 searches covered 12 markets, ten services, two search formulations and two observations of each. Different market groups were observed across different intervals. The final 79-of-80 result covers four markets, while the 61% figure describes the full collection.
Within the earlier eight-market group, AI Overview coverage rose from 38% to 66%. In the later four-market group, it rose from 60% to approximately 99%. The final capture frequently showed business recommendations inside the overview with no separate map pack.
These are observations from a defined manual study. They do not establish a nationwide replacement rate or tell a particular business how often its customers encounter that layout. Retain the market group, capture date, and search formulation when using any of the figures in a client report.
Track the business identity in each module
For a local operator, the unit of interest is often a branch or service-area business. A display name alone can be ambiguous. Record the name, address or service area, website destination, and map link when available. This helps distinguish a branch omission from a spelling variation.
Give AI recommendations and map-pack recommendations separate fields. Add organic website results as a third observation. A branch may appear in more than one place, and its website may be absent even when the business is recommended.
If the answer lists a business in text but provides no direct business link, record the mention and the access path separately. Source links may support a general explanation while business cards offer a call or directions action. Capture the action the customer can actually take.
Set up a small local visibility audit
Choose a few services the business genuinely provides and two formulations customers could reasonably use. Include a named location when that matches the reader’s task. Agree on the phrases before collecting results so that later runs can repeat them.
Use a recorded location, language, device, and account state. Save the effective query as displayed by Google. For each run, capture the whole results page, the AI answer after expansion, and the map pack when present. If a module fails to load, keep that status in the log.
Record each business suggested by the overview and each business in the separate map pack. Then open the relevant links and verify the branch identity. Keep timestamps and destination URLs with the images so another reviewer can trace the entries.
This is a proposed audit procedure. It does not claim new observations of the businesses or markets in the published study.
Use overlap to describe the actual relationship
A simple illustrative example shows why a single rank is insufficient. Suppose a map pack contains branches A, B, and C, while the AI Overview recommends B, C, D, and E. Two of the three pack branches also appear in the overview. Two of the four overview recommendations appear in the pack.
Those fractions answer different questions. The first describes how much of the pack carried into the overview. The second describes how much of the overview came from the pack. Preserve both denominators if the report discusses overlap.
Now consider ten paired observations of one business. It appears in the pack six times and in the overview four times, with both appearances occurring together three times. The business appears in at least one of the two surfaces seven times. Adding six and four would double-count the three shared observations.
The correct calculation is six plus four minus three, giving seven. Label this as observed presence across the two surfaces. It remains a count of captured results, with no customer exposure or click denominator attached.
Investigate an omission through the customer task
If a business repeatedly appears in the pack but is missing from the overview, inspect the answer’s stated requirements. Does the query ask for emergency availability, a specific service, or a geographic boundary? Check whether the business’s public information accurately communicates those facts.
Review branch pages, published hours, service descriptions, and the destinations linked from business profiles. Correct discrepancies you can verify. Keep the edit date and repeat the same test afterward, while recognizing that a before-and-after difference can have several causes.
A credible audit also identifies questions the business cannot satisfy. If it does not provide the requested service, the appropriate outcome may be an accurate exclusion. Do not rewrite a page to imply capabilities the business lacks.
Connect the audit to business outcomes
A recorded recommendation can lead to a website visit, a call, directions, or no measurable action. Use available business reporting to inspect each outcome with its own definitions. Explain any missing attribution before attaching revenue claims to AI visibility.
The Ask Maps audit adds checks for branches, hours, and bookings. The SERP testing guide supports repeatable captures, and the AI visibility crosswalk keeps appearances distinct from referrals.
For the next reporting cycle, collect a stable small sample and retain the separate business inventories. That record can show whether the two modules recommend the same providers and which links or actions a customer is offered.
Keep learning
Continue this topic
Audit citations
Locate the failing answer layer
Compare evidence
Separate mentions, citations and visits
Measure referrals
Check AI search visits in GA4
SEO
Ask a question or join the discussion