For fifteen years, local visibility had one scoreboard: the map pack. Rank in the local three-pack and the phone rang; fall out of it and you didn't exist. The scoreboard assumption broke in 2026 — quietly, and with a number that should be pinned above every local marketer's desk.
For the same local search, 28.5% of businesses that appeared in Google's traditional Local Pack were missing from Google's AI Mode — per a Local SEO Data study of 1,120 query executions across seven service verticals and 13 US cities (May 2026). One in three-and-a-half businesses ranked where every local SEO playbook told them to rank, invisible in the AI surface Google keeps expanding. The direct answer to "does local-pack optimization still cover AI search?" is: only about 71% of it does. The rest is a new game with new selection logic.
Two Systems, Not One UI
The study's design deserves attention, because the finding is methodologically sturdy: 1,120 unique query executions across seven verticals (plumbers, dentists, HVAC, chiropractors, lawyers, restaurants, medical specialists) and 13 cities, each query run in both AI Mode and the traditional Local Pack, with bootstrap confidence intervals and multiple-comparison corrections. Key findings:
- 71.5% overlap, 28.5% gap (95% CI: 69.2–73.8%) between AI Mode and Local Pack business lists for the same query
- Top-3 overlap dropped to 48% — the businesses at the top of each system are nearly coin-flip different
- 999 unique businesses appeared in AI Mode that never appeared in the Local Pack — a parallel visibility universe
- AI Mode results shifted day to day while the structural patterns held — a volatility signature consistent with everything this series has documented about AI citation behavior
The strategic translation: a business tracking only its Local Pack rankings is blind to roughly 30% of the competitive surface, including 999 businesses per study scope that exist nowhere else.
The Selection Logic Differs — and It Reviews-Driven
Where the two systems diverge, the divergence has a pattern. Consumer-service categories: AI Mode favored businesses with substantially more reviews. AI-Mode-only businesses had 49% more reviews at the median than pack-only businesses — and among plumbers the gap reached 337% (median 546 reviews in AI Mode versus 125 in the pack).
Professional services: the pattern flips. For lawyers and medical specialists, AI Mode more often skipped business listings entirely in favor of educational content, and businesses it did show had fewer reviews on average. The answer layer reads professional-service queries as questions to answer, not vendors to list.
One more mechanism makes the two systems formally different: query wording gates visibility. When the bare keyword "divorce lawyer" returned zero AI Mode business listings across all ten cities tested, adding "near me" restored listings in 38 of 38 cases. The local answer layer is phrasing-sensitive in ways the map pack is not — which makes prompt-level testing part of local measurement, not an optional extra.
The GBP Is the New Raw Material
Why does the AI layer select differently? Because it is answering a different question. The map pack ranks; the AI layer recommends — and a recommendation needs material to recommend with. The single richest structured source for local business facts is the Google Business Profile, and industry tracking reported in 2026 found AI search results citing Google's own resources (google.com surfaces including GBP data) at more than 8× the previous rate — the answer layer is increasingly assembled from the merchant entities Google already hosts: hours, phone, categories, service areas, review summaries, booking actions.
The consequence is a role reversal that most local businesses have not internalized: review text has become recommendation content. When an AI answer says a business is "known for fast emergency response" or "praised for transparent pricing," it is frequently paraphrasing patterns extracted from review text — not counting stars. A 4.8-star business with recurring "no-show" complaints can be excluded while a 4.4-star competitor praised for punctuality is featured. Review content specificity is now a GEO input, not merely a reputation-management nicety.
The Nuance: How Big Is This Today?
Honesty requires the counterweight. Local-intent queries have been the most AI-resistant part of search: one 2026 industry analysis put AI Overview appearance on local searches at only about 7% — the map pack still owns most local intent, and Google's stated local ranking basis (relevance, distance, prominence) remains the foundation the AI layer builds on. The 28.5% invisibility finding is about the surface that is expanding, not the surface that dominates today.
The rational reading: local GEO is not an emergency; it is a compounding window. AI Mode's local behavior is early, volatile, and asymmetric — exactly the conditions under which early, disciplined operators accumulate advantages that late movers pay for. The zero-click context adds urgency: local-intent searches already end without a click at very high rates, and the AI answer layer completes the journey (call, hours, directions, book) without the website.
The Local GEO Playbook
Sequenced by evidence strength:
- GBP completeness and accuracy first. Categories matched to real demand language, complete service and product listings, current hours including holiday hours, posted updates, prompt review responses. This is the raw material the recommendation draws from.
- Review velocity and specificity. Volume matters (the 49%/337% gaps), recency matters, and content matters most: encourage customers to mention the specific service and what stood out. Generic five-star reviews give the recommendation layer nothing to quote.
- NAP consistency across every citation surface. AI systems cross-reference sources; contradictory addresses and stale phone numbers read as entity doubt — the silence mechanism from our brand-accuracy work, operating at local scale.
- Genuine locality in content. Named neighborhoods, landmarks, locally-specific FAQs (permits, climate, building codes) give the answer layer material that matches hyper-local queries; "serving the tri-county area" gives it nothing.
- Prompt-level visibility testing. Pull AI Mode results for the top 5–10 local queries plus variants ("near me," city name, "best") — the study shows phrasing determines presence. Track monthly; the results move.
- Professional-services exception. Lawyers, medical specialists, and similar verticals should weight practice-area educational content and authoritative directory citations over review-volume racing, because that is what the AI layer's selection logic currently rewards in their category.
The Honest Caveats
- One study, one market, three days. The Local SEO Data research is the best public evidence available, but it covers seven US verticals over a three-day window with daily result flux. The structural findings (separate systems, review gaps, phrasing gates) are likelier to persist than the specific percentages.
- Local trigger rates remain low. With AI Overviews appearing on only a small fraction of local queries, the ROI case is about trajectory, not current volume. Budget accordingly.
- Tracker numbers conflict at the macro level. Industry estimates of overall AI Overview prevalence vary widely by methodology (from ~48% of queries per BrightEdge's tracking to higher figures from other trackers); this piece relies on the vertical-specific study rather than contested macro numbers.
- GBP-citation concentration is an incentive risk. The more the local answer layer draws from Google's own merchant inventory, the more dependent local visibility becomes on Google's policies — a platform risk worth naming.
The Bottom Line
Local AI search is not the map pack with new paint — it is a separate selection system with its own logic: review-content-driven for consumer services, education-driven for professional services, phrasing-gated at the query level, and built substantially from the structured facts businesses maintain about themselves. The businesses it features are not the ones that outspend; they are the ones that out-structure — complete profiles, specific reviews, consistent citations, genuine locality.
That is a rare structural gift to small operators, and it has an expiration date: the 28.5% of the market currently invisible in AI Mode includes businesses that will fix this. The local answer layer is being written now, review by review and profile field by profile field. The question is only whose material it is written from.
Source and method note
This article relies on the primary sources and studies listed below. Figures are as of their study or announcement dates; claims drawn from vendor or agency self-descriptions are labeled as such, and architecture points not officially documented are marked as reported. Xindar's internal knowledge-base records informed context but are not treated as independent evidence.
- Local SEO Data, "AI Mode vs Local Pack: Where Do Local Businesses Actually Appear?" (May 2026): 1,120 query executions, 7 verticals, 13 US cities; 28.5% pack-business invisibility in AI Mode (95% CI 69.2–73.8% overlap); top-3 overlap 48%; 999 AI-Mode-only businesses; "near me" restoration 38/38; review gaps (49% median, plumbers 337%); professional-services flip; bootstrap CI and multiple-comparison corrections
- Mercury Technology Solutions industry analysis (August 2026): google.com resource citations in AI search up 8×+; GBP as structured merchant object in the answer layer (reported, not Google-confirmed)
- The Valley Marketing Group local analysis (2026): AI Overviews on ~7% of local searches; zero-click local journey
- BrightEdge (early 2026): AI Overview pushing first organic result down ~980px desktop / ~1,400px mobile; local-intent zero-click and review-sentiment analysis per industry reviews coverage
- Cross-context from this series: The Silence Tax (entity doubt mechanisms), The Trust Recession (verification behavior), A Patchwork Not a Layer (vertical trigger variance)
Percentages are as of the study window (May 2026); AI Mode local behavior changes continuously. Macro trigger-rate estimates conflict across trackers — treat vertical-level studies as the more reliable evidence.