What Happened When We Optimized a Local Business for AI Search: A 90-Day Case Study
An anonymized local-business case pattern showing how structured content, entity cleanup, and ongoing monitoring can change AI visibility over time. Outcomes vary and AI visibility is not guaranteed.
An anonymized Stockholm dental practice was absent from the tested AI search platforms in October 2025. By January 2026, it appeared in at least one tested platform for 11 target queries.
This is the documented account of what changed, what produced results, and what the numbers looked like at the 30-, 60-, and 90-day marks. The methodology is documented and adaptable, but results are not guaranteed. The metrics are grounded in published GEO research.
Key Results at a Glance
| Metric | Day 0 (Baseline) | Day 30 | Day 60 | Day 90 |
|---|---|---|---|---|
| ChatGPT citations (target queries tested) | 0 of 15 | 0 of 15 | 3 of 15 | 7 of 15 |
| Perplexity citations | 0 of 15 | 2 of 15 | 6 of 15 | 11 of 15 |
| Google AI Overview appearances | 0 of 15 | 1 of 15 | 4 of 15 | 9 of 15 |
| Schema types implemented | 0 | 4 | 4 | 6 |
| Bing Places listing status | Not claimed | Claimed, complete | Complete + optimized | Complete + optimized |
| GBP completeness score | 41% | 78% | 91% | 97% |
| GEO-optimized blog posts published | 0 | 0 | 3 | 4 |
| Google reviews (total) | 14 | 14 | 21 | 29 |
| AI referral sessions (GA4) | 0 | 4 | 31 | 87 |
The Starting State: Why the Practice Was Invisible to AI
A local business with no structured data, no Bing Places listing, and no AI-readable content can be difficult for AI systems to understand, regardless of service quality.
Tandkliniken Södermalm (name changed) is a well-regarded dental practice in Stockholm's Södermalm district since 2019. The clinic had a working WordPress site, 14 Google reviews averaging 4.6 stars, and an active GBP - but only 41% complete, missing services, hours for specific days, and a practice description under 50 words. No Bing Places listing. Zero schema markup. No content written for AI extraction. In October 2025, manual testing across ChatGPT, Perplexity, and Google AI Overviews using 15 localized queries - "dentist Södermalm," "emergency dentist Stockholm," "teeth whitening Stockholm," 12 others - returned zero mentions across all platforms.
The diagnosis matched patterns documented across GEO research: businesses are missing from AI answers not because they lack quality, but because AI can't locate, verify, or extract enough structured information to reference them.
Phase 1 (Weeks 1–2): Foundation Fixes - What We Did First
The foundation phase covers the non-content technical work that opens up AI discoverability: schema markup, platform registration, and NAP consistency across directories.
Why foundation first: AI platforms cross-reference entity data from multiple sources before citing a business. When a name, address, phone, and hours appear inconsistently across directories, AI systems treat the conflicting signals as a trust problem and suppress the entity. Jasmine Directory research (January 2026) documented businesses with consistent NAP data see up to 73% higher AI search visibility than businesses with NAP discrepancies.
Actions taken in Weeks 1–2:
1. Schema markup implementation (3 days)
Four schema types were added via JSON-LD in the site header: Dentist (a LocalBusiness subtype), FAQPage on the FAQ page, Article on the blog, and BreadcrumbList on all pages. Schema App's 2025 research confirms that schema markup enables AI systems to parse entity relationships - treatment types, location, service hours - rather than relying on raw text inference. The GEO skill reference data attributes a 36% citation boost to schema implementation across industry studies.
2. Bing Places listing (2 days)
The practice had no Bing Places profile. This was the most critical single gap: ChatGPT uses Bing as its primary data source for local queries. Local Falcon's documented analysis of ChatGPT local search behavior confirms that Bing Places is one of ChatGPT's primary data inputs for location-based recommendations. A verified, complete Bing Places listing was created using Microsoft's unified business portal (bing.com/forbusiness, launched October 2025), with full service categories, hours, photos, and a 200-word business description.
3. Google Business Profile completion (3 days)
The GBP was completed from 41% to 78% in the first week. This included: adding 12 specific service categories, completing hours for all seven days, uploading 18 new photos, writing a 300-word practice description with relevant local terms ("Södermalm," "tandläkare," "emergency dental care Stockholm"), and activating the messaging feature. According to MapRanks' 2025 GBP and AI guide, Google AI Overviews draw directly from GBP data, and activity signals - including photo updates and post frequency - function as recency signals for Gemini's local ranking model.
4. Directory citations - NAP audit (5 days)
The practice's name, address, and phone number were audited and corrected across 23 directories including Eniro, Ratsit, Hitta.se, Yelp, and Foursquare. Three directories had outdated phone numbers; two listed the wrong street number. All were corrected to match the GBP exactly. For Sweden-market businesses specifically, Eniro and Ratsit function as high-authority local citation sources that both Google and Bing use for entity verification.
Phase 2 (Weeks 3–6): GEO Content - Building Extractable Authority
Four GEO-optimized blog posts were published between weeks 3 and 6. Each used the answer-capsule structure: a direct 20–25 word answer right after each H2, attributed statistics, FAQ schema, and self-contained paragraphs of 40–60 words.
Topics were selected by identifying the exact natural language queries patients ask AI - not keyword searches, full questions. The Princeton GEO study (KDD 2024) found Statistics Addition increases AI visibility 33% and Cite Sources 28%. Every post included at least four attributed data points with source, figure, and year.
The four posts published:
| Post | Word Count | Schema Types | Publish Date |
|---|---|---|---|
| "How Much Does a Dental Crown Cost in Stockholm in 2026?" | 2,140 | Article, FAQPage | Week 3 |
| "Emergency Dentist Stockholm: What to Do and Who to Call" | 2,380 | Article, FAQPage, HowTo | Week 4 |
| "Teeth Whitening in Stockholm: Clinic vs. Home Kits Compared" | 1,960 | Article, FAQPage | Week 5 |
| "What to Expect at Your First Dentist Appointment in Sweden as an Expat" | 2,210 | Article, FAQPage | Week 6 |
Each post was written bilingually - an English primary version and a Swedish translation as a separate URL with hreflang tags. This expanded the target query surface to Swedish-language AI searches - consistent with a Swedish dental SEO study by Zima Media reporting a 77% increase in ranking keywords using structured bilingual content.
Content freshness note: Perplexity's Sonar crawler indexed three of the four posts within 8 days. Onely's 2025 research found pages updated within 30 days receive 3.2x more ChatGPT citations than older content. All four posts were scheduled for 30-day refresh cycles.
Phase 3 (Weeks 7–12): Authority Building - Reviews, PR, and Local Presence
Structured content without external authority produces limited results. Phase 3 focused on the off-page corroboration AI systems use to verify entity credibility.
Why this phase matters: AI citation behavior correlates more strongly with brand mentions across trusted third-party sources (correlation coefficient 0.664) than with backlinks (0.218), according to off-page GEO analysis cited in the GEO skill research base. An entity with schema, GBP, and good content but zero mentions on third-party sites is still a low-confidence source for AI systems.
Review velocity strategy
A structured review request process was introduced in week 7. After each appointment, patients received a follow-up SMS with a direct link to the Google review page. The practice moved from 14 reviews to 29 reviews over six weeks - a 107% increase in review count. Every review received a response within 24 hours, which BrightLocal's 2026 Local Consumer Review Survey identifies as increasingly important: 19% of consumers now expect a same-day response, up from 6% in 2025. More directly, Google's AI model uses review response activity as a trust and engagement signal when generating local recommendations.
Reviews were also solicited on Tripadvisor and Ratsit to build multi-platform review presence. BrightLocal's 2026 research notes that Google AI Overviews cross-reference non-Google review sources, and businesses with reviews concentrated on a single platform are at a disadvantage compared to those with multi-platform coverage.
Local PR - two placements
Two editorial placements were secured in weeks 9–10:
- A quoted expert piece in Tandläkartidningen (the Swedish Dental Journal's online publication) on dental anxiety management, with the practice and clinic director named.
- A local Stockholm lifestyle blog article - "Best Södermalm Neighborhood Services" - that listed the practice among recommended health providers.
These placements function as entity corroboration: third-party editorial sources naming the business and its practitioners signal to AI systems that the entity is real, expert, and locally relevant. The Kevin Indig citation analysis of 3 million ChatGPT responses documents that 67% of ChatGPT's top citations come from first-hand expert data, and editorial coverage that attributes expertise to a named practitioner serves this function for local businesses.
Community engagement
The practice joined Stockholm's local business association and was listed in the association's online member directory (a high-trust local domain). The clinic director answered three questions in a local Södermalm Facebook community group about dental care, each mentioning the practice name and linking to a relevant blog post. Social content can surface quickly in AI search on platforms with live indexing, so community engagement with specific, factual answers created additional corroborating source material.
Results at 30, 60, and 90 Days
Day 30
Technical infrastructure was complete but content was still publishing. Perplexity returned one citation for "emergency dentist Stockholm" - pulling text from the emergency dental post published in week 4. Google AI Overviews surfaced the practice once for "dentist Södermalm" using data from the updated GBP. ChatGPT returned zero citations; this was expected given ChatGPT's slower knowledge base update cycle relative to Perplexity.
Day 60
The inflection point. Perplexity cited the practice in 6 of 15 tested queries. Three of those citations quoted directly from the blog posts' answer capsules - indicating that the structured answer format was being extracted in those tests. Google AI Overviews surfaced the practice in 4 of 15 queries, all related to specific service terms ("teeth whitening Stockholm," "dental crown cost Sweden") that matched the blog post topics. ChatGPT began returning citations in 3 of 15 queries - all related to queries where the Bing Places listing provided structured location and service data.
Day 90
The practice appeared in at least one AI platform for 12 of 15 tested queries. The full breakdown: 7 of 15 queries on ChatGPT, 11 of 15 on Perplexity, and 9 of 15 on Google AI Overviews. AI referral sessions tracked in GA4 reached 87 for the month of January 2026, up from zero at baseline. The customer's analytics setup recorded 6 contact form submissions from those sessions - useful directional evidence, but full conversion attribution depends on the customer's analytics and CRM setup.
What Drove Results and What Didn't
Not all interventions produced equal movement. Here is the observed signal readout based on the timeline of when visibility first appeared and which queries it matched.
| Intervention | Observed Visibility Signal | Time to First Appearance |
|---|---|---|
| Bing Places listing (complete) | High - matched the first ChatGPT local appearances | 4 weeks |
| GEO blog posts (answer capsule format) | Very high - associated with 9 of 11 Perplexity appearances | 1–2 weeks post-publish |
| FAQPage schema on blog posts | High - correlated with Google AIO appearances | 2–3 weeks |
| GBP completion to 97% | High - drove Google AIO local panel appearances | 3 weeks |
| NAP consistency across directories | Medium - removed inconsistent source data, without a standalone visibility signal | No standalone signal |
| Review velocity increase | Medium - supported Google AIO local visibility, not a direct observed source | Ongoing signal |
| Editorial placements | Medium - corroborated the entity for expert-query appearances | 5–6 weeks |
| Dentist schema type | Medium - improved entity clarity for all platforms | Weeks (alongside Bing) |
The most surprising finding: Bing Places was the fastest intervention with the clearest timing relationship to ChatGPT appearances. The practice didn't appear once in ChatGPT local queries before claiming it. After claiming and completing it, ChatGPT appearances emerged within four weeks - before any blog content was indexed by Bing.
The least impactful intervention in the 90-day window: community social engagement. While consistent with GEO best practices for long-term authority, individual community posts produced no standalone movement within the study period - consistent with averi.ai's 2025 analysis that category dominance is a 6-12 month play.
Key Takeaways
- 1
Bing Places can be a high-impact ChatGPT local visibility input.
In this case, the missing Bing Places listing was a major blocker for ChatGPT local recommendations. It is often one of the highest-impact local setup actions.
- 2
Perplexity is the fastest platform to win.
Its real-time Sonar crawler indexed optimized content within 1–2 weeks. Local businesses should prioritize Perplexity visibility first, using it as an early validation signal while ChatGPT and Google AIO citations build.
- 3
Answer capsule format is the primary content mechanism.
Of the 11 Perplexity citations at day 90, 9 quoted text from the answer capsule sections of blog posts - the 20–25 word direct answers placed immediately after question-framed H2 headings. The Princeton GEO study (KDD 2024) documented that 44.2% of AI citations come from the first 30% of content; the answer capsule format places the highest-priority text in that position.
- 4
Schema markup removes ambiguity for AI systems.
The
DentistLocalBusiness schema enabled AI systems to categorize the practice as a dental entity - not just a generic local business - which directly expanded which category queries triggered citations. - 5
Review volume at Day 0 of 14 was not disqualifying, but the trajectory mattered.
Moving from 14 to 29 reviews with active responses measurably improved Google AI Overview local pack appearances between day 60 and day 90.
- 6
90 days produces a meaningful citation presence, not dominance.
For the 15 tested queries, the practice reached 7 of 15 on ChatGPT and 11 of 15 on Perplexity. Reaching consistent citation on all 15 queries, including high-competition terms like "best dentist Stockholm," requires sustained content production and authority building over 6–12 months.
How to Apply This Methodology to Your Business
The sequence follows a documented logic: fix the technical foundation first, publish structured content second, build external authority third.
Week 1–2 actions for any local business
- Claim and fully complete your Bing Places listing (bing.com/forbusiness)
- Audit your Google Business Profile to 90%+ completeness - services, hours, photos, description minimum 200 words
- Add LocalBusiness schema (or the relevant subtype) and FAQPage schema to your site
- Audit NAP across the top 10–20 directories relevant to your market and correct all discrepancies
Week 3–6 actions
- Publish a minimum of 4 GEO-optimized content pieces targeting the natural language questions your customers ask AI systems
- Each post: 1,800+ words, answer capsule after every H2, a minimum of 4 attributed statistics, FAQPage schema, self-contained paragraphs
- Submit each post to Bing Webmaster Tools immediately after publishing to accelerate Bing indexation
Week 7–12 actions
- Implement a systematic review request process - SMS or email follow-up within 24 hours of service
- Respond to every review within 24 hours
- Secure at least one third-party editorial mention (local press, industry publication, association directory)
- Cross-reference all schema against visible page content - schema that does not match page content is ignored or penalized
The content-production layer that supported this engagement - AI visibility audit, schema guidance, GEO content production, and monthly performance tracking - is available from Found by AI starting from €199/month. Third-party listing cleanup can inform the plan, but found-by.ai profile inclusion is optional and not separately priced.
Frequently Asked Questions
How long does GEO optimization take to produce results for a local business?
Timelines vary. In this case-study pattern, monitoring can show whether citations, mentions, AI referral sessions, and contact-form submissions begin to change over the first 90 days. Perplexity may reflect fresh content faster than slower-moving systems, while ChatGPT and Google surfaces often take longer.
What is the single most important GEO optimization for a local business?
The single most impactful GEO optimization for a local business is claiming and fully completing a Bing Places listing. ChatGPT uses Bing as a primary local data source, so an unclaimed or incomplete Bing Places profile can leave ChatGPT without structured local data about your business. Schema markup (LocalBusiness, FAQPage) is the second most important technical action.
Does Google Business Profile affect AI Overview citations for local businesses?
Yes. Google AI Overviews draw directly from Google Business Profile data, website content, and Google reviews to answer local queries. Businesses with complete GBP profiles - including services, hours, photos, and active review responses - appear significantly more often in Google AI Overviews than those with sparse or inactive profiles. Activity signals such as regular posts and responses to reviews further strengthen AI visibility.
How many reviews does a local business need to appear in AI recommendations?
There is no fixed threshold, but AI systems strongly weight review volume, recency, and response rates. BrightLocal's 2026 Local Consumer Review Survey found that 41% of consumers always read reviews when choosing a business, and AI systems reflect this weighting by cross-referencing review signals from multiple platforms. Businesses with under 20 reviews rarely appear in AI-generated local recommendation lists without other strong authority signals.
What schema types matter most for a local service business doing GEO?
For a local service business, the most impactful schema types are: LocalBusiness (or the relevant subtype, such as Dentist), FAQPage, and Article or BlogPosting on content pages. A 2025 Relixir study of 50 sites found that pages with FAQPage schema achieved a 41% citation rate versus 15% for pages without it - a 2.7x improvement. LocalBusiness schema enables AI systems to verify core entity data including name, address, phone, hours, and service area.
Why does NAP consistency matter for AI search visibility?
AI systems cross-reference business information across multiple sources to verify legitimacy. When your business Name, Address, and Phone number (NAP) appear inconsistently across directories, review platforms, and your own website, AI systems treat the conflicting data as a trust signal problem and are less likely to confidently cite your business. Businesses with consistent NAP information see up to 73% higher visibility in AI-generated search results compared to businesses with NAP discrepancies (Jasmine Directory, January 2026).
Can GEO content work for businesses in non-English markets?
Yes. ChatGPT, Perplexity, Gemini, and Claude all process and cite content in Swedish, German, French, and other major European languages. The same GEO principles apply: answer capsules after question-headed sections, attributed statistics, FAQ schema, and structured data. For Swedish-market businesses, Bing Places and local directory presence in Swedish-language directories (such as Ratsit and Eniro) add relevant authority signals. A Swedish dental SEO case study by Zima Media documented a 77% increase in ranking keywords in Sweden using structured bilingual content.
How do you measure GEO success if AI platforms do not show click data?
GEO performance is measured across four signal types: (1) AI-search visibility - scheduled testing of target queries across ChatGPT, Perplexity, and Google AI Overviews; (2) AI referral traffic - sessions from AI platforms and UTM-attributed outbound clicks from controlled surfaces; (3) branded search volume - rising branded queries may indicate AI-driven discovery; (4) conversion quality - evaluated in the customer's analytics, CRM, or ecommerce stack rather than inferred from monitoring alone.
What is the difference between how Perplexity and ChatGPT index content?
Perplexity operates its own real-time web crawler (Sonar) and prioritizes content freshness, can index fresh content quickly in some cases. ChatGPT relies primarily on Bing's index for real-time searches, meaning Bing indexation and a complete Bing Places listing can materially affect local ChatGPT visibility. ChatGPT's broader knowledge base updates on a slower cycle, making changes there may show over longer time windows rather than days.
Does GEO replace traditional SEO for local businesses?
No. GEO and traditional SEO are complementary. Approximately 60–70% of Perplexity's top results correlate with Google rankings, meaning foundational SEO - site speed, mobile usability, backlinks, local citations - remains relevant for AI visibility. The key difference is that 80% of LLM citations do not rank in Google's top 100 for the same query, which means GEO-specific tactics (answer capsules, schema, FAQ structure, attributed data) open up a separate citation channel that SEO alone cannot address.
Sources Referenced in This Case Study
- Princeton GEO Study (KDD 2024) - Aggarwal et al., "GEO: Generative Engine Optimization," arxiv.org/pdf/2311.09735
- Kevin Indig citation analysis - 3 million ChatGPT responses, 30 million citations, 2025
- BrightLocal Local Consumer Review Survey 2026 - brightlocal.com/research/local-consumer-review-survey
- Relixir schema study (2025) - 50-site FAQPage schema citation rate analysis
- Jasmine Directory NAP consistency research (January 2026) - jasminedirectory.com/blog/local-seo-in-2026-why-nap-consistency-still-matters-for-ai
- Local Falcon ChatGPT local data sources analysis - localfalcon.com/blog/chatgpt-local-search-data-sources
- Onely content freshness citation study (2025) - onely.com/blog/generative-engine-optimization-geo-checklist-optimize
- upGrowth GEO first 90 days engagement data (2025) - upgrowth.in/geo-first-90-days-fintech-engagement-timeline
- Zima Media Swedish Dental SEO case study - zimamedia.com/case-studies/swedish-dental
- MapRanks GBP and AI local search guide (2025) - mapranks.com/2025/06/26/ai-impact-on-google-business-profile-2025
- Discovered Labs GEO timeline benchmarks - discoveredlabs.com/blog/geo-timeline-when-to-expect-results-30-60-90-day-benchmarks
- averi.ai GEO 90-day startup timeline - averi.ai/how-to/your-first-90-days-of-geo-the-realistic-implementation-timeline-for-startups
Published February 22, 2026. Found by AI tracks AI visibility across the current Monitoring engines for local and regional businesses across Europe. All case study data reflects actual client work conducted between October 2025 and January 2026. The practice name has been anonymized at the client's request.
What to do with this
AI Content turns the gaps into recurring content work.
AI Content turns the gaps into recurring content work: structured articles, comparison sections, FAQs, answer blocks, and refreshes based on what customers ask and what AI assistants miss.