Short answer: this was not a vibe check. We asked several AI models the same set of real buyer questions, over and over, and compared the anonymized business against local competitors. The result: the business showed up in only a low double-digit share of answers, while the strongest competitors showed up several times more often.
Methodology: how this local AI visibility audit was measured
The audit used a fixed set of local buyer questions, asked across multiple AI providers, with repeated runs per question. We counted mentions, recommendations, citations, competitor appearances, wrong-business mix-ups, and cited source sites separately. That separation matters: a shop can be mentioned, recommended, cited, or described incorrectly, and each outcome points to a different fix.
- Prompt categories: non-brand local recommendations, service-specific questions, competitor alternatives, and brand-direct checks.
- Provider split: Claude, ChatGPT, Google AI Overview, and Perplexity were kept separate instead of blended into one opaque score.
- Decision rule: repeated appearances and recurring citations counted more than one lucky answer.
The uncomfortable finding: AI did not lack options
When someone asks an AI assistant for a reliable local service provider, the model has to turn messy local evidence into a short list. In this audit, that short list was dominated by competitors. Those competitors simply had a clearer online identity and stronger, more citable pages.
Across the answer set, several established local competitors appeared more reliably. The anonymized business was not necessarily worse in the real world. It was just harder for AI systems to confidently understand, verify, and cite.
| Model | Anonymized business visibility | What it means |
|---|---|---|
| Claude | Meaningful but inconsistent | Claude found enough traces to mention the shop sometimes, but not enough to make it a consistent leader. |
| ChatGPT | Occasional | ChatGPT occasionally surfaced the shop, but competitors had stronger source support and entity clarity. |
| Google AI Overview | No observed inclusion | The business did not break into Google’s AI answer layer for the tested local prompts. |
| Perplexity | No observed inclusion | In a citation-forward engine, the shop lacked the source mix needed to be selected and cited. |
Case 1: the model split showed exactly where the problem lived
The most useful part of the audit was not the overall score. It was the split by engine. The anonymized business appeared inconsistently in some assistants. But it had no observed inclusion in engines that lean heavily on citations. That pattern is diagnostic.
Claude and ChatGPT can sometimes infer a business from looser web context. They may recognize fragments of a brand even when its online identity is messy. Google AI Overview and Perplexity are less forgiving for this kind of local-service question. They need clear, indexed, well-sourced information before they will put a shop into the recommendation set.
| Observed pattern | What it usually means | Why it matters commercially |
|---|---|---|
| Some visibility in Claude / ChatGPT | The brand is not invisible. The model can understand it when enough context is available. | There is recoverable demand; the business is not starting from zero. |
| 0% in Google AI Overview / Perplexity | The source graph is too weak or too ambiguous for citation-driven engines. | High-intent buyers using search-like AI experiences never see the shop. |
| Competitors appear repeatedly | The market has enough evidence for AI to make recommendations — just not enough for the anonymized business. | The gap is not category demand. It is source eligibility and trust. |
This is why “we showed up once in ChatGPT” is not a visibility strategy. The question is whether the business appears consistently across the engines that behave like local search, answer engines, and citation engines.
The citation landscape explained the rankings
The answer set contained hundreds of citations. The source mix showed what AI systems were using:
- A large share of citations went to business-owned websites.
- Some citations went to social or community sources.
- Others went to reviews and directories.
That split matters. Local businesses often obsess over review count alone. But this audit showed that AI assistants leaned heavily on the business's own website to decide:
- What the shop actually does
- Where it serves customers
- Whether it handles insurance claims
- Which repair categories it should be associated with
Case 2: competitors won because they owned the sources AI was already citing
The audit found that one leading competitor was cited strongly across the engines that cite sources. Its edge was not just a better homepage. It had a stronger network of mentions across niche directories, review platforms, social sources, and community references.
Another competitor showed a different lesson: one strong industry listing can become the backbone for one model's recommendations in a local market.
| Source type | Where it mattered in the audit | What the anonymized business lacked |
|---|---|---|
| Industry directory | Backbone for ChatGPT citations; also appeared in Google AI Overview source patterns. | No optimized Industry directory profile acting as a citation anchor. |
| Review platform | Frequently cited source type in the dataset. | Reputation signals were fragmented instead of concentrated under one canonical identity. |
| Reddit / Facebook groups | Recurring sources for Perplexity and Google AI Overview in local recommendation answers. | Little community corroboration that AI could cite or summarize. |
| Niche directories | Niche directories reinforced competitor category confidence. | Thin or inconsistent niche-directory scaffolding. |
The lesson is not “submit to every directory.” The lesson is to identify which third-party sources the engines already cite in the category, then fix and strengthen those sources first. For the audited category, the most-cited directories and review platforms were not generic SEO chores — they were AI visibility infrastructure.
Each AI model had a different source diet
The audit also showed why a business can appear in one AI product and disappear in another. The systems did not use identical evidence.
- Claude was more willing to synthesize from broad web evidence, which is why the anonymized business appeared there more often.
- ChatGPT was more selective in the tested prompt set, producing weaker but non-zero visibility.
- Google AI Overview behaved like a local-search layer: if the web and local entity signals were not clear enough, the shop was skipped.
- Perplexity demanded citation-ready sources; when it could not ground the recommendation, it cited competitors instead.
This is why a single AI search is misleading. One prompt in one model may make visibility look fine. Repeated prompts across providers reveal whether the business is consistently understood.
The core problems were fixable, not mysterious
1. Entity fragmentation
AI systems need to know that a business name, website, local listings, reviews, phone number, and service area all refer to the same business. In the audited case, the public footprint created too much confusion. The problems included:
- Multiple active business names
- Inconsistent listing details
- No declared links tying its profiles together (a "sameAs" graph)
- Fragmented structured-data IDs
Competitors had a cleaner, more consistent identity across their sites, listings, and review pages.
Why it matters: before an AI assistant decides whether to recommend a shop, it first has to work out what the shop is. If that identity is unstable, the assistant either skips the business or borrows facts from nearby competitors.
2. Missing or weak schema
For a local service business, structured data (schema) should make the business easy for machines to read. That means:
- Local business schema markup
- Consistent name, address, and phone number
- Geo and service-area details
- Opening hours
- Links tying its profiles together (sameAs links)
- Services listed clearly
- Reviews, where compliant
- FAQ content for common buyer questions
3. Authority and indexation gaps
AI assistants favored sources they could retrieve and trust. The report found very low authority signals and partial indexing: many pages existed, but only a subset were visible enough for search systems to use. Important pages were not indexed, service pages were shallow, and third-party references were sparse. Because of this, the business was present in the real world but underrepresented in what AI systems could actually find.
Why it matters: AI cannot cite a page it cannot find. And if the only retrievable pages are thin or ambiguous, the model will cite a clearer competitor page instead.
4. Weak answer assets
Competitors won because they were easier to cite for specific buyer questions, such as:
- "Best provider near me"
- "Approved service provider"
- "Specialty repair"
- "Cost questions"
- "How long does the service take?"
The anonymized business needed pages that answer those questions directly, not just a generic services page.
Case 3: entity confusion created hallucinations, not just lower rankings
The most dangerous finding was not omission. It was incorrect visibility. In the report, inconsistent identity signals caused AI systems to fill gaps with adjacent businesses and invented details.
- Inconsistent public naming made it unclear whether related listings described one entity or several.
- Inconsistent contact and location details weakened confidence in how to reach the business.
- Conflicting hours and address formats weakened local entity confidence.
- Missing sameAs links failed to connect major profiles and the website into one machine-readable brand.
- Fragmented structured-data identifiers split the site’s own entity signals.
When asked directly by name, the audited business was sometimes named back, but the answers were not reliably clean. Some engines returned no answer. Others mixed the business up with unrelated, similarly named businesses. At the citation level, a noticeable share of the links in this direct-name test pointed to competitors or unrelated businesses.
That is the expert distinction: a business can be “visible” and still lose demand if AI describes it incorrectly. Wrong phone numbers, invented attributes, missing service lines, or competitor leakage inside branded answers are conversion problems, not vanity-metric problems.
How to diagnose the same problem in your local market
- Ask 10 non-brand buyer prompts across ChatGPT, Perplexity, Gemini or Claude, and Google AI experiences.
- Record mention, recommendation, citation URL, sentiment, and whether facts such as phone, address, hours, and service area are correct.
- List every cited source domain, then classify it as owned site, Google/GBP, review platform, niche directory, social/community, or competitor-owned.
- Compare name, address, phone, website, categories, and hours across the business profile, website, review platforms, social profiles, and niche listings.
- Flag any answer that mixes the business with a similarly named shop. That is not harmless hallucination; it is lost demand.
The concrete fix queue
- Build an entity cleanup map. Normalize name, phone, address, website, social profiles, and major directory listings. Remove stale variants where possible.
- Add local business and service schema. Use category-appropriate structured data, sameAs links, service area, hours, and clear service offerings.
- Create citable service pages. Separate pages for priority services, pricing or estimate questions, process details, warranty or policy questions, and location-specific intent.
- Publish buyer-question content. Turn real customer questions into FAQ and guide pages that answer how repair estimates, supplements, insurance approvals, timelines, warranties, and rentals work.
- Strengthen third-party proof. Improve and reconcile Review platform, Google Business Profile, BBB or industry profiles, social/community mentions, and local citations. Do not spam forums; make the business easy to verify.
- Improve indexation. Submit updated sitemaps, inspect important URLs, fix blocked/noindexed pages, and make sure key pages have internal links from the homepage and navigation.
- Re-scan the same prompt set. Use the original prompt set, model mix, and repeat count so movement is measured against the same baseline.
What success should look like
The first milestone is not "rank #1 in ChatGPT." It is to move from low, inconsistent visibility to reliable inclusion. In practice, that means:
- More mentions
- More recommendations
- Better sentiment
- More citations from pages the business controls or has deliberately improved
For this audit, the benchmark was clear: the anonymized business was far behind the local leaders. That gap is large, but it is specific. The path forward is not generic SEO advice. It comes down to four things: make the business's identity cleaner, make the website easier to answer questions from, make the schema more explicit, and make the third-party evidence easier for AI systems to trust.