AI Search Visibility for Local Businesses: What 1,814 Answers Show
A 32-day test of 1,814 AI answers for one Vancouver-area electrician found that major answer engines agreed on a recommended business only 9% of the time, making repeated measurement more useful than one-off screenshots.
· 14 min read
A 32-day local-business test recorded 1,814 AI answers across ChatGPT, Gemini, Google AI Mode, Google AI Overviews and Perplexity. On the same morning, those systems named at least one business in common on only 37 of 416 question-days (9%)—a result that changes how AI search visibility for local businesses should be measured and reported.
The practical payoff is straightforward: a screenshot showing that a business appeared in ChatGPT, Gemini or Google AI Mode on one day is weak evidence. A useful local AI visibility report needs repeated prompts, controlled location settings, platform-by-platform rates, and a record of the sources cited beside each answer.
The headline finding: five systems agreed only about one day in five
The test came from a Reddit post by an SEO who tracked one electrician serving the Vancouver area. Each morning from September 7 through October 8, the researcher asked 16 local buying-intent questions, such as “best electrician in Burnaby BC” and “who can install an EV charger at my house in Vancouver.”
There were 416 question-days in the agreement analysis. “Agreement” used a generous definition: every system that returned an answer named at least one business in common, or every system named no business at all.
| Same-morning result | Question-days | Share |
|---|---|---|
| At least one business named by every answering system | 37 | 9% |
| All systems named no business | 52 | 13% |
| No agreement | 327 | 79% |
Combined, the two agreement states total 89 question-days, or roughly 21%—about one in five. But only the first category represents shared business visibility. The other 13% means every system declined to name a business, which is not evidence that any local brand was recommended.
That distinction matters for agencies and site owners. “The systems agreed” can sound like broad endorsement when, in this data set, agreement frequently meant that none of them suggested a provider.
What the 1,814-answer local-business test measured
The study design was more useful than a casual prompt check because it controlled several variables:
- One real business: an electrician in the Vancouver area.
- 16 purchase-oriented prompts: including service-plus-location searches and recommendation-style questions.
- Five answer surfaces: ChatGPT, Google Gemini, Google AI Mode, Google AI Overviews and Perplexity.
- A fixed daily cadence: prompts were run every morning for 32 days.
- A local setting: queries were made while located in the business’s city.
- Logged-out sessions: intended to reduce account-history effects.
- Manual entity cleanup: spelling variants of a company name were merged before scoring.
Perplexity was included only for the final approximately two weeks, so not every question-day had five systems available. The author reported 201 question-days where all five systems answered. Across those complete five-platform observations, there was not one instance in which all five named a business in common.
This is a limited sample, not a universal benchmark. It covers one trade, one metro area and one testing period. Still, it is a more transparent data point than a single “we rank in ChatGPT” claim because the prompt set, geography, answer engines and scoring rule are disclosed.
AI search visibility for local businesses requires four separate metrics
A business can appear in an AI answer without being recommended, cited, ranked first, or described accurately. Conflating those outcomes makes reporting unreliable.
1. Mention
A mention means the answer includes the business name anywhere. For example, an assistant might list three Vancouver electricians and place the tracked company third. That is visibility, but not necessarily preference.
2. Recommendation
A recommendation means the assistant presents the business as a suitable choice, such as “a good option for EV charger installation.” Recommendation language is stronger than a plain list inclusion, but it can still vary from one run to the next.
3. Rank or position
Where an answer lists businesses in an ordered sequence, the first listed name is not automatically an objective “number-one ranking.” The model may organize its response by relevance to wording, service fit, review information, distance, or the pages it retrieved that day. A measurement framework should record position, but should not overstate what position proves.
4. Citation and factual consistency
A citation identifies a page or source used to ground an answer. ChatGPT Search can use web search and show source links; Perplexity similarly presents cited sources; Microsoft says Copilot web answers can include citations. (help.openai.com) A business may be named without its own website being cited, or cited without being recommended.
Factual consistency is separate again. A local business might be named with an incorrect service area, old opening hours, inaccurate phone number, or a review count that no longer matches. A practical audit should log these errors rather than treating every mention as a success.
The test found disagreement between platforms and within platforms
The strongest result was not simply that systems differed from one another. Each system was also unstable over time.
The Reddit analysis found that same-day overlap between pairs ranged from 40% for AI Mode versus Perplexity to 63% for AI Mode versus AI Overviews. That higher AI Mode/AI Overviews overlap is unsurprising as a directional observation: Google describes both as AI-powered Search experiences, and says AI Overviews work with its existing Search quality and ranking systems and Knowledge Graph. (search.google) It does not mean the two surfaces return identical local-business recommendations.
More notably, an individual assistant’s answer shared no business with its own answer from the previous day on 29% to 37% of days. The tracked electrician itself flipped between named and not named on 9% to 17% of consecutive days, despite no reported changes to the company website.
| Measurement | Reported result | Interpretation |
|---|---|---|
| All systems naming a common business | 9% of 416 question-days | Cross-platform consensus was rare |
| All systems naming no business | 13% | “Agreement” can still mean no local visibility |
| Five-system common business, when all five answered | 0 of 201 question-days | Full consensus did not occur in this sample |
| Pairwise overlap | 40% to 63% | Similar products can still retrieve different business sets |
| Day-to-day self-disagreement | 29% to 37% | One platform’s daily answer was not stable |
| Tracked-business mention flip rate | 9% to 17% | A brand can appear and disappear without site changes |
Why AI answers can change even when the business website does not
The author did not track competitors’ Google Business Profile edits or review activity, so the data cannot establish why individual flips occurred. One commenter suggested testing whether competitor review changes or profile updates correlate with answer changes; that remains a sensible follow-up rather than a conclusion from this study.
The source-level observations are more revealing. According to the post, ChatGPT’s changes often followed the pages it appeared to use that day: when it drew on a list or directory including the electrician, the business was more likely to appear; when it drew on different pages, it could disappear.
That pattern is plausible because answer engines retrieve and synthesize web material rather than operating as a fixed local pack. Google says AI Overviews provide a snapshot with links to explore the web, while AI Mode supplies answers with links from the web. (search.google) ChatGPT Search, Perplexity and Microsoft Copilot also describe web-grounded answers with source links or citations. (openai.com)
The implication is not that a business can control daily AI outputs by adding a page. It is that local AI visibility is partly a retrieval observation: the answer can vary as the system retrieves different public sources, interprets the prompt differently, or decides not to name businesses at all.
What Google Business Profiles, reviews and directories contribute
Google Business Profile remains foundational for Google’s own local ecosystem. Google advises businesses to verify profiles and maintain accurate details including category, hours, address or service area, website and other attributes. It also says local ranking is influenced by relevance, distance and prominence, and that reviews and links can contribute to prominence. (support.google.com)
That guidance should not be stretched into a claim that a complete profile directly causes ChatGPT, Perplexity or Gemini to recommend a company. The five systems do not publish one shared local-ranking formula, and this one-business test does not prove a ranking factor.
What can be said more carefully is that consistent public business information reduces avoidable ambiguity. For a local electrician, that includes:
- Exact business name, primary phone number, domain and service area on the website and Google Business Profile.
- Accurate service descriptions, such as panel upgrades, emergency repair, EV charger installation or commercial electrical work.
- Current opening hours and holiday hours.
- Genuine customer reviews handled within each platform’s policies.
- Correct listings on relevant business directories and trade associations.
- Location and service pages that explain where the business operates rather than repeating city names without useful detail.
This is an entity-consistency task, not a promise of an AI ranking. It supports the source material that answer engines may encounter across the web. A citation audit can reveal whether those sources are actually appearing in answers.
Google AI Overviews and AI Mode are not the same as ChatGPT or Perplexity
Google AI Overviews and Google AI Mode should be tracked individually even though both sit within Google Search. Google describes AI Overviews as a generative snapshot linked to supporting web content, while AI Mode is a more conversational search experience designed for detailed questions and follow-ups. (search.google)
ChatGPT Search can use web search and location information for local results. Perplexity describes itself as an AI-powered search engine that searches the web and provides citations. (help.openai.com) Microsoft Copilot can also ground web answers in Bing search results. (support.microsoft.com)
So, do Google AI Overviews and AI Mode use the same sources as ChatGPT and Perplexity? Not necessarily. They may overlap on highly visible sources—such as a company website, Google Business Profile-related data, large review platforms, local media or business directories—but their retrieval, ranking and response-generation processes are separate. The 40% to 63% pairwise overlap found in the test is a practical reason not to assume source parity.
For a deeper explanation of how Google, ChatGPT and Perplexity differ in citation behavior, see Google AI Overviews vs ChatGPT vs Perplexity: How Citations Differ. For the broader strategic framework, see Answer Engine Optimization: What It Is and How It Works.
A repeatable monitoring workflow for local AI visibility
A durable process measures rates over time rather than celebrating a single answer. The following workflow adapts the useful parts of the 32-day test without claiming that one market represents every market.
- Build a prompt set of 10 to 25 questions. Include service-plus-location prompts, comparison prompts, urgent-need prompts and recommendation prompts. A plumber might track “emergency plumber near [area],” “who installs tankless water heaters in [city]” and “best-reviewed residential plumber in [city].”
- Specify test conditions. Record date, time, city or simulated location, logged-in status, device/browser, language and whether a search feature was manually enabled.
- Run each prompt across each target surface. For a North American local business, that may include ChatGPT, Gemini, Google AI Mode, AI Overviews, Perplexity and Microsoft Copilot where available.
- Capture the entire answer and its visible sources. Save the prompt, response text, business names, order of appearance, citations, screenshots and errors.
- Normalize entities. Merge obvious spelling, punctuation and legal-suffix variants so “Example Electric,” “Example Electric Ltd.” and “Example Electrical” are not counted as separate businesses.
- Calculate weekly and monthly rates. Track mention rate, recommendation rate, first-position rate, citation rate, factual-error rate and no-business-answer rate by platform.
- Review cited domains. Identify which directories, review sites, publishers and owned pages recur. This is where a CiteMetrix-style source-overlap view is more useful than a vague AI visibility score.
Audra can support this process by combining answer-engine checks with technical SEO, performance, accessibility and link audits in a local desktop workflow. The important principle is that answer visibility and site health should be viewed together, while still reported as separate measurements.
What a client-ready report should say—and avoid saying
A concise report can be both more honest and more actionable when it separates observation from inference.
Useful reporting language:
- “The business was mentioned in 14 of 20 ChatGPT runs this month.”
- “Perplexity cited the company website in 6 of 12 relevant answers and a local directory in 8.”
- “Google AI Mode named the business in 9 of 20 prompts, but not consistently for EV charger queries.”
- “Three listings showed conflicting service-area wording; these are candidates for cleanup.”
Language to avoid:
- “The business ranks number one in AI.”
- “Updating this directory listing will make ChatGPT recommend the company.”
- “A Google Business Profile improvement will automatically improve Perplexity visibility.”
- “One screenshot proves AI search dominance.”
The difference is material. The first group reports observed rates within defined prompts and dates. The second group converts volatile outputs into claims the data cannot support.
Businesses that also see indexing anomalies should resolve those before interpreting AI visibility too aggressively. Google Search Console Only One Page Indexed? Fix It covers one common technical issue that can restrict the set of owned pages available to search systems.
The limits of this data set are part of the finding
The test was transparent about its limits: one electrician, one city, one trade, logged-out queries and a 32-day window. Perplexity had less coverage than the other systems. The business names were normalized manually, which is reasonable but introduces a judgment call around chains, subsidiaries and near-identical names.
It also did not measure conversions, call volume, user personalization, map-pack exposure, competitor review changes, Google Business Profile edits, paid placements, or source-crawl timing. It cannot tell a local business which assistant is universally most likely to recommend it.
Still, the findings support one operational conclusion: AI answer-engine visibility is an observed, time-bound distribution—not a permanent position. Local businesses should monitor it like a volatile share-of-voice metric, then investigate recurring citations and factual errors rather than optimizing against one answer from one morning.
For teams adapting Google strategy around these surfaces, Google AI Overviews to AI Mode: SEO Changes Explained provides additional context on the shift from overview-style answers to more conversational AI search.
FAQ
How often do ChatGPT, Gemini, Google AI Mode, AI Overviews and Perplexity give the same answer?
In this 32-day, one-business test, every answering system named at least one business in common on 37 of 416 question-days, or 9%. Another 52 question-days had agreement because no system named a business. On the 201 observations where all five systems answered, they never all named a shared business.
Which AI assistant is most likely to recommend a local business?
This data set cannot answer that reliably. It tracked one Vancouver-area electrician, not a representative set of industries or cities, and it did not publish platform-by-platform recommendation rates. A business should measure its own prompt set across ChatGPT, Gemini, Google AI Mode, AI Overviews, Perplexity and, where relevant, Microsoft Copilot.
What sources do AI systems use when choosing local business recommendations?
The systems can use public web information, but the exact source mix varies by platform and query. In the study, changes in ChatGPT recommendations appeared to coincide with different retrieved lists and directories. Businesses should inspect visible citations, then check accuracy across their website, Google Business Profile, reviews, directories and relevant local publications.
How should businesses measure visibility in AI-generated answers?
Use repeatable prompts and fixed conditions over several weeks. Record business mentions, recommendation wording, list position, citations, factual errors and no-business answers separately for each platform. Monthly rates are more defensible than a single screenshot because the study found 29% to 37% day-to-day self-disagreement across individual systems.
Do Google AI Overviews and AI Mode use the same sources as ChatGPT and Perplexity?
They can overlap, but they should not be treated as the same source system. Google says AI Overviews and AI Mode work with Search and link to web content, while ChatGPT and Perplexity also use web-grounded answers with citations. Their retrieval and answer-generation systems differ, and this test found only 40% to 63% same-day overlap between platform pairs.
Sources
- https://www.reddit.com/r/SEO/comments/1x353ig/i_tracked_1814_ai_answers_for_one_local_business/
- https://search.google/pdf/google-about-AI-overviews-AI-Mode.pdf
- https://www.search.google/ways-to-search/ai-overviews/
- https://search.google/ways-to-search/ai-mode/
- https://support.google.com/business/answer/7091
- https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt
- https://www.perplexity.ai/help-center/en/articles/10352155-what-is-perplexity
- https://support.microsoft.com/en-us/microsoft-365-copilot/how-web-search-works-in-microsoft-365-copilot-chat-and-agents
- https://openai.com/index/introducing-chatgpt-search/
- https://support.google.com/business/answer/7091?hl=