AI Hallucinations and Local Businesses: Risks and Protection Strategies

Why AI Hallucinations Are Becoming a Business Risk for Local Companies in 2026

AI assistants have become an integral part of local search. According to an international study from the USA,45% of consumers use AI for local business recommendations in 2026, compared to only 6% in 2025. This makes AI one of the most important tools for local recommendations, behind Google and Facebook, but ahead of Yelp and TripAdvisor. This shift also affects the DACH region: When customers ask ChatGPT, Gemini or Perplexity where the nearest tax advisor, hairdresser or plumber is, the AI's answer determines the first contact.

The problem: These answers are not always correct. AI systems regularly generate false information about local businesses. Incorrect opening hours, invented services or made-up phone numbers can cause real revenue losses. Particularly critical is the trust that users place in these answers: According to the same US study,63% of respondents trust AI recommendations, and 64% of AI users trust tools like ChatGPT for local decisions as much as real reviews. Those who are misrepresented lose customers without noticing.

AI has also arrived in everyday business in Austria. According to Statistics Austria,20.3% of companies with ten or more employees already used AI technologies in 2024, almost twice as many as in 2023 with 10.8%. Local businesses therefore stand on both sides: as AI users and as potential victims of erroneous AI representations.

What AI Hallucinations Are and Why They Are Not Solved in 2026

An AI hallucination occurs when an AI system presents information with conviction that is factually incorrect or completely invented. Large language models (LLMs) predict the next probable word. When reliable data is missing, they fill the gap with plausible guesses instead of an honest "I don't know".

Three causes are particularly relevant for local businesses:

• Training data gaps:Small businesses are underrepresented in the training data. The model has learned hardly any reliable facts about them.
• Data voids:For rare search queries, such as for a niche business in a small town, little authoritative content exists online. This data void is filled with guesses or content from competitors.
• Real-time web pollution:Modern chatbots search the web live. This does not solve the problem, but shifts it. According to NewsGuard,ten leading AI tools repeated misinformation on current topics in 35% of cases in August 2025, compared to 18% in August 2024. At the same time,the refusal rate fell from 31% to 0%. The models therefore answer more frequently, but they draw from a polluted information ecosystem and treat unreliable sources as credible.

That even controlled tasks produce hallucinations is shown by the Vectara Hallucination Leaderboard. When summarizing given documents,error rates vary depending on the model from a few percent to double digits. The model choice makes a big difference, but none is error-free.

How AI Obtains Local Business Data and Where Errors Arise

AI systems do not create their own business directory. They aggregate existing sources: Google Business Profile, Bing Places, business directories, review portals, social media profiles and the company's own website. ChatGPT with web search partly relies on Bing data, Google AI Overviews on the Google index.

Errors arise mainly in two places:

• Inconsistent NAP data:NAP stands for Name, Address, Phone. If the company name on the website is "Müller GmbH", in the Google profile "Müller GesmbH" and in the old Herold entry "Müller & Söhne", the AI cannot clearly determine whether it is one or several companies.
• Entity fragmentation:Contradictory data distributed across many platforms leads to the AI perceiving the company as a fragmented entity. It then combines fragments, such as an old phone number with a new address, or transfers the services of a competitor.

The more contradictory the data situation, the greater the room for interpretation and the risk of invented details.

Concrete Risk Scenarios for Local Businesses

The following scenarios are not theory, but direct consequences of data gaps and inconsistencies:

• Incorrect opening hours:The AI states outdated or invented times. Customers stand in front of a closed door and go to the competition.
• Invented services:The AI claims the business offers services that do not exist, or omits the core business. Both lead to incorrect inquiries and frustration.
• Wrong address or phone number:Customers reach no one or end up in the wrong place. With a foreign, taken-over number, a third party may benefit.
• Made-up recommendations to competitors:In a data void, the AI recommends better-documented competitors when in doubt, even though one's own business would be more suitable.
• Invented reviews or review content:The AI summarizes reviews and invents tonality or details that were never written.

Legal and Liability Risks in the DACH Region

The question of who is liable for AI false statements is not conclusively clarified in 2026 either, but is gaining contour. A frequently cited precedent is Moffatt v. Air Canada (Canada, 2024): A court ruled that Air Canada is liable for false information from its website chatbot. The company could not claim that the bot was a separate legal entity. The core lesson: Operators of an AI system bear responsibility for its statements to customers.

For local businesses in the DACH region, two constellations must be distinguished:

• Own chatbot:Anyone who operates an AI assistant on their website is liable in case of doubt as for any other statement by the company. False statements can have competition and warranty law consequences.
• Third-party AI about one's own company:If ChatGPT or Google makes false claims about a business, the legal situation is more complex. Depending on the content, UWG aspects (misleading or disparaging statements, especially with competitor reference) as well as personal and reputational legal questions may apply.

The EU AI Act additionally increases regulatory pressure on AI providers, for example through transparency obligations. However, it does not exempt local businesses from the obligation to actively control their own data representation. In practice, prevention is cheaper and faster than any litigation.

Protection Strategy 1: Data Consistency as Foundation

The most effective protection is a complete, consistent data foundation. If the same facts are everywhere, there is nothing to hallucinate.

• Establish NAP consistency:Name, address and phone number must be written exactly identically on every platform, down to the legal form and spelling of the street.
• Maintain Google Business Profile completely:Keep opening hours, services, categories, photos and description current. This profile is one of the most important sources for AI systems.
• Don't forget Bing Places:Since ChatGPT partly accesses Bing data, this profile should also be complete and maintained.
• Cover Austrian directories:Check and update entries at Herold and in the WKO Firmen-A-Z. Outdated legacy data is a common hallucination source.
• Use structured data:Schema.org markup such as LocalBusiness, Organization, OpeningHoursSpecification and PostalAddress provides AI systems with machine-readable, unambiguous facts directly from your own website.

Protection Strategy 2: Monitoring Your Own AI Representation

What you don't measure, you can't correct. A fixed monitoring routine uncovers errors before customers stumble over them.

• Test directly:Regularly query ChatGPT, Gemini, Perplexity and Google AI Overviews with realistic search queries about your own business, services and opening hours.
• Ask with variety:Test different formulations and languages, as AI gives different answers depending on the prompt.
• Check competitor recommendations:Monitor whether the AI mentions your own business or the competition for relevant queries.
• Document results:Record answers with dates to identify changes and recurring errors.

That monitoring remains necessary is shown by AI Overviews: According to an analysis by Search Engine Land,Google AI Overviews answered correctly in 91% of cases in February 2026, compared to 85% in October, leaving around 9% incorrect. With Google's data volume, even small error rates add up to millions of false answers.

Protection Strategy 3: Correction and Escalation

When an error is discovered, quick, structured action counts.

• Correct the source:First fix the data source behind the error, such as an incorrect directory entry or outdated opening hours in the Google profile.
• Use feedback channels:In Google AI Overviews and in the chatbots there are feedback and reporting functions for incorrect answers. Use these specifically.
• Create authoritative first-party source:Create a clear, well-structured page on your own website with all facts. It serves as a reliable reference point for future AI answers.
• Escalate if necessary:For persistent, reputation-damaging or legally relevant false statements, use the official support and correction channels of the providers and document the process.

AI-SEO and GEO: Becoming the Preferred Correct Source for AI

Generative Engine Optimization (GEO) and AI-SEO aim to make AI systems prefer your own, correct content. Instead of just fighting hallucinations, you proactively provide the AI with the best foundation.

• Clearly structured content:Clear headings, short paragraphs and direct answers to typical customer questions make it easy for the AI to cite correct statements.
• Increase fact density:Anchor concrete, verifiable information about location, services, price range and specialization.
• Strengthen local signals:Explicitly name local reference, catchment area and regional relevance, especially for Austrian businesses with local focus.
• Consistency across all channels:Website, profiles and directories must tell the same story so that the AI recognizes a clear entity.

Reputation Management and Reviews as Shield

Current, consistent reviews stabilize AI representation because they create a broad, consistent data foundation. The more reliable reviews exist, the less room remains for invented details.

• Actively ask for reviews:Specifically invite satisfied customers to reviews, especially on Google.
• Respond to reviews:Responses provide additional, consistent context that AI systems can pick up.
• Understand tonality:Since AI summarizes reviews, recurring themes shape the AI image of the business.

Verification by users remains a second security layer: According to the US study,42% of AI users always check recommendations on native review platforms. If AI statements and real reviews match, this reinforces the first impression. If they diverge, trust is quickly lost.

Common Mistakes

• Treating data maintenance as a one-time task:Profiles become outdated. Without regular updates, inconsistencies return.
• Only paying attention to Google:Bing, directories and one's own website are neglected, even though they are AI sources.
• No monitoring:Errors remain undiscovered until a customer complains or stays away.
• Ignoring old directory entries:Outdated Herold or business directory entries with incorrect numbers or addresses remain as hallucination sources.
• Trusting one platform:Since models hallucinate differently, it is not enough to check only one AI tool.

Metrics and Measurement

Protection measures need measurable indicators:

• NAP consistency rate:Proportion of platforms with exactly matching master data.
• AI accuracy rate:Proportion of correct answers in regular tests across multiple AI tools.
• Error resolution time:Time from discovery of a false statement to the corrected source.
• Schema coverage:Proportion of relevant pages with valid LocalBusiness markup.
• Review currency and volume:Number and currency of reviews as stability indicator.

Practice Checklist: Immediate vs. Ongoing

Immediate measures:

• Check Google Business Profile:Verify opening hours, address, phone, services.
• Compare NAP data:Check website, Google, Bing, Herold and WKO for identical spelling.
• AI quick test:Query your own business in ChatGPT, Gemini, Perplexity and Google AI Overviews.
• Add Schema.org:Implement LocalBusiness markup on the website.

Ongoing measures:

• Monthly AI monitoring:Fixed routine across multiple tools.
• Quarterly data audit:Update all profiles and directories.
• Active review management:Obtain and respond to reviews.
• Correction workflow:Clear process for finding, reporting and fixing false statements.

Further Reading

AI hallucinations are not a temporary phenomenon, but a structural property of today's AI systems. Local businesses cannot switch them off, but can significantly reduce the risk. Those who combine data consistency, monitoring and quick correction become a clearly recognizable, correctly represented entity for AI. Sensible next steps are a complete local SEO setup, the implementation of structured data and the establishment of a fixed GEO routine, so that the AI prefers your own facts and does not invent them.