AI Search and Local Business Recommendations: What a Small Test Revealed

A business searching for an IT provider in Worcester, Massachusetts, might receive very different recommendations depending on whether it uses ChatGPT, Gemini, or Claude. Even when the question is identical, the platforms select different companies, emphasize different qualifications, and use different information sources to explain their recommendations.

Those differences become more interesting when the customer moves beyond a basic search and starts describing specific requirements, business problems, or concerns about making the wrong decision. Instead of just asking for a list of providers, the customer is asking the AI platform to help determine which businesses are suitable and why.

To explore how this works, I tested ChatGPT, Gemini, and Claude using three levels of local search intent. The research focused on managed IT service providers in Worcester, Massachusetts, a category where technical capabilities, responsiveness, cybersecurity, and customer trust can materially influence a purchasing decision.

Across the nine responses tested, the platforms evidenced considerable variation in which businesses they recommended and how they justified those choices. As the questions became more detailed, the answers also became more advisory, connecting individual businesses to specific customer needs and providing guidance on how to evaluate providers.

I also examined the sources referenced within the responses, including company websites, specialized service pages, location landing pages, business directories, and customer reviews. These provided additional context about how the platforms supported their recommendations and exposed some limitations in the evidence they presented.

How I tested local business recommendations across three AI platforms

The test used a simplified version of the five-level local intent framework from my earlier Google Ask Maps research published on Search Engine Land. That framework examines how recommendations change as searches progress from basic business discovery toward more detailed questions involving situations, trust, and decision-making.

For this comparison, I consolidated those five levels into three prompts representing different stages of the customer journey. The goal was to observe whether introducing additional context changed the businesses recommended, the attributes emphasized, and the information used to support each recommendation.

Level 1: Basic provider discovery

Prompt: “What are some managed IT service providers in Worcester, Massachusetts?”

This is a basic category-and-location search. The customer knows the type of business they need but hasn’t provided additional requirements about company size, industry, technical needs, or service expectations. The purpose was to establish which businesses appeared in each platform’s initial recommendations and how much overlap existed between them.

Level 2: Specific capabilities and business requirements

Prompt: “I’m looking for an IT company in the Worcester area that supports businesses with 20–50 employees, including cybersecurity and Microsoft 365.”

The second question introduced the customer’s business size and two specific service requirements. This allowed me to examine whether the platforms changed their recommendations when they needed to match providers against a more detailed set of capabilities, rather than relying primarily on business category and location.

Level 3: Trust, risk, and decision support

Prompt: “Our Worcester-area business has outgrown its current IT provider. We’re worried about cybersecurity, slow response times, and disruptions when switching providers. Which local IT firms should we consider, and why?”

The final prompt described a business that needed a new provider but was concerned about security, responsiveness, and operational disruption. A useful recommendation would need to address those concerns, explain why particular providers might be suitable, and identify information that would help the customer make an informed decision.

For each response, I examined the businesses recommended, the overlap between platforms, the reasons provided for each recommendation, and the types of evidence referenced. I also reviewed the citation URLs captured with the original responses to identify which information was being presented as support for particular business descriptions.

Research scope and limitations

This was a small exploratory test conducted in October 2026, consisting of nine responses: three prompts submitted to each of the three platforms. Each prompt was tested once, and all searches focused on managed IT providers serving Worcester, Massachusetts.

The limited sample means the findings shouldn’t be interpreted as representative of how these platforms consistently recommend businesses. AI-generated responses can vary between searches, and different markets, industries, prompts, model versions, and platform settings may produce different results. The test also didn’t control for every difference in how the platforms access and retrieve information.

The source review adds useful context, but it has limitations of its own. The captured links show information cited or referenced within the responses, not every source the systems may have considered. Some citations were embedded in interactive business results and others were missing. I examined selected source pages, but did not independently verify every claim about the businesses or conduct a full assessment of their service quality.

The results document what the platforms recommended, how they described businesses, and what supporting sources were visible. They do not establish stable rankings, platform-wide preferences, or causal relationships between individual sources and recommendations.

Finding 1: The same local search produced different business recommendations

The first question produced the clearest example of disagreement between platforms. ChatGPT identified seven Worcester managed IT providers, including Domitek, Cinch I.T., Machado Consulting, iMedia Technology, Technology Advisory Group, Bryley Systems, and Triton Technologies. Gemini returned five different companies: Integris, Paragus Strategic IT, Rutter Networking Technologies, Attain Technology, and VTC Tech.

Claude recommended three companies: Triton Technologies, Cinch I.T., and iMedia Technology. Although ChatGPT and Claude shared three providers, not one company appeared in all three platforms’ responses to the basic search.

This is notable because the customer had provided very little information beyond the service category and location. Each platform had broad freedom to determine which providers were relevant. The resulting shortlists differed substantially. A business that appeared prominently in one response could be entirely absent from another.

The answers also varied in how they presented the businesses and the sources they referenced.

ChatGPT emphasized locally listed businesses and general service information

ChatGPT’s response included business addresses, service descriptions, and links to company websites such as Domitek, Cinch I.T., and Bryley Systems. It also linked to a broader Worcester managed IT provider directory for additional options.

The response resembled an expanded local business listing, with additional context about the services each company provided. Most of the visible links directed users to company homepages rather than sources addressing a particular specialization or customer concern.

Gemini drew heavily on Worcester-specific service pages

Gemini included a broader mix of local and regional firms, describing their industry focus and technical capabilities. Its citations were particularly interesting because several pointed to location-specific landing pages.

For example, Gemini referenced Integris’s Worcester location page, Paragus Strategic IT’s Worcester page, and Rutter Networking Technologies’ Worcester managed IT services page. It also referenced Attain Technology’s Worcester IT support page and VTC Tech’s Worcester service page.

Several of these citations contained text fragments identifying specific passages within the pages. The referenced passages related to services, industries, geographic coverage, and technical capabilities.

This doesn’t establish that location pages caused these businesses to be included. It does show that Gemini’s response used location-specific company content to support descriptions of providers serving the Worcester market.

Claude used a smaller set of providers and more varied sources

Claude returned a shorter list and explicitly cautioned that the information it found did not establish actual service quality. Its citations included Triton Technologies’ website, a third-party Triton provider profile, and a Cloudtango listing for Cinch I.T..

For iMedia Technology, Claude referenced a syndicated article about the company’s managed services. Although that article appeared on a third-party website, its promotional subject matter means it shouldn’t automatically be treated as independent evidence of service quality.

The first prompt produced differences in both the recommended businesses and the types of information presented in support of those recommendations. The platforms selected different shortlists while drawing on varying combinations of company websites, local landing pages, directories, and published articles.

Finding 2: Specific requirements changed which businesses were recommended

The second prompt introduced three additional criteria: a company with 20–50 employees, cybersecurity support, and Microsoft 365 services. The responses became a bit more selective, and the explanations placed much more emphasis on matching business capabilities to those requirements.

The supporting links also became more specific in several responses, moving from general descriptions of IT providers toward pages addressing individual services, technical capabilities, and client requirements.

ChatGPT connected specialized service pages to customer requirements

ChatGPT recommended Domitek, Machado Consulting, ClearCom IT Solutions, and Bryley Systems. Rather than only describing their general services, it connected those companies to specific needs involving cybersecurity, Microsoft 365 administration, compliance, and small-business support.

For Domitek, ChatGPT referenced the company’s Microsoft 365 service page and SaaS protection page. These pages describe Microsoft-related services, cloud applications, data protection, and administration capabilities that corresponded to the requirements in the prompt.

The Microsoft 365 page provides a particularly useful example. Domitek explains its offering, describes security and administration features, and outlines the services it can provide to businesses using Microsoft’s cloud applications. The page gives a potential customer more information than a general statement that the company provides IT support.

ChatGPT also linked to ClearCom’s Worcester managed IT services page when recommending the company for organizations in the requested employee range.

These examples demonstrate how a response can use specialized service content to connect a provider’s advertised capabilities with a customer’s more detailed requirements. Whether those services are suitable for a particular organization would still require further evaluation.

Gemini relied on specific descriptions of capabilities and client fit

Gemini recommended Bryley Systems, Paragus Strategic IT, Integris, and Applied Computer Sciences. Its response emphasized technical capabilities such as endpoint detection and response, multifactor authentication, cloud management, and compliance support.

The supporting citations again included location-specific pages, but the passages referenced were more directly connected to technical services and client requirements.

Gemini cited Bryley’s managed IT services page for Worcester, Paragus’s Worcester page, and Integris’s Worcester location page. In Integris’s case, one citation pointed specifically to content describing the sizes of organizations the company typically supports.

This was particularly relevant because the customer had identified a business with 20–50 employees. The AI response was no longer evaluating providers against a broad category alone; it was attempting to match organizational requirements to advertised service capabilities and target markets.

One limitation was that Gemini also recommended Applied Computer Sciences without an accompanying source link in the captured response. The recommendation included several service claims, but the available citation record didn’t provide the same opportunity to trace those claims as it did for the other businesses.

Claude used customer reviews to differentiate providers

Claude recommended Cinch I.T., Bryley Systems, iMedia Technology, CRCIT, and Technology Advisory Group. Its response drew heavily on customer review ratings and feedback rather than detailed citations to individual service pages.

Cinch I.T. was associated with long customer relationships and responsive support. Bryley Systems was connected to technology planning and assistance with cybersecurity insurance requirements, while iMedia Technology was associated with data security, backups, and ongoing support.

Claude also acknowledged an important gap in the evidence gathered. While the reviews provided information about customer experiences, they didn’t establish whether the providers offered the specific Microsoft 365 capabilities the customer are looking for. Claude recommended confirming those services directly before making a decision.

The captured Level 2 response didn’t include direct links to the reviews Claude referenced. When asked about its sources, Claude indicated that the reviews came from Google Places. However, some of the review counts and average ratings differed from those reported in the original response. These discrepancies raise questions about the accuracy of the review data and make it difficult to verify the specific customer experiences Claude described or determine whether the supporting evidence was accurately represented.

Bryley Systems was the only provider appearing across all three platforms for this question. The overlap was greater than in the basic search, but the platforms still differed considerably in their shortlists and the evidence they emphasized.

The comparison illustrates how professional-service businesses can be evaluated through different kinds of information. Advertised technical capabilities, descriptions of client fit, and customer experiences may all be relevant, but they support different aspects of a purchasing decision.

Finding 3: Trust-oriented questions moved the platforms toward decision support

The third question introduced a more complicated situation. The customer was dissatisfied with an existing IT provider and concerned about cybersecurity, response times, and the consequences of switching.

All three platforms responded by offering recommendations alongside guidance about how to evaluate potential providers. The supporting information became particularly interesting because the customer was asking for help managing the risks of a decision, rather than simply identifying businesses capable of providing IT services.

ChatGPT emphasized documented processes and business fit

ChatGPT recommended six companies, leading with ClearCom IT Solutions, Machado Consulting, Domitek, and Cinch I.T., followed by Bryley Systems and TOSS C3. Its explanations connected providers to different aspects of the customer’s situation rather than presenting them as interchangeable businesses offering similar services.

ClearCom was recommended first because of its published response-time targets and onboarding procedures. Machado was associated with cybersecurity and industry-specific compliance requirements, while Domitek was positioned around its discovery assessment, risk identification, and tailored implementation process.

These descriptions demonstrate how company-specific information can be used to explain why a particular business may be suitable for a customer’s circumstances. In this case, published processes and operational details helped ChatGPT differentiate providers according to the problems the customer was trying to solve.

Gemini emphasized specialization and technical expertise

Gemini recommended Machado, Bryley Systems, and Integris, focusing heavily on cybersecurity, business continuity, compliance, and the capacity to support growing organizations. Machado was framed around regulated industries and a dedicated support structure, Bryley around continuity planning and regional experience, and Integris around advanced security leadership and co-managed IT services.

Gemini also provided a detailed transition framework covering system documentation, security tools, provider offboarding, and employee support during the changeover. Its response demonstrated substantial interpretation of the customer’s situation, although some provider-specific descriptions were more assertive than the evidence visible in the captured answer could independently establish.

Claude emphasized due diligence and verification

Claude recommended Machado, Triton Technologies, Braided Technologies, Cinch I.T., Bryley Systems, and IT Solutions. Its response was comparatively cautious about the difference between a company’s advertised capabilities and independently verified performance.

Claude acknowledged that much of the available information came from company marketing materials and stated that it couldn’t verify service quality, response times, or pricing. Rather than treating the recommendations as definitive, it explained how the customer should investigate cybersecurity practices, contractual service-level agreements, client references, and transition plans before choosing a provider.

This represented a different type of decision support. Claude used its recommendations as a starting point for evaluating potential providers, while making clear that additional verification would be necessary before reaching a conclusion about their suitability.

How the decision-making process changed

By the third prompt, the platforms were responding to more than the category “managed IT services.” They were addressing the customer’s underlying concerns about operational risk, service quality, security, and whether a new provider could manage a complicated transition.

A company’s services remained relevant, but information about its processes, experience, approach, and ability to address the customer’s particular circumstances became more prominent. The responses increasingly resembled an evaluation process, where the AI platform helped connect the customer’s situation to the qualifications and characteristics of potential providers.

The source citations reinforce this observation. Pages describing specific processes, industry capabilities, cybersecurity services, and transition planning were particularly relevant to explaining why certain providers might suit the customer’s situation.

However, the presence of a citation doesn’t establish that a recommendation is accurate or that the cited page caused the company to appear. Those questions require separate evaluation.

How much agreement was there between ChatGPT, Gemini, and Claude?

The recommendation overlap changed across the three levels of intent. The basic search produced no provider appearing in all three responses. The second produced one shared recommendation, and the third produced two.

Search intent	ChatGPT	Gemini	Claude	Providers appearing in all three
Basic discovery	7	5	3	None
Specific capabilities	4	4	5	Bryley Systems
Trust and advisory	6	3	6	Machado, Bryley Systems

There were also substantial differences in which providers appeared repeatedly within each platform. ChatGPT included Domitek in all three responses, eventually positioning it among the companies worth contacting first for an IT transition. Domitek did not appear in Gemini’s or Claude’s primary recommendations during this test.

Machado Consulting also appeared in all three ChatGPT responses and became a shared recommendation across all three platforms when the question introduced trust and transition concerns. Bryley Systems appeared across all three platforms in both the second and third prompts, despite not appearing in Gemini’s or Claude’s initial basic recommendations.

These patterns illustrate two distinct aspects of AI-driven local search: whether a business gets included and how it is positioned when it appears. The reasons a company is recommended, the capabilities associated with it, and the supporting evidence can be as important to understanding the response as the business’s inclusion.

Although the results showed increasing agreement across the three prompts, the sample is too small to establish a general relationship between query complexity and recommendation consistency. A broader test with additional markets, service categories, and repeated searches would be necessary to determine whether the pattern holds.

What evidence appeared in the AI recommendations?

The platforms differed in the businesses they recommended, but also in the kinds of information they cited and the claims those sources were used to support. Several evidence categories appeared across the test, ranging from general business information to specific service descriptions, customer experiences, and published operational processes.

Business identity, location, and service information

In the basic search, company names, locations, service descriptions, and industry focus played prominent roles. This information helped establish which businesses appeared relevant to the service category and geographic area, although the platforms differed in how narrowly they interpreted the Worcester market.

Gemini’s references to Worcester-specific landing pages were particularly noticeable. Pages from Integris, Paragus, Rutter, and other providers explicitly connected their services to the geographic market being searched. For regional and multi-location businesses, accurate information about office locations, service coverage, and the relationships between locations can help communicate where and how the organization operates.

Specialized capabilities and industry experience

When the prompt introduced Microsoft 365 and cybersecurity requirements, the responses emphasized more specific qualifications. These included technical services, certifications or credentials, compliance experience, and descriptions of the businesses served.

ChatGPT’s references to Domitek’s Microsoft 365 and data protection pages demonstrated how specialized service content could support an explanation of provider relevance. Gemini’s citations to descriptions of Bryley’s security capabilities and Integris’s client-size range served a similar purpose.

The same principle applies across professional services. A financial advisory firm may serve specific types of clients, an accounting practice may specialize in particular industries, and a cybersecurity provider may focus on regulated organizations. Clear descriptions of these capabilities give customers information they can use to evaluate whether a business is appropriate for their needs.

Customer experience and reputation

Claude’s Level 2 response was particularly notable for using review content to describe responsiveness, long-term client relationships, and strategic support. Instead of treating reviews only as a source of ratings, it referenced details about how customers experienced the services delivered. The review examples associated providers with characteristics such as ongoing technology planning, assistance with cybersecurity insurance requirements, reliable support, and established customer relationships.

This illustrates the additional context customer feedback can provide beyond a company’s own service descriptions. Having said that, the absence of individual review URLs in the captured response limits the ability to determine whether Claude accurately represented those experiences.

Reviews can offer valuable evidence about customer experiences, but they don’t independently establish a provider’s current technical capabilities or whether it can meet every requirement in a prospective customer’s request.

Documented processes and supporting evidence

The advisory question produced some of the clearest examples of specific business content being used to support recommendations. ClearCom’s switching-provider page addressed the customer’s concern about transition disruption through documented onboarding procedures and customer testimonials. Domitek’s managed services and company information described its approach to assessments and planning, while Gemini’s cited cybersecurity pages supplied additional context about specialized services.

These sources helped the responses move beyond generic descriptions of what each provider offers toward explanations of how a provider might address a particular customer problem.

For professional-service businesses, this suggests value in documenting how important services are delivered. Detailed service pages, case studies, implementation processes, FAQs, and examples of previous work can provide information that helps customers evaluate the organization beyond its general capabilities.

A citation doesn’t necessarily verify a recommendation

One additional finding from reviewing the source links is the need to distinguish between a citation being present and the citation adequately supporting the accompanying claim. The sources used in these responses ranged from business homepages and specialized service pages to industry directories and syndicated articles. Each can provide useful information, but they don’t necessarily carry the same evidentiary weight.

A company website may establish that a provider advertises Microsoft 365 support, but it doesn’t independently verify service quality. A third-party directory may confirm that a company operates in a particular market, but it may contain information supplied by the business itself. A syndicated article may appear on an external publication while primarily reproducing promotional material.

The source review also surfaced a possible citation mismatch worth investigating. In its Level 3 response, Claude described Bryley Systems as a managed IT and cloud provider, but the accompanying Elioplus citation URL identifies Charland Technology. The URL suggests that the cited profile may not correspond to the company being described. Without confirming the current contents and any historical relationship between those businesses, it would be premature to conclude that the citation is definitively incorrect.

Even so, it illustrates why source inspection matters. A business recommendation can contain a plausible description and an apparently supporting citation, while the relationship between the two still requires verification. This is particularly important for professional services, where claims about qualifications, certifications, response times, regulatory experience, and service quality can influence consequential purchasing decisions.

The platforms also differed in how they communicated uncertainty. Claude explicitly distinguished between advertised capabilities and verified performance. ChatGPT acknowledged that its recommendations were candidates for further evaluation, while Gemini generally presented more confident assessments of provider suitability.

None of these approaches eliminates the need to check the underlying evidence. For research purposes, business inclusion, supporting citations, factual accuracy, and independently verified performance should be treated as separate measures.

What the findings mean for professional-service and regional businesses

The implications extend beyond managed IT providers. Accounting firms, cybersecurity consultancies, financial advisory businesses, healthcare organizations, and other professional-service companies often compete on qualifications that are difficult to communicate through a business category or short service description alone.

Customers may need to know whether a provider works with organizations of their size, understands their industry, has experience with particular challenges, or follows a process that reduces the risks of hiring them. AI-driven search allows customers to combine those considerations into a detailed request, potentially bringing more of the evaluation process into the initial search experience.

The source review also reinforces the importance of considering where those qualifications are documented and whether the available information supports the business’s intended positioning.

Make specializations and client fit explicit

A general statement such as “comprehensive IT solutions for businesses” establishes a category but offers limited information about who the provider is best equipped to serve. A more useful description explains the types of organizations served, relevant industries, supported technologies, operational requirements, and particular areas of expertise.

The Level 2 test demonstrated how business size and technical requirements could change the recommendations and the information used to justify them. Pages describing supported technologies, client types, and specialized services gave the platforms specific material to reference when explaining provider suitability.

For a multi-location professional-service organization, that might include how services are delivered across offices, whether support is centralized or local, and what level of operational complexity the organization can manage.

Explain how the business solves important customer problems

The third prompt demonstrated why service delivery information matters. A customer concerned about switching IT providers may be looking for transition planning, documented response commitments, clear responsibilities, and safeguards against operational disruption.

ClearCom’s published onboarding process was a particularly useful example because it directly addressed the customer’s stated concern. The cited page described what the company says prospective customers can expect during a transition, allowing ChatGPT to connect that information to the request.

Businesses in other industries can apply the same principle by documenting processes that help customers evaluate risk and understand what working with the organization involves. Relevant evidence can include service pages, case studies, client examples, implementation procedures, and answers to common questions that arise before a purchase.

Pay attention to what customer reviews actually say

Review counts and average ratings remain useful indicators, but the language of customer feedback provides additional context. In this test, Claude used review details to associate providers with responsiveness, established client relationships, cybersecurity guidance, and technology planning.

Businesses should examine whether their customer feedback accurately reflects the qualities and capabilities that matter in their market. Encouraging genuine, detailed feedback about real customer experiences can create a more informative public record than relying primarily on ratings or generic praise.

The emphasis should remain on authentic experiences rather than directing customers to use predetermined wording. Businesses should also recognize that AI-generated descriptions of reviews may omit context or misrepresent individual experiences, making source verification important.

Evaluate how the business is described and what supports the description

A business might be recommended for a service it no longer prioritizes, associated with a specialization it doesn’t offer, or described in terms that don’t reflect its intended market position. Prospective customers may use these AI-generated descriptions to decide which businesses deserve further consideration.

When testing AI recommendations, examine whether the business appears for general searches and more specific situations, which capabilities are associated with it, and what evidence supports those descriptions.

The original source links should be part of that review. A citation to a company homepage communicates something different from a citation to a detailed service page, a case study, an independent customer review, or an industry publication.

Looking at competitors can also be informative. The objective is to understand how the market is being interpreted, which companies appear in recommendations, how they are described, and whether the available evidence supports the organization’s intended positioning.

From local discovery to evaluation and choice

Traditional local search leaves much of the comparison work to the customer, from reviewing search results and visiting websites to reading reviews and evaluating potential providers. AI-driven search can bring more of that process into a single interaction by interpreting customer needs, recommending businesses, and explaining why they might be suitable.

The Worcester managed IT test illustrated this shift, with recommendations and supporting evidence changing as the prompts became more specific. The platforms differed in which businesses they recommended, how they described them, and what sources they cited. The research also highlighted the importance of examining whether those citations actually support the claims being made.

For businesses, the findings reinforce the importance of maintaining accurate, detailed information about their services, expertise, experience, and processes across the sources customers and AI platforms may consult. As AI becomes more involved in evaluating providers, the evidence supporting why a business is a suitable choice may become as important as the information establishing what it offers.