A construction business comparing accounting firms may receive different advice depending on whether it prioritizes industry experience, financial guidance, or a dependable relationship with a new accountant. Even when the same firms are being considered, AI platforms can emphasize different qualifications and reach different conclusions about which provider appears better suited to the customer.
To examine these comparisons, I collected 27 responses from ChatGPT, Gemini, and Claude using three accounting firms serving the Worcester, Massachusetts, area. The prompts varied the customer’s highest priority and the order in which the firms were named, allowing me to compare the recommendations, the explanations provided, and the evidence presented in support of them.
The responses showed meaningful differences in how the firms were positioned, particularly when the priority shifted from construction experience to ongoing financial advice. They also revealed a recurring limitation: some answers extended documented services or business characteristics into assumptions about capability and reliability. Construction experience became presumed job-costing expertise, while firm size became a basis for predicting continuity or personal attention.
Key Takeaways
Customer priorities changed how businesses were evaluated. Industry experience, financial guidance, and service dependability produced different explanations and, in some cases, different preferred providers.
Platforms weighed qualifications differently. Agreement was strongest when the customer prioritized industry experience; other scenarios produced more conditional advice and differences between platforms.
Business descriptions were not always complete. Some answers emphasized only part of a firm’s advertised services, changing the apparent differences between competitors.
Some recommendations exceeded the evidence. Industry positioning and firm size became assumptions about specific capabilities, continuity, or personal attention.
Different sources supported different parts of the comparison. Service pages, professional biographies, reviews, directories, and news coverage provided context, but did not establish every claim about suitability or performance.
Name-order variation did not establish a position advantage. The construction preference remained stable, but the small sample could not separate order effects from ordinary response variation.
A small study of three firms and three customer needs
The comparison included Bollus Lynch, LLP; Richard Mood CPA; and Masiello Accounting Services, LLC. The firms were selected through a random draw in Google Sheets from a pool screened for location or service coverage and advertised accounting, tax, and advisory services. None was a current client of mine, and the same three firms were used throughout the test.
Every prompt described the same prospective customer: an established construction business in Worcester with 20 employees seeking ongoing accounting, tax planning and preparation, and financial reporting. I varied the customer’s highest priority across three scenarios:
| Scenario | Highest priority |
|---|---|
| Construction experience | Documented industry experience, job costing, and project-level reporting |
| Financial advice | Interpreting reports, planning cash flow, and making business decisions |
| Service relationship | A dependable transition, clear communication, and defined support responsibilities |
For each scenario, I submitted three arrangements of the firm names to each platform. The remaining comparison instructions stayed the same, including requests for source links, meaningful differences, unresolved information, and questions to verify before deciding. That produced nine responses per scenario and 27 overall.
Because every prompt named the three firms, the test examined how the platforms evaluated an existing shortlist. It did not measure which businesses would appear in an unprompted search for an accountant. I reviewed the recommendations, explanations, citations, and stated uncertainties, distinguishing an explicit preference from a conditional shortlist or a firm left unranked because information was insufficient. A firm’s position in a comparison table was not treated as a recommendation ranking.
I also examined the types of information referenced within the responses, including company service pages, professional biographies, business directories, news coverage, customer reviews, and interactive business results. Selected source checks considered whether the information supported the accompanying claim and whether it described advertised services, attributed experience, historical background, or reported customer experiences.
Research scope and limitations
This was a small exploratory study conducted in October 2026, covering three firms, one market, three customer scenarios, and 27 responses. Each exact prompt was tested once. The findings document the captured comparisons and should not be interpreted as stable rankings or representative platform-wide preferences. The study did not measure actual service quality, market-wide recommendation rates, or the effect of changing website content.
Source checks were selective, and some pages were inaccessible. The captured citations show information presented in support of the answers, rather than every source a system may have considered. I did not independently verify every business claim, and the source review does not establish which information caused a recommendation.
Construction experience produced the clearest agreement
All nine responses in the construction scenario favored Bollus Lynch, regardless of where its name appeared in the prompt. The recurring explanation was its documented construction focus. Its construction page identifies contractors and developers among its audiences and describes industry-related accounting, tax, audit, and advisory experience.
The page provides a reasonable basis for considering the firm for an interview about construction accounting. However, some responses extended that information into more specific capabilities, inferring familiarity with construction accounting software, work-in-progress reporting, or superior project-accounting support. The cited industry page did not establish those particular services or demonstrate better delivery than the alternatives.
One Gemini answer described the firm as structurally equipped for work-in-progress schedules and bonding requirements. Another predicted greater familiarity with construction software. Those may be possibilities to investigate, but the public industry statement did not demonstrate them.
Several responses handled the distinction more carefully, recommending Bollus Lynch as the strongest candidate to interview while acknowledging that job-costing and project-reporting depth remained unconfirmed. These answers connected the documented industry focus to the customer’s requirements while leaving the specific capabilities for further verification.
A different priority changed the comparison
When the priority changed to financial interpretation, cash flow, and decision support, Masiello received greater consideration from ChatGPT and Gemini. Its financial management and CFO page advertises outsourced controller and CFO services, cash management, financial projections, and financial statements. Those offerings aligned with the revised customer need.
Claude continued to favor Bollus Lynch across the three advisory prompts, often emphasizing construction experience and organizational depth. In some of those answers, it also acknowledged that Richard Mood’s published advisory offering closely matched the request. Its explanations gave greater weight to industry fit and scale, even when another firm’s advertised services aligned more explicitly with the customer’s stated priority.
These differences illustrate how the customer’s priority can change the way a business is evaluated. Masiello received stronger consideration for financial guidance without becoming the preferred option for construction specialization. Testing only a general comparison of the three firms would have provided a less complete picture of how each was positioned for different needs.
The description of a business could also change
Richard Mood’s representation varied substantially. Some responses described a largely tax-focused practice. Others recognized a broader advisory offering, including part-time CFO services and cash-flow support.
The accessible business-services page advertises outsourced accounting, part-time CFO support, cash-flow and profitability assistance, and help with business decisions. Those are relevant to the advisory scenario, even though they do not establish construction expertise or successful delivery.
Claude provided a particularly useful example of this variation. Some responses referenced the sparse moodcpa.com website and third-party listings, describing a practice focused largely on tax and small-business services. Other responses cited business-advisory and cash-flow pages on moodcpa.net and described a much broader offering. In the advisory scenario, Claude sometimes identified Mood’s published services as the closest match to the customer’s priority, even while preferring Bollus Lynch overall.
Some Gemini responses also cited Mood’s tax pages while emphasizing tax and compliance. Taken together, these examples show differences in which parts of the business were represented in the answers. They do not establish every page a platform accessed or prove that a particular source caused its recommendation.
This matters because an incomplete description can change the apparent differences between providers. When one firm is presented as offering CFO support and another is described primarily through tax services, a customer may not realize that both advertise relevant advisory capabilities. An answer can therefore underrepresent a business without making an explicit statement that the service is unavailable.
Dependability was harder to establish than recommend
The service scenario asked about switching accountants, communication expectations, and responsibilities for ongoing support. Many answers acknowledged that they could not establish actual onboarding arrangements, response commitments, or backup coverage. Nevertheless, some used firm size to predict dependability.
Larger teams were associated with continuity and capacity. Smaller practices were associated with direct access and personal attention. Those characteristics suggest useful interview questions, but they do not establish how a particular account will be handled.
Gemini provided especially clear examples. One answer predicted that staffing would prevent turnover from disrupting service. Another suggested that a boutique relationship would provide same-day answers. Neither captured response established a firm-specific commitment supporting that assurance.
The answers often supplied useful questions for consultations, including who would handle the work, who would provide backup, and how records would transfer. This guidance helped identify what the customer needed to establish before choosing a firm. However, the verification questions did not substantiate the earlier predictions about which provider would be more dependable.
Links did not always close the evidence gap
The platforms referenced different combinations of company content, professional experience, third-party background information, and customer feedback. These sources helped explain how the firms were presented, but each supported a different part of the comparison. A source describing a service offering provided different evidence from a biography, an older news report, or a customer review.
ChatGPT referenced professional experience and customer reviews
One ChatGPT construction response included partner biographies, adding more specific information about professional experience to the firm’s general industry positioning. For example, the company-published Michael Bollus biography identifies construction audit experience. That supports an attributed statement about an individual’s background, although it does not establish who would handle a new account or whether the assigned team provides recurring job-cost reporting.
In the service-relationship scenario, another ChatGPT response cited a Birdeye listing for Richard Mood. The inspected listing contained both favorable feedback and older complaints, introducing reported customer experiences into a comparison that otherwise relied heavily on advertised services and firm characteristics.
The source check confirmed that the listing contained material consistent with the response, but did not independently substantiate the complaints or establish current service quality. The original Google review records were not independently checked, and equivalent review searches across all three firms were not completed. This was therefore an example of customer feedback appearing in the comparison, without a sufficient basis for a review-based quality ranking.
Claude combined company information with historical coverage and directories
Claude referenced a wider mix of background sources, including Worcester Business Journal reporting, staffing estimates from directories, and Cause IQ nonprofit-client records. Its citations to merger coverage supplied historical context about Bollus Lynch’s scale. Some responses used that context when explaining a preference for organizational depth and continuity.
However, a reported headcount from an earlier merger does not establish current staffing, available capacity, or the coverage a particular client would receive. Directory estimates for the smaller firms also conflicted, making them an uncertain basis for comparisons of resources. Claude acknowledged several of these inconsistencies, although some recommendations still relied on assumptions about scale and service risk.
The Cause IQ profile for Richard Mood supplied another example of evidence with a limited scope. Claude described a small set of nonprofit clients from that source. Those records could provide information about the clients identified, but could not establish the practice’s full client base or demonstrate whether it serves construction businesses of the requested size.
Gemini presented both source links and interactive business information
Eight of Gemini’s nine responses contained inline links, all pointing to the firms’ own websites. Across those responses, I recorded 31 URL occurrences, including repeated pages and links to different passages on the same page. These occurrences do not represent 31 independent sources or 31 verified claims.
First-party pages are useful evidence of what a firm advertises. A service listing can support a statement that the firm offers CFO assistance, and an industry page can support an attributed claim of construction experience. Neither automatically supports comparative excellence, successful delivery, or dependable communication.
Citation groups also sometimes followed paragraphs covering multiple firms or several distinct assertions. Their presence did not establish that each linked page supported every statement. Evaluating the recommendation required looking at the claim and the source together.
One Gemini advisory response used interactive business and map cards instead of inline claim links. The captured interface displayed ratings and a business panel containing review excerpts and Google Maps controls. That added another kind of information to the comparison experience, but it did not reveal whether the reviews influenced the generated recommendation. Nor did a displayed rating establish financial-advisory competence.
Across the three platforms, the useful distinction was between what a source documented and what the answer concluded from it. A biography could support professional experience, a service page could establish an advertised offering, and news coverage could provide historical context. None automatically established better delivery, current capacity, or a more dependable relationship for the prospective customer.
Changing name order did not produce a simple position advantage
Order was part of the study because the position of a named candidate might affect how a comparison is presented. Using A for Bollus Lynch, B for Richard Mood, and C for Masiello, I tested ABC, BCA, and CBA.
Bollus Lynch remained the preferred construction candidate whether it appeared first or last. Other aspects varied across the differently ordered prompts, including the treatment of alternatives, cited material, and whether the answer selected one lead or offered conditional choices. Gemini’s service responses, for example, included both a preferred firm and an answer assigning different firms to different needs without an overall winner.
These observations do not isolate an order effect. Each exact prompt was run once, and the three arrangements were not balanced across every firm’s position. Retrieval differences, timing, and ordinary response variation could also contribute. The results establish neither a general first-position advantage nor an absence of order sensitivity.
What the findings mean for professional-service businesses
For professional-service businesses, the findings suggest value in examining how public information addresses the situations prospective customers bring to a comparison. A broad advisory label provides less information than an explanation of the decisions supported, the reports delivered, the customers served, and the responsibilities included in an engagement.
For an accounting firm, that could mean explaining whether construction support includes job-cost reporting, what recurring financial advice involves, and how a new client transitions from another provider. Relevant client examples, anonymized deliverables, and descriptions of the people responsible for the work can give customers a more specific basis for evaluation. The study did not test whether publishing this information would increase recommendations, but it identified questions that general service descriptions left unanswered.
When reviewing an AI comparison, I would examine three things together: whether it accurately describes the business, whether its preference addresses the customer’s stated priority, and whether the cited evidence supports the strength of its conclusion. An appearance in the answer or a favorable ranking alone cannot answer those questions.
How AI comparisons shape the apparent choice
The accounting-firm test showed how AI platforms can interpret the same shortlist differently depending on what a customer needs. Construction experience produced a consistent lead across the captured responses, while the financial-advice scenario brought different service offerings into consideration. The answers also represented different parts of the firms’ public records, including service descriptions, professional experience, historical coverage, and customer feedback. Those differences affected the information available to the reader, although the study could not establish which sources caused individual recommendations.
For businesses, this makes the explanation accompanying a recommendation worth examining alongside the recommendation itself. Relevant capabilities can be understated when an answer presents only part of the service offering, while general claims about experience or scale can be extended into unsupported predictions about performance. The useful question is whether the description accurately represents the business and gives a prospective customer a sound basis for deciding which providers deserve further consideration and what still needs to be verified.