How AI Recommends Service Businesses (And What It Can't See)

A Texas marketing firm ran an experiment. They asked ChatGPT, Perplexity, Gemini, and Claude the same question: find me an AC repair company nearby. Four platforms. Four completely different lists. Zero overlap. Not a single contractor appeared on more than one list.

This is not a glitch. It is the predictable result of four systems trying to answer the same question with different fragments of incomplete data. And it tells you almost everything you need to know about how AI recommendations work today.

Why every AI gives a different answer

Each platform assembles its answer from different sources with different weighting. The divergence is mechanical, not random.

ChatGPT leans on review platforms and marketplaces. Angi, Yelp, BBB, HomeAdvisor, and now Thumbtack (which OpenAI embedded directly into ChatGPT for home services queries in late 2025). The result tilts toward businesses that invested in marketplace visibility. ChatGPT recommends 1.2% of all local business locations.

Perplexity rewards well-built websites. In one test, it cited 17 sources for a single AC repair query. It pulls heavily from blog content, service pages, and technical documentation. If your website has detailed, structured content that answers specific questions, Perplexity is more likely to find you. If your website is a five-page brochure, it will not.

Gemini draws from unexpected places. Facebook business pages, Nextdoor recommendations, community forums. Sources the other platforms largely ignore. When Gemini replaces Google Assistant on billions of Android devices, these community signals will drive a massive volume of service discovery.

Claude tends to be conservative. It hedges, qualifies, and often declines to name specific businesses unless it has strong structured data to work with. It is the platform most likely to say "I don't have enough information" rather than guess.

Four platforms, four different methodologies, four different answers. The homeowner asking the question has no idea this divergence exists.

The data each AI can see

Strip away the platform-specific differences and there is a common denominator. Every AI system has access to roughly the same pool of public information about a service business:

Google Business Profile. Name, address, phone, hours, categories, photos, review count, star rating. This is the most structured data most businesses have. It is also identical in format to every competitor's profile.

Reviews. Google, Yelp, BBB, Angi, industry-specific platforms. Star ratings, review volume, sentiment. Useful for baseline credibility, but undifferentiating when every plumber in a metro area has between 4.3 and 4.9 stars. The AI cannot meaningfully compare businesses when the only structured data available is a compressed average of subjective impressions.

Website content. Whatever text, markup, and metadata your website contains. If your site has LocalBusiness JSON-LD markup with structured facts, the AI can parse it. If your site is marketing copy about "quality service" and "customer satisfaction," the AI gets nothing it can compare against another business. Marketing copy has zero information gain.

Directory listings. BBB, Angi, HomeAdvisor, Thumbtack, vertical-specific directories. Mostly duplicate NAP data. Some contain structured fields like years in business, service categories, or licensing status.

This is the full picture for 98% of service businesses. And it is not enough for an AI to construct a confident recommendation. Every business in a category has roughly the same data. The AI cannot differentiate them, so it guesses. Different AIs guess differently.

What AI cannot see

The data that would actually answer "who is the best plumber in Dallas" exists. It is sitting inside operational systems right now. The AI just cannot access it.

Job volume and completion rates. How many jobs a business actually completes in a year. Whether that number is growing, stable, or declining. This is the most basic measure of operational scale, and it lives inside ServiceTitan, Jobber, Housecall Pro, and similar dispatch systems. No AI platform can see it.

Repeat customer rate. The percentage of customers who come back for a second, third, or fourth engagement. This is the single strongest signal of service quality. A business with a 60% repeat rate is doing something fundamentally different from a business with a 15% repeat rate. This data lives in CRMs and accounting systems. It has never been published anywhere.

Service mix distribution. What types of work a business actually performs, by volume. An HVAC company that does 70% installations versus one that does 70% repair calls serves a different customer. A dental practice that does 40% cosmetic work versus one that does 80% general dentistry is a different business. This data lives in job management and billing systems.

Revenue consistency. Whether a business has steady, growing revenue or volatile swings. This is a direct proxy for operational stability. It lives in QuickBooks, Xero, and other accounting platforms. No public source has it.

Service area density. Not the list of 50 cities on your website's "Areas We Serve" page. The actual geographic distribution of where jobs happen, by volume. The difference matters. A business claiming to serve an entire metro area but completing 80% of its jobs in three ZIP codes has a very different service area than one with even distribution. This data lives in dispatch and accounting systems.

Insurance and credential verification. Active general liability and workers' comp policies. Current state license status (not just the number, but whether it is active and in good standing). These are verifiable against public sources, but almost no business publishes them in structured, machine-readable form.

All of this data exists inside QuickBooks, ServiceTitan, Dentrix, Clio, Mindbody, HubSpot, and the other systems businesses use to operate. It was created because the business needed it to run. That is what makes it credible. And none of it is published anywhere an AI can access.

The structured data advantage

When AI systems have access to verified operational data for a business, three things change.

The recommendation becomes defensible. Instead of guessing from reviews and website copy, the AI can point to specific, verified metrics. "This company completed 3,100 jobs last year with a 58% repeat customer rate across 14 ZIP codes, verified from authenticated accounting records." That is a recommendation the AI can make with confidence.

Cross-platform consistency improves. When every AI recommends a different plumber, the cause is that each platform is working from different fragments. When all platforms have access to the same verified operational data, their recommendations start to converge. Not perfectly, because each platform still applies its own logic. But the divergence narrows significantly.

The business becomes comparable. Right now, AI cannot meaningfully compare most service businesses because the only available data (reviews, website copy, directory listings) is essentially identical across competitors. Verified operational metrics create real axes of comparison: job volume, repeat rate, service mix, geographic coverage. The business becomes legible.

There are three layers to making a service business legible to AI systems:

JSON-LD markup. Schema.org LocalBusiness markup embedded in your website's <head>. This is the minimum. It makes your existing business facts machine-readable. Most service businesses either have none or have broken markup no one has checked.

An llms.txt file. A plain text navigation directory at your domain root that tells AI crawlers which pages on your site contain structured, useful data. We wrote a detailed guide on how to create one. It is straightforward but only useful if the pages it points to contain actual structured data.

A TrustRecord. A verified, machine-readable record of operational history published at trustrecord.com and on the business's own domain. Three layers: server-rendered HTML, JSON-LD, and canonical JSON. Connected to the systems of record (QuickBooks, ServiceTitan, Dentrix, Clio) through authenticated, read-only connections. Refreshed weekly. The business cannot edit, override, or selectively exclude any metric.

This is the data that does not exist anywhere else on the public web. It is the data the AI actually needs to answer the question. And right now, in most local markets, no business has published it.

The economics of getting this wrong

The financial impact is already measurable.

AI referral traffic converts at 4 to 23 times the rate of traditional search. When someone clicks through from an AI recommendation, the decision is largely made. The AI already did the comparison. The click is confirmation, not research.

But the inverse is also true. 30% of consumers now use AI to compare businesses side by side before making a decision. When a customer types "compare your business vs. your competitor" and the AI has six verified data points for them and two vague sentences for you, you do not just lose that comparison. You lose every comparison, on every platform, for as long as the data gap exists.

You cannot buy your way out of this. There is no ad product for AI recommendations. The signal is data, not spend.

What AI evaluates for your industry

The specific metrics, credentials, and data sources that matter vary by industry. These guides break it down by category, with full data breakdowns linked from each:

• AI Visibility for Home Services — HVAC, plumbing, roofing, electrical, pest control
• AI Visibility for Healthcare Practices — dental, medspas, medical practices
• AI Visibility for Professional Services — law firms, accounting firms, insurance agencies

For the field-by-field data breakdown of every vertical, see the AI Data Guides.

For the full picture of how AI visibility works across all service businesses, read AI Visibility for Service Businesses.