We Asked AI to Recommend Service Businesses. Here's How Often It Got It Wrong.

We ran an experiment. We asked four AI platforms — ChatGPT, Claude, Perplexity, and Gemini — to recommend and describe service businesses across five industries: HVAC, roofing, dental, medspa, and law. Same prompt. Same 10 cities. Same day.

Then we checked their work.

We verified every recommended business against Google Places. We extracted every factual claim — founding years, addresses, certifications, owner names — and graded each one against BBB records, state licensing boards, and company websites.

The results are not encouraging.

The platforms almost never agree

Across all five verticals and 10 cities, we collected 735 unique business recommendations. Only 12.4% of them appeared on more than one platform.

Ask ChatGPT for the best law firm in Dallas and ask Perplexity the same question. You will almost certainly get two completely different lists.

This is not a rounding error. In medspa and law, fewer than 1 in 15 recommended businesses appeared on more than one platform. Which AI your customer uses determines which business they find.

Most of what AI says can't be verified

Recommendations are one problem. Descriptions are worse.

For each industry, we selected 20 well-established businesses (2 per city, chosen by Google review volume) and asked each platform to describe them in detail. We extracted every specific factual claim and graded it against verified sources.

Across 7,759 specific claims extracted from all five industries:

• 65% couldn't be verified from any public source — not confirmed, not denied, just unverifiable
• Only 24% were confirmed correct
• 311 claims were provably wrong — directly contradicted by BBB records, state licensing boards, or company websites

The platforms aren't just guessing about obscure details. They're getting basic facts wrong: founding years, owner names, addresses.

Same business, different "facts"

The most striking finding is what happens when you ask two platforms the same question about the same business.

Thompson Law Injury Lawyers, Dallas — "When were they founded?"

• Perplexity: 2017
• Gemini: 2011
• Verified answer: 2017

Tate Law Offices, Dallas — "Who is the owner?"

• Perplexity: Tim Tate
• Gemini: Stephen Tate
• Verified answer: Tim Tate

Moncrief Heating & Air Conditioning, Atlanta — "When were they founded?"

• ChatGPT: 1898
• Gemini: 1976
• Verified answer: 1898

Gemini said Moncrief was founded in 1976. The company has been in business since 1898. That is not a minor discrepancy. That is 78 years of operating history erased.

Viva Day Spa + Med Spa, Dallas — "What is their address?"

• Perplexity: 8300 Preston Rd, Suite 250, Dallas, TX 75225
• Gemini: 3223 W Mockingbird Ln, Dallas, TX 75235
• Verified answer: 8300 Preston Rd, Suite 250

Gemini gave the wrong address entirely. A potential customer following that recommendation would end up in the wrong part of Dallas.

Dentistry of Nashville — "When were they founded?"

• Perplexity: 1919
• Gemini: 2014
• Verified answer: 1919

A 105-year-old practice described as a decade old. These aren't edge cases. Across 218 fact comparisons where two or more platforms stated an answer, they disagreed 65% of the time. In law, the disagreement rate was 78%. In medspa, 89%.

The confidence problem

What makes this worse is that none of these platforms hedge. They don't say "we think" or "according to one source." They state founding years, addresses, and credentials as fact. A business owner searching their own name on ChatGPT will find a description that reads like a verified profile — but may contain invented details they've never seen before.

Claude was the exception. It declined to recommend specific businesses in most cities, citing its inability to verify current information. That is the honest answer. The other three platforms recommended confidently, with no way for users to verify those recommendations.

Why this happens

AI platforms don't have access to the data that would actually answer these questions accurately. The operational facts that define a business — when it was founded, who runs it, what services it offers, where it's located — live inside the business's own systems: accounting software, practice management platforms, CRM databases.

That data has never been extracted, structured, and published in a format AI can read. So each platform does its best with whatever fragments it can find: an outdated BBB listing, a years-old blog post, a directory entry with the wrong address. Each platform finds different fragments, synthesizes them differently, and arrives at a different (often wrong) answer.

The problem isn't that AI is broken. The problem is that AI doesn't have a verified source to draw from.

What this means for your business

If you run a service business, AI is already describing you to potential customers. The description may be accurate. It may not be. You have no way to know unless you check, and you have no way to correct it by writing a better "About Us" page.

The only thing that fixes this is structured, verified operational data — published in a format AI systems can read, from a source they can trust. Not marketing claims. Not self-asserted copy. Verified data from your own systems of record.

That is what a TrustRecord is. And these studies are why it exists.

Full methodology, platform-by-platform breakdowns, and raw data are available in the individual study pages: HVAC · Roofing · Dental · Medspa · Law