Every AI Recommends a Different Plumber

A Texas marketing firm ran an experiment earlier this year. They typed the same query into four AI platforms: find me an AC repair company nearby. ChatGPT, Perplexity, Gemini, and Claude each returned a list of recommended contractors.

Zero overlap. Not a single company appeared on more than one list.

This is not a bug. It is the logical consequence of a broken data layer.

What each AI actually does

Each platform assembles its answer from different sources, with different weighting, using different logic. The divergence is predictable once you see how each one works.

ChatGPT leans heavily on review platforms. Angi, Yelp, BBB, HomeAdvisor. It pulls star ratings, review volume, and directory descriptions. The result is a list that looks a lot like what you would have found on Google five years ago, tilted toward businesses that invested in review generation. OpenAI recently embedded Thumbtack directly into ChatGPT for home services queries, which means marketplace listing data is becoming the default structured source. The businesses ChatGPT recommends are the ones that paid for visibility on the platforms ChatGPT happens to index.

Perplexity rewards well-built websites. In one test, it cited 17 sources for a single AC repair query. It is pulling from blog content, service pages, and technical documentation more than any other platform. If your website has detailed service descriptions, geographic coverage pages, and structured content that answers specific questions, Perplexity will find you. If your website is a five-page brochure with a phone number, it will not.

Gemini pulls from unexpected places. Facebook business pages. Nextdoor recommendations. Community forums. Sources the other platforms ignore entirely. When Gemini replaces Google Assistant on 3.3 billion Android devices later this year, these community signals will drive a massive volume of service business discovery. The contractor who has never thought about their Nextdoor presence is about to have a visibility problem.

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

Four platforms. Four completely different recommendation methodologies. Four different lists of contractors. The homeowner asking the question has no idea this divergence exists.

Why the lists diverge

The surface explanation is that each AI uses different data sources. That is true but incomplete.

The deeper reason is that none of these platforms have access to the data that would actually answer the question.

When a homeowner asks "who is the best HVAC company in Dallas," they are really asking: who does reliable work, at a fair price, in my area, consistently?

The data that answers that question is job volume, completion rate, repeat customer behavior, service area coverage, revenue consistency, average ticket size. Operational performance over time.

That data exists. It sits inside QuickBooks, ServiceTitan, Jobber, and the other systems these businesses use to run their operations every day. But it has never been extracted, normalized, and published in a format AI systems can read.

So each AI does the best it can with the proxies available to it. ChatGPT uses reviews. Perplexity uses website content. Gemini uses community chatter. Each proxy captures a different sliver of the truth, and none of them capture the whole thing. The divergence across platforms is not a technology problem. It is a data availability problem.

The Thumbtack problem

In October 2025, OpenAI announced a partnership to embed Thumbtack's marketplace directly into ChatGPT. When you ask ChatGPT for a plumber, it now pulls from Thumbtack's network of 300,000+ service providers. On the surface, this looks like progress. Structured data from a real marketplace, surfaced directly in the AI interface.

Look closer at what Thumbtack's data actually contains. Response time to leads. Project type matching. Customer reviews collected through the platform. Price estimates. Availability.

What it does not contain: how many jobs that plumber completed last year. Their repeat customer rate. Their service area density. Their revenue consistency. Whether they have been in continuous operation for two years or twelve. Thumbtack knows who paid for a listing and responded to leads. It does not know who actually does good work, consistently, at scale.

Marketplace data is a different category of information than operational data. It measures participation in a marketplace, not performance in a business. A plumber who is great at converting Thumbtack leads may or may not be the plumber who has completed 2,400 jobs across 18 zip codes with a 62% repeat customer rate. Thumbtack cannot tell you. Neither can Angi, HomeAdvisor, or any other marketplace.

When ChatGPT treats marketplace data as ground truth for contractor recommendations, it is making an assumption the data does not support.

What the numbers say

The scale of this problem is not small.

ChatGPT currently recommends just 1.2% of all local business locations. Gemini recommends 11%. The vast majority of service businesses are invisible to AI recommendation systems entirely.

Meanwhile, 45% of consumers now use AI tools to find local services, up from 6% a year ago. And 42% of those consumers trust AI recommendations as much as traditional online reviews. The traffic that does come through AI search converts at 4 to 23 times the rate of traditional search traffic.

The channel is real. The trust is building. And the data underlying the recommendations is weak.

What consistent recommendations would require

If all four AI platforms had access to the same verified operational data for the same business, their recommendations would start to converge. Not perfectly, because each platform still applies its own ranking logic. But the inputs would be the same, and the outputs would be far more consistent than they are today.

Consistent recommendations require a common data layer. Structured, machine-readable, verified against systems of record, and published at a canonical URL that any AI system can retrieve. Not marketing copy. Not review aggregation. Not marketplace participation metrics. Actual operational performance data, computed from authenticated sources, that no business can edit, override, or selectively exclude.

This is what a TrustRecord is. It connects to the systems where operational truth already lives (QuickBooks, ServiceTitan, the accounting and dispatch software a business already uses), extracts the data through authenticated read-only connections, normalizes it across platforms, and publishes it as a structured record in three layers: server-rendered HTML, JSON-LD, and canonical JSON.

The business with a TrustRecord does not need to optimize separately for ChatGPT, Perplexity, Gemini, and Claude. The data is published once, in a format all of them can read. The divergence problem disappears when the data layer is the same.

The real question

Every homeowner who asks an AI for a contractor recommendation is placing a bet. They are betting that the AI has access to enough reliable information to give a good answer. Right now, that bet is not well-placed. The AI is guessing from fragments, and different AIs guess differently.

The businesses that will be consistently recommended across platforms are the ones whose operational reality is legible to machines. Not the ones with the most reviews. Not the ones with the best websites. Not the ones that paid for the right marketplace listing. The ones whose verified performance data is published where every AI system can find it, read it, and act on it.

That is the gap TrueSignal closes.