We fact-checked what AI tells agents about lead companies

Plenty of agents now ask ChatGPT about a lead company before they ever talk to a rep. So we took ChatGPT, Perplexity, Claude and Gemini and asked every one of them the same 112 questions about 23 insurance lead companies. Nothing opinion based. Just stuff with a right answer, like who owns the company, what leads it sells and for how much, how returns work and whether there’s a contract or minimum spend. Then we sat down with the company’s own pages and checked the answers fact by fact.

How often each engine got a fact wrong

• ChatGPT2.8%2.8% of facts wrong (22 of 785 checked). 15 of 112 answers had a wrong fact.
• Perplexity7.9%7.9% of facts wrong (41 of 518 checked). 30 of 112 answers had a wrong fact.
• Gemini9.6%9.6% of facts wrong (67 of 699 checked). 40 of 112 answers had a wrong fact.
• Claude15.2%15.2% of facts wrong (102 of 669 checked). 55 of 112 answers had a wrong fact.

That rate counts anything wrong or out of date, out of the facts we could check. For all four engines together it was 8.7% of 2,671 facts. The answer count measures something else. One answer holds many facts, so it counts the answers with at least one wrong fact.

What the engines got wrong most

• 10.0%Prices. Agents care about this one most and it got missed most
• 9.4%Owners, leaders and founding years
• 8.6%Lead types and exclusive vs. shared
• 7.8%Contracts, minimum orders and deposits
• 7.6%Return and refund policies

Where the wrong facts came from

It matters a lot where the engine is getting its info. Facts it took from the lead company’s own website were wrong 3.4% of the time. If it used some other site that went up to 19%, and on blogs from competing lead companies half of it was wrong.

• 3.4%wrong when the source was the company's own site
• 19%wrong when the source was a third-party site
• 50%wrong when the source was a competing lead company

Don’t assume the company’s own site is clean either. For 9 of the 23 companies we found their own pages contradicting each other. For example an old FAQ with a different return window than the terms page they use today. If the engine lands on the old page, that’s what you get told.

Examples we checked by hand

Every one of these we looked up ourselves on the company’s own page, Oct 6 to 8, 2026.

What this means if you buy leads

1. If you use AI to find lead companies don't take everything at face value. Be sure to confirm their terms. Return windows, minimums and prices are where we saw the most mistakes.
2. Ask the engine where it got its answer. If it wasn't the lead company's own site, go look for yourself.
3. Read the current terms page before you pay anything. Skip the FAQ and the blog for this. The terms are the thing you're signing up for.
4. Let us do it for you. Just check InsuranceAgencyTools.com. See how we research and check every company.

How we did it

• Same 112 questions for every engine, up to five per company, asked once each with web search on. 20 companies got five questions and three got four (G.O.A.T. Leads, Speakeasy Leads and TTC Leads). We only ask who owns and founded a company when we have those facts on record. We ran them from the United States in English, Oct 6 to 8, 2026.
• We only graded facts that have a right answer.
• Each answer got split into single facts, and each fact went up against our research records. Anything marked wrong then got a second look from a person on the company's current pages. If the company's page agreed with the AI, that meant our record was the one that was off. We counted it as right for the AI and fixed our record.
• If we couldn't confirm a fact either way it's not in the rates. 38 times an engine just said it couldn't find the answer. We count those separately.

What comes next

We’ll run the same questions again and update this page each time so you can see how the answers change.