Where do ChatGPT mortgage leads land?
Mostly on the homepage or contact page. Across 23 broker websites over six months, 4 in 5 ChatGPT enquiries arrived there, even though about 3 in 4 of them were niche and every site already had a page for that niche. ChatGPT reads the specialist page, then links to the firm’s main website.
You’ve probably had one. A long, polite, perfectly punctuated enquiry that reads like it was written by a very organised robot. Because it was.
More of our clients are getting them. So we wanted to know two things. Where on the website do these people land? And what are they actually asking for?
What we looked at
We went through every enquiry from the last six months where the visitor came from ChatGPT or another AI assistant, across 23 mortgage and protection broker websites. Then we looked at where each person landed and what they asked for.
What we found
| What we measured | Result |
|---|---|
| Sample | Every AI-referred enquiry, 23 mortgage and protection broker websites, last 6 months |
| Landed on homepage or contact page | 4 in 5 |
| Of those, niche enquiries | About 3 in 4 (about 2 in 3 excluding a single-niche site) |
| Enquiry length vs Google | Around 6x longer |
| Sites that already had a page for that niche | All of them |
| ChatGPT visits to theleadengine.co.uk landing on our homepage | 68% (since January) |
4 in 5 of them landed on the homepage or the contact page. Not the specialist page. The front door.
You might expect those to be the vague ones. “First-time buyer, can you help?” Some were. But most weren’t. About 3 in 4 of the homepage and contact page enquiries were niche. Even when we took out a site that only covers one niche (so its homepage is really a niche page), it was still about 2 in 3.
These weren’t “tell me about mortgages” enquiries. They were things like:
| Enquiry | Matching page on the site? | Where they arrived |
|---|---|---|
| CIS contractor | Yes | Homepage or contact page |
| NHS doctor / GP partner | Yes | Homepage or contact page |
| Skilled Worker visa (one written in Ukrainian) | Yes | Homepage or contact page |
| Repaying a Help to Buy loan | Yes | Homepage or contact page |
| Self-employed, one year of accounts | Yes | Homepage or contact page |
| Leasehold flat plus self-employed income | Yes | Homepage or contact page |
| Buying after a divorce | Yes | Homepage or contact page |
| Parent and grown-up child buying together | Yes | Homepage or contact page |
They were long, too. ChatGPT enquiries were around 6x longer than Google ones, because people paste in the message ChatGPT helped them write.
Here’s the interesting bit. All the websites already had a page for exactly that niche. A doctor’s page. A visa page. The person still arrived on the homepage.
We checked our own site too. Since January, 68% of the visits ChatGPT sent to theleadengine.co.uk landed on our homepage.
Why this happens
We can’t see inside ChatGPT, so this is our best explanation rather than proof.
When someone asks an AI for help getting a mortgage on a Skilled Worker visa, it’s rarely one search. It’s a conversation: their situation, their deposit, their worries, then follow-up questions. The broker recommendation normally comes at the end of that thread, not the start. Along the way the AI reads around the web, and your specialist page is probably what convinced it you’re the right adviser. But when it finally gives a link, it often gives the firm’s name and main website. Not the page it read.
It’s a bit like a friend recommending a restaurant because of one amazing dish. They give you the address, not the menu page.
What to do about it
- Keep building niche pages. Just like Google, AI needs something specific to read before it recommends you. The niche page is still doing the heavy lifting, even when it doesn’t get the click.
- Make the contact form copy and paste friendly. A simple message box lets people paste the enquiry ChatGPT wrote for them.
What about AI visibility tools?
There are more and more tools that promise to show your “AI visibility”, like Profound, Peec AI, Otterly, Semrush’s AI Toolkit and Ahrefs Brand Radar. Most of them ask an AI a list of sample questions and record whether your name comes up.
The trouble is, AI rarely gives the same answer twice. Rand Fishkin of SparkToro ran the same brand-recommendation prompts nearly 3,000 times across ChatGPT, Claude and Google’s AI. They gave the same list of brands less than 1% of the time, and the same list in the same order less than 0.1% of the time.
So a “you rank #2 in ChatGPT” score doesn’t mean much. How often you get mentioned across lots of questions holds up better, but it’s still a sample of made-up prompts, not what your prospects actually saw. (Source: Search Engine Journal on the SparkToro study.)
What you can actually see
When ChatGPT looks something up for someone, it often sends a bot to read the page there and then. That visit shows up in your website’s server logs, or in Cloudflare if you use it, labelled “ChatGPT-User”. So you can see which of your pages ChatGPT really read, and when.
It won’t catch everything, because sometimes it answers from what it has already stored. But line those visits up against your enquiries and you’re much closer to the truth than any score.
| AI visibility tools | Server logs / Cloudflare | |
|---|---|---|
| Examples | Profound, Peec AI, Otterly, Semrush AI Toolkit, Ahrefs Brand Radar | "ChatGPT-User" bot visits |
| What it measures | Whether your name comes up for sample prompts | Which of your pages ChatGPT actually read, and when |
| Real prospect data? | No, made-up prompts | Yes, live lookups for real users |
| Repeatable? | Same brand list less than 1% of the time (SparkToro, ~3,000 runs) | Fixed record of what happened |
| Blind spot | What your prospects actually saw | Answers from stored memory, with no visit |
Not a silver bullet
AI enquiries are still a small slice of most brokers’ inbound leads. Google still sends far more. We can see which pages AI read, but not how it used them. But the pattern was clear enough across lots of different brokers that it’s worth creating content around specific scenarios.
Ask yourself: would ChatGPT, Claude or Gemini recommend your website for a specific enquiry? Do you have the content?
ChatGPT mortgage leads: FAQs
Do mortgage brokers get leads from ChatGPT?
Yes, but it's a small share for now. Across our broker clients, Google still sends far more enquiries. The ChatGPT ones tend to be longer, more detailed and niche, and most arrive through the homepage or contact page.
Is ChatGPT a form of AI mortgage lead generation?
Not in the way most AI lead tools are. You can't buy ads on ChatGPT for mortgages, and there's no Search Console for it. The leads come from ChatGPT recommending your firm, and it does that after reading in-depth pages about specific scenarios.
How does ChatGPT decide which mortgage broker to recommend?
We can't see inside it. In our data, the firms that got recommended had a page for that exact scenario, and ChatGPT tends to lean on firms with lots of reviews. Specific content plus reviews is the best bet.
How do I see ChatGPT traffic in Google Analytics?
In GA4, go to Reports, Acquisition, Traffic acquisition and look for the source chatgpt.com. Change the table to show landing page as well, and you'll see most of it hits your homepage. Your server logs or Cloudflare show "ChatGPT-User" visits too, which tell you the pages it read.
Are AI visibility scores reliable?
Not very. They're based on sample prompts, and AI rarely gives the same answer twice. SparkToro found the same list of brands came up less than 1% of the time.
Why do ChatGPT enquiries land on the homepage?
Our best explanation: ChatGPT reads your niche page to decide you're the right adviser, then gives the person your firm's main website rather than the page it read.
Watch: 60 seconds