How to Tell If AI Is Sending You Customers

You've heard that AI assistants are starting to recommend businesses like yours. Maybe a customer even mentioned they found you through ChatGPT. The natural next question is the practical one: is it actually happening to you, and how would you know? Some of it leaves a trace you can measure. A lot of it doesn't. This is how to see what's visible, ask about what isn't, and read the trend without fooling yourself.

Is AI actually sending me customers?

Probably a little, and more each month, but less than the headlines suggest. AI referral traffic is real and growing fast. It's also still small next to Google. Both are true at once, and holding both in mind is the key to reading your own numbers honestly.

The scale looks like this. AI tools sent over 1.13 billion referral visits to the top 1,000 websites in June 2025, up 357% from a year earlier, with ChatGPT responsible for more than 80% of them (TechCrunch). That's fast growth. But in that same month Google Search sent those sites about 191 billion visits (TechCrunch), roughly 170 times more. So if you're waiting for AI to become a flood of clicks, you'll be waiting a while. What's changing faster is influence you can't count: people who ask an assistant, read the answer, and walk in already knowing your name.

How do I see AI referral traffic in my analytics?

Start with the traffic you can actually attribute. When someone clicks a link inside an AI answer, most tools pass along where they came from, and that shows up in your website analytics as a referral from the AI's web address. This is the clearest, cheapest AI search analytics you have, and you already own the tool.

Look for AI hostnames in your referral report

In Google Analytics 4, open Reports, then Acquisition, then Traffic acquisition, and look at the session source. In most analytics tools there's an equivalent 'referrals' or 'sources' view. Scan it for the web addresses AI assistants use:

• chatgpt.com and openai.com — by far the biggest AI referrer for most sites.
• perplexity.ai — the answer engine that cites its sources on nearly every reply.
• gemini.google.com and copilot.microsoft.com — Google's and Microsoft's assistants.
• claude.ai and others — smaller today, worth watching as the mix shifts.

If you only have time to watch one, watch ChatGPT. In one study of 6.77 million AI-referred sessions across 166 analytics properties, ChatGPT accounted for 92.4% of trackable AI referral traffic (Search Engine Land). That doesn't mean the others don't matter for your business. It means chatgpt.com is where the visible signal is strongest, so it's the first line to look for.

Group them into one 'AI' channel so you can track the trend

Checking hostnames one by one gets old fast. In GA4 you can build a custom channel group, or a saved filter, that matches all those addresses at once, so 'AI' sits next to Organic Search and Direct in your reports. Then you're not hunting, you're watching a line on a chart. The number itself will be small. The direction it moves, month over month, is the useful part.

Why does so much AI traffic show up as 'direct'?

Because a lot of it arrives with no return address. AI platforms don't always pass referrer information, so many clicks from an assistant land in your 'direct' bucket instead of a tidy 'chatgpt.com' row (Semrush). Direct is where analytics files anything it can't source: typed-in URLs, some app clicks, and, increasingly, AI. So your neat AI referral line almost certainly undercounts the real thing.

This is why you don't treat the referral number as the whole story. Two supporting signals help. Watch whether 'direct' traffic is quietly climbing, especially to the specific pages you'd expect an assistant to send people to. And watch branded search, people Googling your actual name, which tends to rise when something is putting your name in front of them, an AI answer included. Neither proves AI on its own. Together with your referral line, they sketch the shape of it.

What does 'how did you hear about us?' actually tell me?

It fills in what your analytics can't see. The most reliable way to learn how customers found you is still to ask them, because it catches the invisible paths: the ChatGPT answer with no click, the friend who asked an assistant on their behalf, the recommendation that never touched your website at all.

Make it one easy question at the moment they become a customer, on your booking form, at checkout, or in your first reply:

1. Keep it open, not multiple choice. 'How did you hear about us?' surfaces answers you'd never have listed, like 'I asked ChatGPT for a plumber near me.'
2. Ask once, at the right moment, right when they book or buy, not in a survey three weeks later that they'll ignore.
3. Write down the exact words. A tally of 'ChatGPT,' 'Google,' 'a friend,' and 'saw your Instagram' over a month tells you more than any dashboard about where trust is forming.

It's low-tech and a little imperfect, since people misremember. But it's the only method that captures zero-click AI influence at all. If 'I asked an AI' starts showing up in those answers, that's your signal, no analytics required.

Can I tell if AI bots are reading my website?

Yes, and it's worth a look, because before an assistant can recommend you it has to be able to read you. AI companies send crawlers to fetch web pages, and they identify themselves in your server logs by name: GPTBot and OAI-SearchBot from OpenAI, PerplexityBot, Google-Extended, ClaudeBot, and others. If you can see your traffic logs, and most hosts and site platforms expose them, searching for those names tells you whether AI is at least visiting.

Two honest caveats. Being crawled is not the same as being cited: a bot reading your page doesn't guarantee an assistant will name you. And if you see no AI bots at all, that's a red flag worth fixing, because a page nothing can read is a page nothing can recommend. That's usually a mechanical problem, and it's the reason to understand why AI can't find your business before you blame the algorithm.

How much of AI's impact is just invisible?

A lot of it, and it helps to say that plainly rather than chase a number that doesn't exist. The whole design of AI answers, and of Google's own AI Overviews, is to answer the question on the spot so the person doesn't have to click. That's great for the reader and awkward for measurement.

The trend is clear even outside AI referrals. In the first four months of 2026, 68% of Google searches ended without any click to the open web (SparkToro), as AI summaries answered more questions in place. Add assistants that often send no click at all, and a real share of AI's influence on your business, someone hearing your name in an answer and remembering it, will never appear in any analytics tool. That's not a failure of your tracking. It's the nature of the channel.

What should I do with what I learn?

Use the visible signal to double down, not to chase a perfect count. If a particular page keeps showing up as the entry point for AI referrals, or its direct traffic keeps climbing, that page is doing something right: it answers a real question clearly, and you should write more like it. If 'I asked an AI' starts appearing in your 'how did you hear about us' answers, that's confirmation the work of being findable is paying off.

The action is the same one that got you cited in the first place: publish clear pages that answer the exact questions your customers ask, and earn honest mentions on sites AI already reads. That's the substance behind showing up in AI search and getting recommended by ChatGPT. Measurement tells you it's working; it doesn't do the work.

That steady publishing is the gap Laspi's WordPress plugin is built to fill: it writes SEO- and GEO-shaped articles for your site, grounded in your real business, so the answers your customers ask an assistant actually exist on pages a model can read and cite. You still decide what's true about your business. It just makes sure that truth is written down often enough to be found.

So track what you can, chatgpt.com and its peers in your referrals, an honest 'how did you hear about us,' AI bots in your logs, and read the trend instead of demanding a decimal point. Being found by AI looks less like a spike on a chart and more like a quiet, steady rise in people who arrive already knowing who you are. That's the number that matters, even when you can't fully see it.