Why Your Analytics Are Hiding How Many Customers AI Is Sending You

Straight answer: if a customer asks ChatGPT or Perplexity for a recommendation and then types your business name straight into Google, your analytics records that as "direct" or "organic search" and never tells you an AI sent them. Most small business owners are getting customers from AI tools right now and have no idea, because the tracking was built for the old web, not this one.

The referral that doesn't look like a referral

I run my own analytics every Monday morning, coffee in hand, same routine I've had for years. Late last year I started noticing something odd in the traffic acquisition report: a source called chatgpt.com, sitting quietly in the list next to LinkedIn and email, that I hadn't seen a year before. Not huge numbers. But not zero, and climbing month on month.

Here's the bit that matters more than the raw number. That small chatgpt.com line was the tip of something much bigger sitting underneath it, invisible in the data. Because when someone asks ChatGPT "who's a good AI marketing consultant in the UK" and gets my name back, most of them don't click a link inside the chat. They open a new tab and Google "Lilach Bullock." Google Analytics logs that as organic search, or sometimes direct. It never once mentions the AI conversation that started the whole thing.

What I found when I checked my own numbers

Between March 2025 and February 2026, the chatgpt.com referral line on my own site went from around 40 sessions a month to 380. That's the visible part. When I then cross-referenced branded search terms (people typing my exact name or "Lilach Bullock AI consultant" into Google) in Search Console, that volume had grown faster than anything else on the whole site over the same period, faster than my LinkedIn traffic, faster than any paid campaign I ran. There's no clean way to prove every one of those branded searches started with an AI chat. But the timing, the phrasing people used in the search terms, and the fact I hadn't done any PR push that would explain it, all point the same way.

This is the part almost nobody writing about "AI search visibility" tells you: being cited by ChatGPT rarely shows up as a clean, countable click. It shows up as a delayed, disguised bump in branded search and direct traffic weeks later, and your dashboard just quietly folds it into "organic" like nothing happened.

Why GA4 gets this wrong

Google Analytics 4 classifies traffic by the referring domain in the browser header. When a large language model answers a question inline, in the chat window itself, there's often no referring domain at all, because the user never left the AI tool to get their answer. If they do click through, some AI tools (Perplexity is decent about this) pass a proper referrer. Others, particularly voice assistants and in-app AI answers, strip that information entirely. So you end up with three buckets, all invisible to a normal glance at your dashboard:

• Answered inline, never clicked: the user got your name and details from the AI, never visited your site at all until they were ready to buy, then searched you directly. Recorded as "direct" or "organic search."
• Clicked through but referrer stripped: shows up as "direct" traffic, no source data.
• Clicked through with a clean referrer: this is the only bucket you can see, and for most small sites it's the smallest of the three.

If you're only checking the third bucket, you are looking at the tip of an iceberg and reporting the tip as the whole thing.

The bathroom fitter in Kent

I worked with a small bathroom and wet room fitting business earlier this year, family firm, three fitters, one office manager who answers the phone and books quotes. Good work, decent reviews, completely invisible online beyond a basic website. We added one line to their quote request form: "How did you hear about us?" with a free text box, nothing fancy.

Within six weeks, 12% of new quote requests included some version of "asked ChatGPT for bathroom fitters near me" or "an AI app recommended you." Twelve percent, from a form field that cost nothing to add and took the office manager about ninety seconds to set up in their existing booking software. None of that would have shown up in their Google Analytics as anything other than direct traffic, because most of those customers phoned the number listed on the site rather than clicking through anything trackable at all.

That one form field told them more about where their business was coming from than six months of staring at a traffic dashboard ever had.

How to track this yourself, step by step

You don't need a developer or a new tool for most of this. Here's what to do this week:

• In GA4, go to Reports, Acquisition, Traffic acquisition, and add a secondary dimension for "Session source." Search the list for chatgpt, perplexity, copilot, and gemini. Note the numbers, then check the trend over the last twelve months.
• In Google Search Console, check the Performance report and filter queries that are your exact business name or close variants. A sharp, unexplained rise in branded search volume, with no PR or ad spend behind it, is often your clearest signal that AI tools are mentioning you.
• Add a "how did you hear about us" field to every contact form, quote request, and checkout on your site. Free text, not a dropdown, because customers describe it in their own words and you learn the phrasing they use.
• Train whoever answers your phone to ask new callers the same question and log the answer. This caught the Kent bathroom firm's 12% figure, and it cost nothing.
• If you sell anything with a clear value calculation, build a small interactive tool for it. AI tools love citing pages with clear, structured, calculable answers, and something like an interactive calculator gives an AI a clean, quotable data point to pull from when it's summarising an answer for someone.

Being cited isn't the same as being chosen

Here's the uncomfortable part most posts about "optimising for AI search" skip over: getting mentioned by ChatGPT is not the win everyone's selling it as. I've seen businesses get excited that an AI tool named them in an answer, only to check three months later and find zero measurable change in leads. Citation without conversion is just a vanity metric with better PR.

What moves the needle is what happens in the seven seconds after someone reads that AI answer. If your name gets mentioned alongside four competitors and your website looks dated, loads slowly, or buries your phone number three clicks deep, the AI did its job and you still lost the customer. This is where behaviour data earns its keep. Watching real recordings of what people do on your site (Hotjar is the tool most people know for this, and I've written before about how Hotjar built its whole business around exactly this kind of behavioural insight) will show you whether AI-referred visitors bounce in three seconds or read your pricing page. Most small business owners have never once checked.

The other uncomfortable truth: you cannot fully control what an AI says about you. You can influence it, by having clear, structured, factual content about your pricing, your process, your location and your reviews, but you can't buy your way into a better answer the way you can buy your way up a Google Ads auction. Anyone promising guaranteed "AI ranking" is selling you something that doesn't exist yet in any reliable form.

What helps an AI tool recommend you

A few things consistently show up in the businesses that do get mentioned well:

• Clear, specific pricing on the page itself, not "contact us for a quote." AI tools struggle to recommend a business when the price is a mystery.
• A simple, well-structured FAQ section answering the exact questions customers type into AI chats ("how much does X cost," "how long does Y take," "do you cover Z area").
• Consistent, repeated messaging across your own site, so an AI summarising you isn't stitching together contradictory claims from five different pages. Duolingo's whole content approach, which I covered in a piece on Duolingo's marketing strategy, works partly because the same simple message repeats everywhere: consistency an AI can summarise cleanly, rather than a mess of half-updated pages.
• Real, recent reviews with specific detail in them, not just star ratings. AI tools pull phrases from reviews when they can.

None of this is complicated. Most of it is unglamorous admin that gets skipped because it doesn't feel like "AI strategy." It is, quietly, the only AI strategy that's working for small businesses right now.

What to do about it this week

Don't wait for a perfect tracking solution, because one doesn't fully exist yet even for large companies with data teams. Do the cheap, fast things instead:

• Add the "how did you hear about us" free text field today, on every form you have.
• Check your branded search trend in Search Console right now and note this month's number so you have a baseline.
• Ask your last ten customers directly, by phone or email, whether they used an AI tool at any point before finding you. You will be surprised.
• Automate the alert so you see it when it happens. IFTTT and similar automation tools can ping you the moment a form submission mentions "ChatGPT" or "AI" in the free text field, which is the kind of small, unglamorous automation I talked through in this look at IFTTT's marketing approach, so you don't have to remember to check.
• Rewrite one page this week, your pricing or services page, so it answers the three questions customers ask most, in plain sentences an AI could lift directly.

If all of this feels like more than you have time for around running your business, this is exactly the kind of gap an outside pair of eyes closes fast. I work through this audit with clients regularly as part of hands-on AI consulting for small businesses, and it's usually a half-day job, not a six-month project.

Frequently asked questions

How do I know if ChatGPT is sending me customers?

Check GA4 for a source called chatgpt.com or similar under Traffic acquisition, then cross-check Google Search Console for an unexplained rise in searches for your exact business name. Add a free text "how did you hear about us" field to your contact forms, it's the single most reliable signal and it costs nothing to set up.

Why doesn't Google Analytics show AI referrals ?

Most AI tools either answer inline without sending a click, or strip the referrer information when a user does click through, so the visit gets logged as "direct" or "organic search" instead of showing the AI tool as the source. This isn't a bug you can fix in GA4 settings, it's a structural gap in how referral tracking works for AI-driven traffic.

Can I pay to get ranked higher in AI answers?

Not reliably, no. Unlike Google Ads, there is no verified paid mechanism to guarantee placement in an AI tool's answer. What helps is clear pricing, structured FAQ content, and consistent messaging across your site, the same things that help human readers, which AI tools then find easier to summarise accurately.

Is it worth optimising my small business site for AI search at all?

Yes, but treat it as a byproduct of good, clear, honest content rather than a separate campaign. Businesses seeing real results are the ones with specific pricing, plain-English FAQs, and recent detailed reviews already in place, not businesses running a dedicated "AI SEO" push with nothing solid underneath it.

Related reading: Why AI Is Quietly Wrecking Your Email Open Rates (And How to Fix It) and Tesco Marketing Strategy: How They Built a Brand That Wins.

This builds on my main AI marketing guide, my main guide on the topic.