How did you hear about us? The drop-down is the problem.

Why don't my website stats tell me where my enquiries come from?

Because your website is no longer told where visitors came from. AI assistants and phone apps strip that information out before it reaches you. So the visit turns up with no source, and your stats file it under "Direct". Direct doesn't mean someone typed your address. It mostly means nobody knows.

Open your website stats and look at how much of last quarter's traffic is listed as Direct. For most businesses that number has been climbing for two years. The nice explanation is that more people know your name now.

That isn't it. The internet stopped telling your stats where people came from.

There's a fix. It costs nothing, it takes ten minutes, and almost everyone who does it ruins it the same way. Six questions sort it out, and each one answers the objection the last one raises.

What changed

When someone clicks a link, their browser normally passes on a note saying which page they came from. Your stats read that note. It's the only thing they read.

An AI assistant answering a question inside its own app doesn't pass on a note. Neither does a link opened inside a phone app. So the visit turns up with nothing attached, and a visit with nothing attached looks exactly like someone who typed your web address in by hand.

Google added a line for AI traffic in May 2026, which helps a bit. It still doesn't count Google's own AI answers. And it can't bring back a visit that arrived with nothing attached.

So more and more of your enquiries have no traceable source. It's happening fastest with the people who arrive most ready to buy.

Which leaves one thing that still works.

1. Should I just ask people how they found me?

Yes. Asking is the only way to find out about the things that happen away from your website. A friend recommending you. A comment in a group chat. An AI naming you in an answer. No tracking code can see any of that, whatever you spend. But a person can just tell you.

It does sound unscientific. It is. You can't check whether the answer is true.

Then look at the alternative.

Watch the software get it wrong, in a way you can prove

An agency put an open box on a client's enquiry form and left it there for a year. 314 enquiries came in.

The website stats put 88% of them down to search and to Direct, and just under 4% down to paid ads in search.

The client doesn't run any.

That's the useful bit. The 88% is only vague. The 4% is a confident number, with a decimal point, about money that was never spent. Meanwhile 43% of those people wrote in that a client or an industry association had recommended them. Software has no way of seeing that at all.

That's an agency reporting on its own client, so treat the exact figures as a rough guide. What matters is the pattern, and the pattern isn't in doubt. A vague answer about something that really happened beats a precise answer about something the software never saw.

2. Should it be an open box or a list of options?

An open box. A list only offers the answers you thought of when you built it, and plenty of people just pick the first one. Lists used to win because counting written answers took hours. That stopped being true, so the reason they won has gone.

Watch a drop-down give you a wrong answer

One company looked at a client's answers and found that 47% of the people who picked the first option on the list had picked the wrong one. People choose the top item because it's quick, and the form can't tell the difference.

Now add a problem that didn't exist when most of these lists were written. A list built in 2023 offers Google, Facebook, "a friend" and "other". There's no option for an AI assistant.

So a physio clinic gets a new patient who asked ChatGPT for a knee specialist nearby. On the form that becomes "other". Or it becomes "Google", because the patient half-remembers searching for something.

Give the same person an empty box and they write what actually happened. That sentence, in their own words, is the most useful thing you will ever learn about how people find you. A list throws it away.

Why lists used to win

Never because they were more accurate. Because written answers are a pain to add up. Three hundred replies, forty ways of spelling "someone recommended you", and somebody sorting them by hand at the end of the quarter.

That job now takes a computer about a minute. The reason lists won has gone, and most advice on this hasn't caught up.

Put one line on your enquiry form or your quote form. Don't put it on a newsletter sign-up, where the person hasn't decided anything yet.

An electrician, an employment lawyer and a commercial cleaning company will all get different sentences out of that box. None of them could have written the others' list.

3. What if they only remember the last thing?

They usually will, and that's a real limit. The fix is to ask two questions instead of one. "How did you first hear about us" and "what made you get in touch today" are different questions with different answers, and the gap between them is the interesting part.

Watch two things get squashed into one

A cleaning company advertises in a local business directory. Someone reads the ad in March and does nothing. In June their current cleaner lets them down. They ask an AI assistant who does contract cleaning nearby, your name comes up, and they get in touch.

Ask them how they heard about you and they write: ChatGPT.

So the ad got nothing? Try taking each piece away. Take away the ad and your name means nothing to them when it appears, so they keep looking. Take away the AI and the interest you paid to create goes to whoever does show up.

You needed both. Neither would have worked alone. "Which one gets the credit" has no answer, because credit isn't a real thing. It's a way of splitting a budget between departments in a big company. That isn't your problem.

And notice what your stats say about that same enquiry. Not the ad. Not the AI. Direct.

Ask two questions

❌ BEFORE "How did you hear about us?"

One box, one answer, and the answer will be whatever happened most recently.

✅ AFTER "How did you first hear about us?" "What made you get in touch today?"

The same ten seconds of their time. The first box gets the jerseys. The second gets the burst hot water. When the same pair keeps showing up you've learned something that counting sources would have destroyed. "Jerseys, then Google", eleven times over, is a finding. "Google: 11" is a number.

There's one more way to see the offline half, and it doesn't rely on anyone's memory. People who see an ad, a van or a sign don't click anything. They go and type your business name into Google later. So if searches for your own name are rising in Search Console, that's the print ad leaving a mark.

4. Should people have to answer it?

No. Making it compulsory puts a hurdle in front of someone who has just decided to contact you, and anyone without an answer ready will type anything to get past it. Leave it optional. You get fewer answers and better ones. If you want more answers, ask again later instead.

Two ways to lose

Leave it optional and about three in ten people skip it. Make it compulsory and the answers keep coming, but you've put an obstacle in front of the submit button on the one form that matters. Some of those answers will be rubbish, typed by someone who wanted to get on with their day.

Losing an enquiry costs more than losing a piece of data. That settles it.

The worry about getting too few answers is still fair, and there's a better fix than forcing it: ask again once the pressure is off. In the confirmation email. On the first phone call. While you're writing the quote. By then the person is relaxed and already committed, and you get a sentence instead of a word. An accountant has that moment on the welcome call. A builder has it while pricing the job. Neither needs a form.

5. What if the answers don't match my stats?

Don't pick one. Your stats show the last link someone clicked. People tell you what convinced them. Those are two different things, so it's normal for them to disagree. When they do, it usually means something is working that your stats were never able to see.

Read the gap, not the winner

If your stats say a page is quiet, but three people this month mentioned reading it, the page isn't quiet. It just isn't being counted. A business that trusts only the dashboard is about to cut the thing that's working, for the wrong reason.

So keep three answers next to each other for every enquiry: what you would have guessed, what your stats say, and what the customer wrote. The ones where all three differ are a list of the marketing you're doing that nothing can measure. No software will ever hand you that list.

6. What about the people who never fill in a form?

Plenty of people ring you, walk in, or send a message without ever touching your website, and none of them fill in a form. If you only ask online, you're only hearing from the customers who happen to type instead of call. In some businesses that's the smaller half.

In a trade or a clinic, that's most of the work.

Where the question has to live

The form is one place to ask. The others are the phone, the front desk, the quote visit and the message. Every one of those is a moment when one person is already talking to another, which makes them better places to ask, not worse.

The problem isn't the customer. It's that nobody writes it down. A receptionist who asks forty times a month and records none of it has produced nothing.

So it needs somewhere to go. A column on the job sheet. A box on the booking form. A note on the customer's record. Anywhere you'll read it later. Asking is free. Writing it down is the work.

And if you're spending real money on something printed, put a phone number or a web address on it that appears nowhere else. Then that flyer answers its own question, forever, and nobody has to remember anything.

Sources: Google Analytics Help Centre, What's new (13 May 2026) — the new line for AI traffic and what it leaves out; the list of AI tools it recognises has already changed once since launch, so check it before relying on it. Medium Giant, Making the case for self-reported attribution (2025) — the year-long test on 314 enquiries, including the paid-search figure for a client who runs no paid ads. Ruler Analytics — the 47% first-option finding, published by a company that sells tracking software. Worth saying plainly: almost every number in this field is published by someone selling either tracking software or tracking services, and I couldn't verify a single independent one. Where a figure couldn't be separated from the company selling it, I've said so above rather than quietly leaning on it.

Every enquiry happened for a reason.

Your stats can see fewer of those reasons every year, and they lose the good ones first. The recommendation. The AI answer. The ad someone remembered three months later. No amount of money spent on tracking brings those back, because the information is gone before your website loads.

So stop asking who gets the credit and start asking what would break if you stopped. One open line on a form, one question on a call, and somewhere to write the answers down will teach you more than any system built on what's left.

A dashboard nobody believes is worse than three numbers that are true.

All of it is a build job.

I do this for a living →

— From the Off your plate Desk