Where Did Your Customers Come From? What Cafes Can Know
An owner looks at a busy Saturday morning and asks a fair question. Where did all these people come from? The honest answer is that some of it is knowable and a good amount of it is not, and no dashboard changes that no matter how long you stare at it.
Short answer: analytics tools can trace a search click, a link with tracking added to it, and an action taken on a Google Business Profile. They cannot trace a walk in after someone noticed the sign, a text from a friend describing a shop, or an answer given by an AI assistant with no tracked link attached to it. Both categories are real. Only one of them shows up in a report.
This post is about telling those two categories apart so an owner reads their own numbers correctly, without either over trusting a dashboard or giving up on measurement altogether.
What a report can actually see
A web analytics tool works by recording a visit and trying to explain where it came from. It does that well when the visit carries a clue. A click from a Google search result carries a clue, because the browser hands the site a referring page. A click on a link that has tracking parameters added to it carries a clue, because those parameters were built specifically to be read later. An action on a Google Business Profile, like a website click or a request for directions, is recorded by Google itself and can be checked in the profile's own performance view, separate from the website's own analytics.
Search Console works the same way for organic search specifically. It reports impressions and clicks tied to actual queries typed into Google, because Google is the one handing that data back to the site owner. None of this is a guess. It is a direct record of an action that happened on a platform built to keep that record.
The common mistake is assuming that because some sources are this traceable, all of them must be. They are not, and the gap is not a bug in the tool. It is a structural limit of how the web works. A browser only tells a site what it was told to tell it, and a person who never clicked a tracked link simply never generated that information in the first place.
It is worth being specific about why the traceable sources work at all, because the mechanism matters for how much you trust the number. A search click works because Google's own results page passes a referring value along when someone clicks through. A tagged link works because someone, usually the cafe itself, built that link with extra parameters on purpose, meaning the traceability was designed in ahead of time rather than discovered after the fact. A Google Business Profile action works because the action happens inside Google's own systems, not on the cafe's website at all, so Google simply reports what it already recorded. Each of these three has a clear, explainable reason it can be traced. That is exactly what the untraceable sources lack.
What a report cannot see, and why
A person who walks past a shop, sees the sign, and comes in later never generates a digital trail at all. There is nothing to trace because nothing online happened. The same is true of a customer who hears about a shop from a friend over lunch and shows up the next day with no link, no search, and no click involved anywhere in the chain.
A more modern version of the same gap involves AI assistants. When someone asks an AI assistant a question and gets an answer that mentions a cafe, that answer may or may not include a link, and even when it does, not every assistant passes along the kind of clean referrer information a browser can read and label. The result, on the analytics side, often lands in a bucket usually labeled direct, the same bucket that also holds someone typing a web address straight into their browser because they already knew it.
This is worth spelling out because it is easy to get backwards. A large direct number is not proof of nothing. It is proof that the analytics tool could not identify a source, which is a different statement entirely. Some of that bucket really is people who already knew the address. Some of it, going forward, is likely people who arrived by a path the tool cannot label. There is no way to split that bucket apart with certainty from the data alone, and claiming otherwise would be overstating what the report shows.
It also helps to picture a single real morning to see how quickly the traceable and the untraceable get tangled together. A regular walks in because she passed the shop on her normal route, invisible to any tool. Ten minutes later, a first time visitor arrives after clicking a Google search result for coffee near a specific neighborhood, fully traceable through Search Console. An hour after that, someone orders a drink because a coworker mentioned the shop by name at their office, invisible again. By closing time, the analytics report will show one traceable visit and nothing for the other two, not because the other two did not matter, but because nothing about them was ever recorded anywhere online.
A small, dated example: hillcane.co
It helps to see this play out on an actual site rather than in the abstract, so here is one dated case, with the sample size stated plainly because it is small and should not be read as a benchmark for any other business. Hillcane's own site, hillcane.co, launched on September 9, 2026. In the first days after launch, Google began showing the site in search results, with impressions building toward roughly 124 per day by day 8. The first organic clicks, meaning someone actually clicking through from a Google search result, appeared on day 7. The first visits tagged as coming from ChatGPT appeared on days 8 to 9.
Those are the traceable pieces: search impressions and clicks that Search Console recorded directly, and a small number of AI referred visits that happened to carry a label. Separately, in the first 12 days after launch, about 2,000 of the roughly 7,738 total hits recorded on the site were attributed to AI bots crawling or fetching pages, which is a different thing entirely from a human visit and is worth naming so it does not get confused with customer traffic. It is evidence that AI systems were reading the site early, not evidence about where human visitors came from.
On the human side, Thomas has an observation worth stating as exactly that, an observation and not a study: on his other sites, direct traffic has tended to include visits that look, on closer inspection, like they originated from an AI assistant answer with no tracking attached. That pattern has not been formally measured for hillcane.co specifically in this small a window, so it is offered here as context for how to read a direct traffic number, not as a claim about hillcane.co's own direct bucket. A single new site with days of data is a case, not proof of a rate that applies elsewhere.
Why bother sharing numbers this small at all, rather than waiting for a larger dataset? Because the shape of the pattern, not the size of it, is the useful part. A brand new site with almost no history already shows search impressions arriving before organic clicks, organic clicks arriving before any AI referred visit, and AI bot activity running well ahead of any of it, all inside the first two weeks. That ordering is a real, dated fact about one specific launch, and it illustrates the mechanics this post is describing better than a hypothetical example could. It is offered for that reason, stated with its sample size every time, not as a number a cafe owner should expect to reproduce.
A one question habit that fills the gap
Since no report will ever close this gap completely, the most useful tool an owner has costs nothing and takes five seconds: ask new customers how they found the shop. Not every customer, and not as an interrogation, just a light question at the register when it fits naturally, logged somewhere simple, even a running note on a phone.
This habit does something a dashboard cannot. It captures the walk in who saw the sign, the friend referral, and the AI assistant answer that carried no tracked link, all of which are invisible to analytics but very real to the business. Over weeks, a simple tally of answers starts to sketch a picture that no software report will ever hand you on its own, and it costs nothing but a moment of attention.
The two approaches are meant to sit side by side, not to compete. The analytics reports tell you, with real precision, about the sources that can be traced: search clicks, tagged links, and profile actions. The one question habit tells you, with less precision but real coverage, about the sources that cannot. An owner who uses both gets a fuller, more honest picture than either one gives alone, and an owner who relies on only the dashboard is quietly missing a meaningful share of their own customer base, whatever that share happens to be for their shop.
A running note does not need structure to be useful. A short list on a notes app, a few words per entry, saw the sign, friend told her, ChatGPT mentioned us, is enough. The value is not in the format. It is in doing it consistently enough, across enough weeks, that patterns start to repeat and become visible on their own, the same way a busy weekend rush becomes obvious after a few Saturdays even without a single spreadsheet involved.
What to check monthly, and what to let go of
A realistic monthly routine does not need to be elaborate. Check Search Console for search impressions and clicks tied to actual queries. Check the Google Business Profile performance view for website clicks, call clicks and direction requests recorded directly by Google. Check the analytics tool's own source breakdown, understanding that its direct bucket is a mix of the genuinely direct and the genuinely untraceable, not a single clean category. Keep the running note from the one question habit next to those three reports, not instead of them.
What an owner should let go of is the expectation that every visit and every customer can eventually be traced to a single source with enough tools or enough effort. That expectation is not realistic, and chasing it wastes time that could go toward the shop itself. The honest goal is a monthly habit that shows the traceable sources clearly, treats the untraceable share honestly instead of pretending it does not exist, and uses a five second question to put a human face on the part no dashboard will ever fully explain.
None of this requires new software or a paid tool. It requires reading the reports you already have for what they say, not for what a dashboard's confident looking labels might imply, and adding one small habit that no analytics platform can replace.
It is also worth resisting the urge to over interpret month to month swings in the traceable sources alone. A jump in search clicks one month and a dip the next can reflect a genuine change in visibility, or it can reflect ordinary variation in how many people happened to search that particular month, and a single month rarely settles which one it was. Reading the traceable numbers alongside the one question notes, over a few months rather than one, gives a steadier picture than either source gives by itself, and it keeps a single good or bad month from being read as more meaningful than it is.
Related Reading
More from Hillcane on this topic, plus the pages on the site that sit next to the work.
Frequently Asked Questions
Can I know exactly where every customer came from?
No. Analytics tools can trace a search click, a link with tracking parameters, and an action recorded on a Google Business Profile, because each of those carries a clue the tool can read. A walk in after noticing the sign, a friend's recommendation, or an AI assistant answer with no tracked link generates no digital trail at all, so no tool, however good, can trace it after the fact.
What does direct traffic mean in my analytics?
It means the tool could not identify a source for that visit. That bucket includes people who typed your address straight into their browser because they already knew it, and it can also include visits that arrived through a path the tool cannot label, such as an AI assistant answer with no tracking attached. A large direct number is not proof of nothing happening. It is proof the tool could not explain what did happen.
Should I trust Google Search Console over my regular analytics?
For organic search specifically, yes, treat it as the more direct record. Search Console reports impressions and clicks tied to actual queries because Google itself hands that information back to the site owner. It is a direct record of a search interaction, not an inference, which is a meaningfully different level of certainty than a source label your general analytics tool has to guess at.
What did hillcane.co's own launch data show about traceable sources?
Hillcane's site launched September 9, 2026. Search impressions began building toward roughly 124 per day by day 8, the first organic clicks appeared on day 7, and the first visits tagged as coming from ChatGPT appeared on days 8 to 9. These are small, dated numbers from one new site and are stated as a case, not as a benchmark other cafes should expect to match.
Were those AI bot hits the same as customer visits?
No, and this distinction matters. In hillcane.co's first 12 days, about 2,000 of roughly 7,738 total site hits were attributed to AI bots fetching or crawling pages. That is a record of automated systems reading the site, not a record of human customers arriving through an AI assistant. The two should never be added together or presented as the same kind of traffic.
Why can't analytics see an AI assistant recommendation?
Because not every AI assistant passes along a clean, labeled referrer the way a click from a Google results page does. Without that label, the visit usually lands in the direct bucket beside people who typed the address by hand, and nothing in the data separates the two with certainty. Some visits do carry a label: hillcane.co, launched September 9, 2026, saw its first ChatGPT tagged visits on days 8 to 9. Untagged ones stay invisible, which is why the one question habit helps.
Is asking customers how they found me useful, or just a nice gesture?
It is useful, because it captures exactly the sources analytics tools cannot see: the walk in who noticed the sign, the friend referral, and the untracked AI assistant mention. A light question at the register, logged in a simple running note over weeks, builds a picture no dashboard can hand you on its own, at no cost beyond a moment of attention.
How often should I check my traceable sources?
A monthly rhythm works for most cafes. Review Search Console for search impressions and clicks, the Google Business Profile performance view for website clicks and direction requests it records directly, and your analytics tool's source breakdown, reading the direct bucket as a mix rather than a single clean category. Pair that with your running note from the one question habit.
Does a big direct traffic number mean my marketing isn't working?
Not necessarily. A large direct number can include people who already know your shop well enough to type the address in directly, which is often a sign of loyalty rather than a problem. It can also include untraceable arrivals from other sources. Without more context, such as your own one question tally, the number alone does not tell you which is which.
What should I stop expecting from my analytics?
Stop expecting that every visit and every customer can eventually be traced to one clean source with enough tools or effort. That expectation is not realistic given how the web works. A more honest goal is reading the traceable sources clearly, acknowledging the untraceable share honestly, and using a simple customer question to put a human face on the part no report will ever fully explain.
Work with Hillcane
An owner looks at a busy Saturday and asks where all these people came from. Analytics can answer part of that question. It was never built to answer all of it.
If you want a straight read on what your own site's numbers can and cannot tell you, reach Hillcane at hillcane.co/contact or call (256) 384-2449.
Comments