Automated requests reached 57.5% of web-page traffic by June 2026, according to Cloudflare data reported by PPC Land. Bots now outnumber people on the web.
The detail underneath that headline matters more than the headline does: roughly 52% of identified AI crawler activity is gathering material to train models, while only about 2.6% is a real-time fetch triggered by an actual person asking an actual question.
So the overwhelming majority of AI crawling on your site is not a customer. It is not even a customer's assistant looking something up on their behalf. It is a machine reading your pages to get better at answering questions in general, and it will never send anyone back.
This quietly breaks the assumption most businesses still run their reporting on: that traffic went down, therefore interest went down.
That is no longer safe.
Two things are happening at once
More machines are visiting your site. Every assistant that wants to answer questions about your business has to read your pages first, and as the 52%-versus-2.6% split shows, most of that reading is stockpiling rather than serving a live query. Crawlers from OpenAI, Google, Anthropic, Perplexity and others arrive repeatedly. None of them will ever call you.
Fewer humans are arriving to find things out. When an assistant can answer "what are their opening hours" or "do they service my area" directly, the visit that used to answer that question does not happen.
So your raw traffic number is now made of a growing bot component and a shrinking informational-visit component, moving in opposite directions. Whatever it does, it is not telling you what it used to.
Why this matters commercially, not just technically
A business looking at a 20% drop in sessions concludes something is wrong and often reacts: cut the content budget, question the SEO spend, move money to ads.
But if the visits that disappeared were people checking your hours, nothing commercially relevant has changed. And if the visits that remain are people who asked an assistant to compare options and were sent to you, the remaining traffic is better qualified than what you lost.
The opposite error is just as expensive. A business whose bot traffic is climbing sees total sessions holding steady and concludes everything is fine, while human visits quietly fall underneath.
Both mistakes come from reading one number that now contains two unrelated stories.
What to do about your reporting
Separate bot traffic from human traffic, and look at them apart. Most analytics platforms filter known bots by default, but AI crawlers are newer than those lists and coverage varies. Your server logs are the honest record here. If you have never looked at them, this is the reason to start.
Stop treating total sessions as the headline metric. Move it down the page. Promote the numbers that survive this shift:
- Calls and form fills
- Branded search volume, meaning people searching your name rather than your category
- Direct traffic
- Booked work, matched back to where it came from
None of these move because a crawler visited. All of them move when demand actually changes.
Watch branded search closely. This is the most useful signal in an AI-mediated market and almost nobody tracks it. If an assistant recommends you, plenty of people will not click the link. They will search your name separately, or type your domain directly, or just call. That looks like "direct traffic" or "branded search" in your reports, with no visible source. Rising branded search alongside flat organic traffic is a business being recommended somewhere you cannot see.
Check whether you are being cited, not just crawled. Being crawled means a machine read you. Being cited means it decided you were worth naming. These are entirely different outcomes and only one of them brings you work. The test is manual and takes minutes: ask ChatGPT, Gemini and Perplexity the questions your customers ask, and see who gets named.
Should you block AI crawlers?
Some businesses have started to, and Google has made this easier: Search Console now offers a site-level opt-out from AI Overviews, AI Mode and Discover AI features, as Search Engine Journal reported.
For publishers whose product is the content, blocking has a real logic. If an assistant answers the question your article exists to answer, the visit and the ad revenue both vanish.
For most service businesses, the logic runs the other way. You do not sell page views. You sell work, and you want to be the name that comes up when someone asks who to hire. Blocking the crawlers is choosing not to appear in the answer, which is the opposite of what you want. The publisher's problem is not your problem, and borrowing their solution would be a mistake.
The nuance worth keeping: opting out of AI Overviews is not the same as opting out of search. Read carefully before touching anything, because these controls are new and the wording is not always as narrow as it first appears.
The uncomfortable part
There is currently no clean, universal way to attribute a booked job back to an AI recommendation. The assistant does not pass a referrer the way a search engine does. Someone who was told to call you and then called you looks, in your reporting, like they arrived from nowhere.
That gap is real and anyone claiming to have solved it entirely is overselling. What you can do is narrow it: track branded search, ask on the enquiry form how people heard about you, and watch direct traffic against periods when you know your visibility changed.
Imperfect attribution is not a reason to ignore the channel. It is a reason to stop judging the channel with metrics built for a different one.
What to do this week
- Pull your server logs and find out what proportion of your traffic is automated. The web-wide figure is 57.5%; yours will be its own number, and most business owners have never looked.
- Move calls, form fills and branded search to the top of your reporting. Demote total sessions.
- Run the citation test. Ask three assistants the question your best customer would ask. Write down who gets named.
- Do not block AI crawlers unless you sell content directly and have thought hard about it.
The bottom line
The web now has more machine visitors than human ones. Your analytics were built for a web where that was not true.
The businesses that misread this will cut spending on the exact work that is making them recommendable, because a number designed for a different era went down. The ones that get it right will change what they measure, not what they invest in.
Want to know whether assistants are citing you or only crawling you? Request a visibility check.
PresenceKit Team
Helping small businesses grow their online presence
