Skip to the content

Case study

Facebook link visits are bots: telling readers from robots

On our own blog, a small crowd arrived after every Facebook or Threads post, read nothing and left. Our analytics counted them as people because they pressed the cookie banner. They were Meta's robot, and the fix changed what we count as a visitor.

Key takeaways

  • A short visit from a Facebook or Threads post may be Meta checking the link, not a person reading it.
  • A press at exactly the top left corner of a page is the mark of a program pressing a button by name.
  • Judge social posts by pages read, scrolls and time on page, not by the first visits in your analytics.
  • When a visitor rule changes, judge the past month again with the same recorded actions so the old and new figures can be compared.
  • You can often separate robots from readers with the data already collected, without adding tracking.

On our own blog, we changed what counts as a visitor after finding that visits from Meta's networks were being counted as people when no one had read the page. The change was live the same afternoon, and the past month could be judged again without collecting anything new. If you share links on Facebook or Threads, the same pattern can make a post look better read than it was.

A post nobody read looks like it worked

If you decide what to post next from the visitor count, a visit that never read can make a weak post look strong. On our own blog, visits from Meta's networks arrived within seconds of each post, read nothing and left. In one week, most of the people our blog counted were visits of this kind. An earlier alert about a sudden group of people on our main website turned out to be the same thing. The dashboard does not label them as robots. It shows people, and that is the figure an owner uses to judge whether social media is worth the time.

How our engineer traced the visits to one source

We opened the suspect visits one by one instead of trusting the total. Our analytics already set robots apart from people: a visit counted as a person only when somebody pressed something, scrolled, filled in a form or stayed half a minute. The engineer read what each suspect visit had done, press by press. Each came from Facebook right after a post and did the same few things. It pressed Accept on the cookie banner, put the cursor in the search box without typing, scrolled nowhere and left within seconds. Its time on the page was always exactly the same, to the thousandth of a second.

Every press on the banner landed at the very top left corner of the page, the point at exactly nought and nought. A person with a mouse or a finger never presses exactly there. A program that presses a button by its name, without moving a pointer, always does. That made the press itself, which our rules took as the surest sign of a person, the robot's signature. The search box told the same story: it was only given focus, never used.

Why the robot got through, and what to count instead

The robot got through because one press was enough to count a visit as a person. Meta checks every link shared on its networks by opening the page with a program, and a page that asks visitors to accept cookies gives that program a button to press. For your own analytics, do not let a single press decide. Require a second sign, such as a scroll, a form actually sent or half a minute on the page, and treat a form that was only started as no sign at all.

What we changed and what the figures show now

We changed what counts as a sign of a person. A scroll, a press anywhere but the top left corner, and a form actually sent still count. A press at exactly the corner, and a form only begun with nothing sent, no longer do. Such a visit must show a person another way, by a scroll or by staying half a minute, or it is set apart as showing no sign of a person.

The cases that had to keep working

A real reader who accepts the banner and then scrolls through the article stays a person, and so does someone who presses Accept with the keyboard and then reads. We judged the past month again by the new rule, using only the presses, scrolls and forms our analytics already recorded. Nothing new is collected about anyone, and our privacy policy did not need to change.

What the figures show now

It was live the afternoon it was found, during a weekly review of how our posts and videos were doing. Judged again, about half of the people our blog had counted over the month before were Meta's robot, and in the week the trace began about three in five were. On our main website, where most visitors arrive another way, it was about one person in eleven. Every report now counts people without it, and each social post shows only the visitors who actually read.

What you can check yourself

  • Open your analytics right after you post a link on Facebook or Threads and see how many visits arrive in the seconds after you post and how long they stay.
  • Compare the visits a post brings with the clicks Meta itself reports for that post. A big gap is a sign of robots.
  • Look at visits that pressed only the cookie banner and nothing else. If they all last a few seconds, they are probably not people.
  • Check whether your analytics treats accepting cookies as proof of a person. If it does, ask whoever looks after it what else a visit must do to count.
  • Look at visits from Facebook with no pages read and the same time on page to the fraction of a second. A repeated exact time is a sign of a program.

Questions people ask

Why do visits from my Facebook posts leave after a few seconds?

Short visits right after a post often come from a program checking the link, not from a reader. A program can open the page, arrive and leave without reading. A real reader who is interested usually leaves a sign such as a scroll or a further page view, so judge the post by those signs rather than by the arrival count.

Does Meta open my website when someone shares my link?

Meta checks links shared on its networks by opening them with a program. That is separate from a person choosing to read the page. If your page asks for cookie consent, the check may clear the banner on its own.

Should I stop counting cookie banner clicks as real visitors?

A press on its own is a weak sign, so do not let it decide that a visit was a person. Set the rule to require a second sign, such as a scroll, a form actually sent or half a minute on the page. Keep the banner press as one signal, not as proof.

Will changing this rule make my past reports wrong?

A rule change only breaks the comparison if the old figures stay as they were. We judged the past month again with the same recorded actions, so the old and new figures could be compared. No new information was collected about visitors, and the privacy policy did not need to change. If you cannot re-judge past figures, note the date the rule changed so you know where the line falls.

How can I tell real visitors from robots without collecting more data?

Use what your analytics already records: pages read, scrolls, forms sent, time on page and where a press landed. A robot that presses a button by name tends to land at the top left corner and do nothing else. A person usually leaves other signs, such as reading further or staying longer.

From Tech Concierge

Want this handled for your business?

Your own engineer builds and runs the website, the app, the CRM and the AI assistant that answers your customers.

The first 7 days are free.

Read next

Case studyMarketing

Why robot traffic in analytics looks like new customers

A burst of sign-ups after an app update looks like growth, and a busy website looks like demand. Here is how we found that almost all of it was machines, and how to tell people from robots in the figures you already record.

5 min read

Case studyAI and automation

When the AI video tool changed its interface, the ads stopped

Presenter lines stopped being made while the screen still promised a finish time of a few hours, and tall clips came out wide. What a quiet change at an AI video service did, and how to check for it.

5 min read

Case studyWebsites

Customers hit an error and leave: making reporting one press

On a client's online tool, a red error message gave people no quick way to tell the team, so most said nothing. Reporting is now one press, with the page carrying the details instead of the person.

3 min read

We count visits to this blog with our own analytics, never Google's, and share what you read with nobody. May we remember this browser with a cookie? See the privacy policy.