Detecting scraper bots through scroll behaviour
niki.cat
niki.cat
If you're providing a service, and AI agents truly take off to where people say "Claude, go buy me a new shower hose about 5' long, brushed nickel. Use my usual evaluation criteria and don't spend over $20." then a lot of ecommerce stores are very very incentivized to be bot friendly.
And of course every "content" site is the reverse, wanting to not waste their bandwidth and hoping to score exclusive paid content deals with any entity that might want to scrape them.
Adafruit, last year: They blocked bots very heavy-handedly, including those that were there to gather information at the behest of a present-at-the-keyboard human. This made their products nearly invisible for those who were using ChatGPT instead of Google to find particular things to buy. Even when they were discoverable through forum postings, this added friction meant that I was unable to make a swift comparison betwixt their wares and those of others, and I wasn't buying anything from them because they were making deliberate choices that I felt were hostile to my process.
But that's changed.
Adafruit, at least a week or so ago: The bot was able to get in and find details about the stuff they sell. This lack of friction meant that I was able to make a swift comparison between their wares and those of others. I still bought the particular sensor-chip-on-a-board widget that I wanted elsewhere, but at least Adafruit's offerings were in the running this time.
the blocking of automated attacks or purchase attempts wasn't aimed at you. when a scarce part is in stock here cheaper than everywhere else, scripts that some human decided to go after us with, show up and try to clear the store before a human person can click buy when it's in stock, first fix is a blunt instrument, then filing it down. your bot getting through is the filing working, good to know, most people only post when it's broken. most people are only complaining, most people. but you, us, and others, are not most people - we work here.
you bought the widget elsewhere, fair, that's what comparison shopping is for, is there is a personal axe to grind, could probably help with that too. we'd still like to know where the price, stock, or a spec buried three clicks deep. we can fix just about anything, you can help us.
if you've got a project that keeps stalling on tooling instead of the idea, send that too and ladyada and i will look at it as a problem to solve, and share the results, like we always do. if the answer is nothing adafruit makes, that's a useful answer. and next time the site snags your setup, tell us exactly where.
today is circuitpython day too, it's taken almost 10 years for being able to do anything i want in less than 5 minutes to get started, and then i can spend as much time on the best parts.
I remember the days of the Great Pi Shortage very well. You guys were trying very hard to sell Raspberry Pis at sensible prices to real people instead of to computer programs, which was/is a commendable effort.
The part I was trying to find last year was a USB canbus interface to do some automotive hacking with. That was a frustrating experience because trying to find such a widget with Google retrieves endless pages of impossible-to-exclude OBD-II interfaces and "CANBUS" replacement automotive LED lamps -- stuff that obviously wasn't applicable to my hacking project at all.
So I used the bot (ChatGPT) to help sort the good stuff that I might want to buy from the irrelevant stuff that I did not want to buy. It got stalled hard when trying to access adafruit.com, which did not help my process.
But, like I said: It's working now. Last week, when I set forth on a small project to characterize daylight for home automation, the same bot worked well with finding your pages on the AS7341 boards you list. No issues. It all flowed very well and Adafruit was listed prominently in the bot's output as a potential place to buy a board from. :)
Anyway, for full disclosure: I see that this board is out of stock right now and I don't remember if that was the case last week or not. I was nearly broke at that time (insert "in this economy" joke here), so I wound up looking on Aliexpress so that maybe I could move the project forward and also buy a cheeseburger. A seller there had a little board which is supposed to have an AS7341 and appears to have the right interface glue brought out to sensible pins, for like $2.50 -- delivered to my door in the US.
You guys can't compete with that. There's no way. Someone lost money on that transaction.
But! The bot wasn't able to find this on Aliexpress. It got stuck, and didn't recommend that as a potential source at all.
So score one for the team: Your stuff is showing up in the bot's results, while your (ruthless) competition is not! That exposes more people to the things that you guys work so hard to accomplish, and more exposure tends to lead to more sales, which tends to help keep the lights on. :)
It's good stuff. Right now, you're ahead of the game.
And thanks. I will shoot you an email directly if I find an aspect that's broken or buried in the future.
The end result was that I got millions of requests daily trying to download every HTML page (every variation of action, commit hash, branch, whatever else cgit allows) and basically no traffic to the proper clone route.
Turns out the bots just don't care about your bot-friendly access method and will instead throw more illicitly-proxied residential IPs at the problem.
plus, AI would also skip search result ads, and, if those have to legally be displayed as such, then there goes the market for them, I'd think
detect all element in currently loaded page I want to do things with. element I want to click. scroll to element I want to click. random viewport scrolling action, go slightly over viewport, scroll back up, determine mouth position where I will click - x, y center of clickable item, determined x,y of bounding box of clickable item, choose some point within bounding box where I can click with some randomness for getting to center because obviously humans don't get the exact center every time do they. set random mouse move speed. mouse choppiness function (determine a number of x.y coordinates on the way towards the x.y you will actually be clicking move towards those each in turn with randomness of mouse speed, so that nobody can say huh, this mouse is moving towards the place it will click with an unnaturally direct and unnaturally smooth movement.) Get to where I will click. wait random time to click. click.
If I do not continually get new elements I want to click based on my criteria just by this method now I need to scroll to see if I can detect new.
random number of times I will scroll. random scroll speed. random scroll distance.
start scroll. hit scroll distance stop. random up down scroll.
My automation goal was not to make my bot able to do the work of hundreds of people. It was to do the work of a few people.
However I eventually gave it up since the target platform was continually deciding I, as a human, was somehow behaving as a bot, it followed that it was somewhat irrational to try to make a bot that behaved as close to a human as possible.
(note there was all sorts of other erratic behavior boiled into my bot, for example how it wrote text, since it had to do keypresses, and randomized typing speeds, and making errors, and back spacing to fix errors)
on edit: before you start telling me about how it's not randomized enough or whatever, I am of course giving a curtailed high level description of the kinds of noise I added in to my bot behavior, because I just really wanted to spin up a bot that could spend a few hours doing stupid work while I did other things, and then write up a report and then later I could come in and do some more clever things.
I was experimenting with Google Ads traffic to one of the websites I cared about, and even just looking at the existence of mouse move and scroll events, let alone the patterns, made me understand how many bots, or users who just (mis)clicked the ad and never did any action on my website, were counted as paid clicks in Google Ads. Analyzing for patterns like in this article would've made even more sense.
It all averages out.
You bid $3 a click because your conversion ratio is x, and paying $3 is worth it. Even if 90% of the traffic is bots, it works out. If Google filtered out all the bots, the conversion ratio for you and your competitors would go up, and your competitors would be willing to pay $27 per click and you'd outbid at $30, being at the exact same result at the end of the month spent for the exact same results.
I'm a human, I don't execute javascript programs and by default my browser does not either. Anyone using this will label me a bot.
i have been finding my locked down experience has been becoming degraded lately fwiw. i am not economically "interesting", and the consequences of valuing privacy in 2026 is limiting information access. shame really.
This made me hit the bandwith cap of my hosting provider and had to pay extra per TB used that month from my own pocket, since the website was not making a profit at all.
I never thought of increasing profits for any institution, but to stop the waste of bandwith many independent websites suffer.