The article doesn't really get into the details. Does it analyze the user agent and compare it to a list of known bot user agents? What about all the bots that spoof user agent values – does it do something special to detect those?
Yes exactly — it matches against a database of 18
known AI bot user agent tokens (GPTBot, ClaudeBot,
CCBot, Bytespider etc.) plus their known IP ranges
where available. GPTBot for example publishes its
IP ranges officially so we can match on both UA
and IP.
The spoofing problem is the hard one. Bots that
fully spoof Chrome headers are invisible to any
UA-based tool including this one. The honest answer
is that BotCost catches the "polite" bots that
identify themselves — which covers the major AI
companies (OpenAI, Anthropic, Google, Meta) since
they all self-identify. The truly malicious scrapers
that spoof identities are a harder problem requiring
behavioral analysis.
So it's accurate for what it is — catching known
AI training and search crawlers — but not a complete
bot detection solution.
Ha, that's a fair point — vulnerability scanners are
a whole separate category I haven't tackled yet.
BotCost is specifically focused on AI training and
search crawlers right now (GPTBot, ClaudeBot,
Bytespider etc.) rather than security scanners.
Good news: 0% AI bot traffic on an unadvertised
landing page makes sense — those bots tend to
follow links and sitemaps. If you run it on a
site with real content and traffic you'll likely
see a different picture.
Vulnerability scanners on the other hand... that's
a different problem worth solving too.
Good question. The cost estimate uses two components:
1. Bandwidth: total bytes served to bots divided by
1GB, multiplied by $0.09/GB (AWS/Cloudflare
blended average rate)
2. Compute: total bot requests divided by 1 million,
multiplied by $0.40 (Vercel/Lambda average per
million invocations)
Both rates are configurable assumptions — the real
value is seeing the relative breakdown between bots
and the order of magnitude of waste. Your actual
cost depends on your specific hosting provider.
Got it, makes sense. Worth noting intra-region vs inter-region transfer can differ a lot too. As a blended average for an order-of-magnitude estimate, this is really useful.
Good point — intra vs inter-region transfer costs
can vary significantly, especially on AWS. The
$0.09 is deliberately conservative as a blended
estimate. A future version could let users input
their actual hosting provider rates for a more
precise number. Adding that to the roadmap. Thank you.
Fair question given the context. I used AI tools
to help build the product, and I do use AI to help
draft responses — but I read, edit, and post every
reply myself. Nothing is auto-posted.
Thanks! Curious what you're running — are you seeing
AI bot traffic on your site? Would love to know if
the log formats you use are covered (Nginx, Apache,
Cloudflare CSV, Vercel JSON supported right now).
Also wrote up the background on Dev.to if anyone wants more context on how it works: [https://dev.to/plaintos_app_fd54e75a054e/i-built-a-free-tool...]
The article doesn't really get into the details. Does it analyze the user agent and compare it to a list of known bot user agents? What about all the bots that spoof user agent values – does it do something special to detect those?
Yes exactly — it matches against a database of 18 known AI bot user agent tokens (GPTBot, ClaudeBot, CCBot, Bytespider etc.) plus their known IP ranges where available. GPTBot for example publishes its IP ranges officially so we can match on both UA and IP.
The spoofing problem is the hard one. Bots that fully spoof Chrome headers are invisible to any UA-based tool including this one. The honest answer is that BotCost catches the "polite" bots that identify themselves — which covers the major AI companies (OpenAI, Anthropic, Google, Meta) since they all self-identify. The truly malicious scrapers that spoof identities are a harder problem requiring behavioral analysis.
So it's accurate for what it is — catching known AI training and search crawlers — but not a complete bot detection solution.
I ran it on my personal domain that has just some kind of landing page and nothing else on it, and it's not advertised anywhere:
0% of traffic is non-human · <$0.01/yr projected
Well, 0% of traffic is AI bots. 99% of traffic is vulnerability scanners actually.
Ha, that's a fair point — vulnerability scanners are a whole separate category I haven't tackled yet. BotCost is specifically focused on AI training and search crawlers right now (GPTBot, ClaudeBot, Bytespider etc.) rather than security scanners.
Good news: 0% AI bot traffic on an unadvertised landing page makes sense — those bots tend to follow links and sitemaps. If you run it on a site with real content and traffic you'll likely see a different picture.
Vulnerability scanners on the other hand... that's a different problem worth solving too.
nice, how are you calculation the cost?
Good question. The cost estimate uses two components:
1. Bandwidth: total bytes served to bots divided by 1GB, multiplied by $0.09/GB (AWS/Cloudflare blended average rate)
2. Compute: total bot requests divided by 1 million, multiplied by $0.40 (Vercel/Lambda average per million invocations)
Both rates are configurable assumptions — the real value is seeing the relative breakdown between bots and the order of magnitude of waste. Your actual cost depends on your specific hosting provider.
Got it, makes sense. Worth noting intra-region vs inter-region transfer can differ a lot too. As a blended average for an order-of-magnitude estimate, this is really useful.
Good point — intra vs inter-region transfer costs can vary significantly, especially on AWS. The $0.09 is deliberately conservative as a blended estimate. A future version could let users input their actual hosting provider rates for a more precise number. Adding that to the roadmap. Thank you.
Was that response written and/or auto-replied by AI?
Fair question given the context. I used AI tools to help build the product, and I do use AI to help draft responses — but I read, edit, and post every reply myself. Nothing is auto-posted.
Hopefully not by using another AI bot... ( ͡° ͜ʖ ͡°)
Nope, pure JavaScript in your browser. No AI bots were harmed or employed in the making of this tool.
interesting!
Thanks! Curious what you're running — are you seeing AI bot traffic on your site? Would love to know if the log formats you use are covered (Nginx, Apache, Cloudflare CSV, Vercel JSON supported right now).