One trick is having an tarpit email adress on your website. It is hidden using CSS so no real visitor sees it but it is visible in source. If your mail server recieves mail for that adress you can just block that IP for 24h.
This sounds like bad advice and would result in blocking google and other major ESPs.
I occasionally get spam from people who took the time to create gmail accounts. Based on this advice, the honey pot email address would get spam from a Gmail account and your script would block Gmail servers.
Some time ago i was wondering if the common "me at foobar dot com" you still see a lot of people do actually helps at all, especially now with LLMs, so i searched for some common "obfuscation" techniques and found this site (not the 2026 update, but the previous - it was a few months ago). Then i wrote a simple LLM query with a bunch of examples from the site[0] (the tool is just a frontend for a commandline program that uses llama.cpp and Mistral Small 3.1 in Q4_K_M quantization since it loads relatively fast and is fine for simple prompts). AFAICT it could reveal anything that wasn't relying on CSS tricks or JavaScript.
Like others mentioned, though, personally i haven't bothered by email harvesting for years now since spam filters seem to do a decent job. I have my email posted in plaintext here (which i bet is harvested very often) and in various other places and the occasional spam i get is eclipsed from "spam" from services i've actually signed up for (coughlinkedincough).
IMO a better approach would be individualized addresses.
So someone visiting your blog who wants to e-mail you can burn some CPU cycles to "earn" an address that hasn't been given out to anybody else, e.g. user+TOKEN@example.com, where it is algorithmically-unlikely for them to be able to guess a different TOKEN that will work. Then if abuse occurs, you can just retire that one address.
Naturally, this would be best with an e-mail client that is aware of the scheme, and with a mail-service that has some API for generating new addresses, such as if you want to cold e-mail somebody and use a new from/return address.
Some years ago I had the fanciful idea of doing it with a phone-app, where it manages creating new addresses as-needed, disabling them, and keeping notes about who you gave them to.
When I wrote my own brainf*ck interpreter (in C) at the start of the year I was really struggling to find a use for the language. Eventually I had the idea to obfuscate emails on my websites with the language.
Basically each email gets written as a brainf*ck program and stored in a "data-" attribute. The html only includes a more primitively obfuscated statement "Must enable Javascript to see e-mail." by default which then gets replaced by another brainf*ck interpreter (in JS) with the output of the brainf*ck code. Since we only output ASCII we can reduce the size of the brainf*ck code by always adding 32 to each value it outputs. The Javascript is loaded from what seemingly looks like a 3rd party domain. There we filter basing on heuristics and check if the "referer" matches before sending out the actual interpreter code.
Of course all this would not help if a scraper properly runs things through Javascript too.
Recently I read you soon will be able to run DOOM via CSS, so certainly it should be possible to have a brainf*ck interpreter in CSS? That would be the next step… just to get rid of the Javascript, but then I'm okay with all the downsides of using Javascript just for the e-mail obfuscation.
Anyway… I also regularly (at least once a year) rotate those public contact addresses.
WTH, a 302 into a "mailto:" (search for "HTTP redirect" in the featured article) opens up my e-mail client without clicking a mailto link!? This seems wrong.
Some browsers ask whether to open the email client in that case. I don’t see it as significantly different from a redirected download link that would open a program based on the mime type or file ending. Or from a redirect to another URL pattern associated with an app, like for example how YouTube links may open in the YouTube app.
Yes, people using “email” for “email address” in contexts where it could also mean “email message”, which “email” more frequently means, is really annoying.
Contact details: [any mailbox] [at] [the domain name of this web site]. Please don’t ask me to give interviews, sign books, appear on podcasts, attend conferences or conventions, or provide feedback or endorsements for works of fiction, scientific theories, or slabs of text disgorged by chatbots.
> HTML entities are often decoded automatically by server-side libraries, which means that even the most basic harvesters can get your email addresses without any special effort. This technique should be worthless—and, yet, it still stops most harvesters.
Anecdotal, but I’ve used HTML entities on a public static website for a long time using an href tag with mailto, and yet I’ve not seen any spam.
I guess any spammer who uses some level of GenAI to process and extract email addresses would have a lot more success against all the methods listed in this article.
Same. I have a normal mailto link on a Google-indexed page (a top hit with the right search terms) with a dedicated email address for over a decade, and rarely ever received spam for it. This is after DNSBL filtering.
I use SVG where I created a text object in Affinity Designer and converted it to curves so the SVG doesn't have text any more, just vectors for the glyphs of it. Seems to work pretty well at keeping spammers at bay.
I'm sorry, but that is not how email address are spammed in bulk.
The data-source are the enormous data breach that are more and more frequent.
There is more intensive to collect more information on someone you already know something about than spamming an email you don't even know if it's a valid one.
The spam can also be very more effective as it present itself with personal information about the spammed.
If you're only passing the address in private to some service, you can just use [some-string-unique-to-that-service]@yourdomain.com. Or, more classically, plus addressing to do the same. Then you just block that recipient.
That solution doesn't apply to the use case in the article.
Surely spammers just turn `me+leaked/sold@mail.com` into `me@mail.com` as well as `me+apple@mail.com`, `me+softbank@mail.com` etc. The cost of stripping any `+postfix` must be about zero even at volume.
Some people block all mail to non-plus-addressed emails on that inbox, so a plus address is required to be received at all. You could say then spammers will just add a random one, but they wouldn't be getting bounces and would have to guess as much. Still, even stripping the +'ed part is beyond what most of them even bother to do. That dropoff plus normal spam filters works well enough.
This is such a waste of effort. Your E-mail address is not and can't be a secret. It will get into spammer databases eventually, no matter what you do. You will spend a lot of effort doing all these fancy tricks, and eventually you will get spam anyway.
Also, a note to those who make fancy "me+someservice@somedomain.com" addresses: make really sure you are in control and these work. Some services (including mine) will need to E-mail you one day, for example to tell you that your account will be deleted because of inactivity. If you don't receive that E-mail because of your fancy spam defenses, your account will be deleted. I've seen people hurt themselves like this and it makes me sad.
On a constructive note: what works very well is spam filtering using LLMs. We have AI to help us with this problem today. I wrote an LLM despammer tool which processes my inbox via IMAP using a local LLM (for privacy reasons). I see >97% accuracy in my benchmarks on my (very difficult) testing corpus. It's nearly perfect in real life usage. I've tested many local models in the 4-32B range and the top practical choice is gpt-oss:20b (GGUF, I run it from LM Studio, MLX quantizations are worse) — not only does it perform very well, but it's also really fast.
Plus-addressing is built in to most email services. There's no 'fancy' set up to break; it just works. That is, there's no way me@gmail.com works but me+someservice@gmail.com doesn't, unless you explicitly configure it not to work. Similarly for custom domains on most services.
If you use a catch-all on a domain, i.e. someservice@somedomain.com, I guess in theory that might break. But it seems about as likely as messing up the overall domain setup.
Also, my account on your service is likely much more disposable to me than my email address/domain. Anything I care about, I'd back up. Not just assume some random website is going to preserve it for me forever.
The techniques in the article right now have had around 95%-100% success at avoiding spam and take about 5 min. to implement. Your approach of putting an LLM in front of your inbox gives 97% accuracy, may have false positives (so you may not receive that account deletion email after all), requires to run inference and, I assume, would take at least an hour to setup.
Also, the two can be complementary, anyways, so I am not sure what your point is.
I stopped being concerned about email harvesting years ago, I just simply leave the email on my website. Spam handling is okay enough, I guess.
But I like this review of techniques, even the simplest ones are very effective, that surprised me.
Same here, the address will eventually leak some way anyway.
I never got SpamAssassin working very well, but since moving my email hosting to Apple (from my own server), spam has not been a problem.
I agree that email addresses get leaked eventually.
However, LLMs are quite good at generating spam and I think soon will evade most filters.
One trick is having an tarpit email adress on your website. It is hidden using CSS so no real visitor sees it but it is visible in source. If your mail server recieves mail for that adress you can just block that IP for 24h.
This sounds like bad advice and would result in blocking google and other major ESPs.
I occasionally get spam from people who took the time to create gmail accounts. Based on this advice, the honey pot email address would get spam from a Gmail account and your script would block Gmail servers.
Some time ago i was wondering if the common "me at foobar dot com" you still see a lot of people do actually helps at all, especially now with LLMs, so i searched for some common "obfuscation" techniques and found this site (not the 2026 update, but the previous - it was a few months ago). Then i wrote a simple LLM query with a bunch of examples from the site[0] (the tool is just a frontend for a commandline program that uses llama.cpp and Mistral Small 3.1 in Q4_K_M quantization since it loads relatively fast and is fine for simple prompts). AFAICT it could reveal anything that wasn't relying on CSS tricks or JavaScript.
Like others mentioned, though, personally i haven't bothered by email harvesting for years now since spam filters seem to do a decent job. I have my email posted in plaintext here (which i bet is harvested very often) and in various other places and the occasional spam i get is eclipsed from "spam" from services i've actually signed up for (coughlinkedincough).
[0] https://i.imgur.com/ytYkyQW.png
IMO a better approach would be individualized addresses.
So someone visiting your blog who wants to e-mail you can burn some CPU cycles to "earn" an address that hasn't been given out to anybody else, e.g. user+TOKEN@example.com, where it is algorithmically-unlikely for them to be able to guess a different TOKEN that will work. Then if abuse occurs, you can just retire that one address.
Naturally, this would be best with an e-mail client that is aware of the scheme, and with a mail-service that has some API for generating new addresses, such as if you want to cold e-mail somebody and use a new from/return address.
Some years ago I had the fanciful idea of doing it with a phone-app, where it manages creating new addresses as-needed, disabling them, and keeping notes about who you gave them to.
Relevant xkcd: https://xkcd.com/1808/
When I wrote my own brainf*ck interpreter (in C) at the start of the year I was really struggling to find a use for the language. Eventually I had the idea to obfuscate emails on my websites with the language.
Basically each email gets written as a brainf*ck program and stored in a "data-" attribute. The html only includes a more primitively obfuscated statement "Must enable Javascript to see e-mail." by default which then gets replaced by another brainf*ck interpreter (in JS) with the output of the brainf*ck code. Since we only output ASCII we can reduce the size of the brainf*ck code by always adding 32 to each value it outputs. The Javascript is loaded from what seemingly looks like a 3rd party domain. There we filter basing on heuristics and check if the "referer" matches before sending out the actual interpreter code.
Of course all this would not help if a scraper properly runs things through Javascript too.
Recently I read you soon will be able to run DOOM via CSS, so certainly it should be possible to have a brainf*ck interpreter in CSS? That would be the next step… just to get rid of the Javascript, but then I'm okay with all the downsides of using Javascript just for the e-mail obfuscation.
Anyway… I also regularly (at least once a year) rotate those public contact addresses.
How does this approach meaningfully differ from having javascript that XORs the email with a random sequence of bytes stored in that JS?
It's more fun? :)
/edit
And you can combine both approaches: XOR'ing the code first for good measurements. :)
How does that work if the scraper takes a screenshot to feed to a LLM or OCR?
That seems like a very expensive way to crawl the internet
WTH, a 302 into a "mailto:" (search for "HTTP redirect" in the featured article) opens up my e-mail client without clicking a mailto link!? This seems wrong.
Some browsers ask whether to open the email client in that case. I don’t see it as significantly different from a redirected download link that would open a program based on the mime type or file ending. Or from a redirect to another URL pattern associated with an app, like for example how YouTube links may open in the YouTube app.
Good stuff, but I think the title should be Email address obfuscation. Thank you for sharing I guess, but spammers can now learn from this too (:
Yes, people using “email” for “email address” in contexts where it could also mean “email message”, which “email” more frequently means, is really annoying.
https://www.gregegan.net/
Contact details: [any mailbox] [at] [the domain name of this web site]. Please don’t ask me to give interviews, sign books, appear on podcasts, attend conferences or conventions, or provide feedback or endorsements for works of fiction, scientific theories, or slabs of text disgorged by chatbots.
I have no idea how to decipher this obfuscation.
What's difficult about it? You know the domain, gregegan.net. You know the @ symbol, presumably. Then put literally any valid text before the @.
Completely unrelated to the conversation, but our user names are remarkably similar.
What I often see is js that fetches the email from the server separately and inserts it.
> HTML entities are often decoded automatically by server-side libraries, which means that even the most basic harvesters can get your email addresses without any special effort. This technique should be worthless—and, yet, it still stops most harvesters.
Anecdotal, but I’ve used HTML entities on a public static website for a long time using an href tag with mailto, and yet I’ve not seen any spam.
I guess any spammer who uses some level of GenAI to process and extract email addresses would have a lot more success against all the methods listed in this article.
Same. I have a normal mailto link on a Google-indexed page (a top hit with the right search terms) with a dedicated email address for over a decade, and rarely ever received spam for it. This is after DNSBL filtering.
I wouldn't think it's very cost effective to apply GenAI to extract email addresses
Very interesting. It seems for his own email the author has opted for a combination of the CSS display none technique and a XOR cipher:
I'm surprised that html entity supstitution performs so well. I would have assumed that scrappers could at least speak proper html.
I use SVG where I created a text object in Affinity Designer and converted it to curves so the SVG doesn't have text any more, just vectors for the glyphs of it. Seems to work pretty well at keeping spammers at bay.
It also keeps visually impaired people at bay.
But now users cannot copy the email nor can they click it to email you?
I filter everything that does NOT include “+asdf” in the to:
I'm sorry, but that is not how email address are spammed in bulk.
The data-source are the enormous data breach that are more and more frequent. There is more intensive to collect more information on someone you already know something about than spamming an email you don't even know if it's a valid one.
The spam can also be very more effective as it present itself with personal information about the spammed.
The OP put those addresses on that web page, and only on that web page. Some addresses received spam.
Edit: that’s not to deny that big data leaks are a serious problem
If you're only passing the address in private to some service, you can just use [some-string-unique-to-that-service]@yourdomain.com. Or, more classically, plus addressing to do the same. Then you just block that recipient.
That solution doesn't apply to the use case in the article.
Surely spammers just turn `me+leaked/sold@mail.com` into `me@mail.com` as well as `me+apple@mail.com`, `me+softbank@mail.com` etc. The cost of stripping any `+postfix` must be about zero even at volume.
Some people block all mail to non-plus-addressed emails on that inbox, so a plus address is required to be received at all. You could say then spammers will just add a random one, but they wouldn't be getting bounces and would have to guess as much. Still, even stripping the +'ed part is beyond what most of them even bother to do. That dropoff plus normal spam filters works well enough.
This is such a waste of effort. Your E-mail address is not and can't be a secret. It will get into spammer databases eventually, no matter what you do. You will spend a lot of effort doing all these fancy tricks, and eventually you will get spam anyway.
Also, a note to those who make fancy "me+someservice@somedomain.com" addresses: make really sure you are in control and these work. Some services (including mine) will need to E-mail you one day, for example to tell you that your account will be deleted because of inactivity. If you don't receive that E-mail because of your fancy spam defenses, your account will be deleted. I've seen people hurt themselves like this and it makes me sad.
On a constructive note: what works very well is spam filtering using LLMs. We have AI to help us with this problem today. I wrote an LLM despammer tool which processes my inbox via IMAP using a local LLM (for privacy reasons). I see >97% accuracy in my benchmarks on my (very difficult) testing corpus. It's nearly perfect in real life usage. I've tested many local models in the 4-32B range and the top practical choice is gpt-oss:20b (GGUF, I run it from LM Studio, MLX quantizations are worse) — not only does it perform very well, but it's also really fast.
Plus-addressing is built in to most email services. There's no 'fancy' set up to break; it just works. That is, there's no way me@gmail.com works but me+someservice@gmail.com doesn't, unless you explicitly configure it not to work. Similarly for custom domains on most services.
If you use a catch-all on a domain, i.e. someservice@somedomain.com, I guess in theory that might break. But it seems about as likely as messing up the overall domain setup.
Also, my account on your service is likely much more disposable to me than my email address/domain. Anything I care about, I'd back up. Not just assume some random website is going to preserve it for me forever.
The techniques in the article right now have had around 95%-100% success at avoiding spam and take about 5 min. to implement. Your approach of putting an LLM in front of your inbox gives 97% accuracy, may have false positives (so you may not receive that account deletion email after all), requires to run inference and, I assume, would take at least an hour to setup.
Also, the two can be complementary, anyways, so I am not sure what your point is.
> Also, a note to those who make fancy "me+someservice@somedomain.com" addresses:
Just wait until one of these companies demands an email from the registered email address of your account!
My email provider allows me to send from + email addresses, just change the from header.