The exact regex, filter list, and python script i use to scrape agency emails from any website in 3 seconds

i've scraped 570+ agency websites across 43 countries in the last two weeks. the script is 55 lines of python. here's the whole thing, explained.

the core: 3 lines that do 90% of the work

requests.get hits the URL. re.findall with an email regex pulls every email-shaped string from the HTML. a filter list removes the garbage. that's it.

the regex: [a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+.[a-zA-Z]{2,}

this catches 95%+ of emails on web pages. it's intentionally loose — i'd rather catch false positives and filter them than miss real emails.

the filter list (this is the real secret)

after scraping 570+ sites, my filter has 25+ patterns. without it, about 30% of "emails" are garbage:

• image filenames: .png, .jpg, .webp, .avif, .svg (yes, regex matches hero-banner@2x.png)
• tracking pixels: sentry, cloudflare, googleapis
• CMS artifacts: wix, wordpress, schema.org
• generic placeholders: example.com, your@, name@, email@, test@, john@doe, john@company
• department inboxes you don't want: hr@, jobs@, careers@, training@, complaints@
• free email providers: hotmail, yahoo (agencies using these are usually not real agencies)
• platform emails: support@wix, info@w3, noreply

every time i scrape a new batch and get a weird false positive, i add it to the filter. the list grows every round.

the multi-path trick

most agency websites don't put their email on the homepage. the contact page is where it lives. so for each agency, i try 3-5 URL paths:

• / (homepage — catches about 20% of emails)
• /contact (the money path — catches 50%+)
• /contact-us (american english variant)
• /contacto (spanish)
• /kontakt (german/scandinavian)
• /en/contact (multilingual sites)

this one change tripled my hit rate. before i was only hitting homepages and getting ~25% extraction rate. with multi-path it jumped to ~65%.

deduplication matters

many agencies list the same email 3-4 times across their pages (header, footer, contact form, sidebar). the script uses a set() to deduplicate per-agency, and checks against all previously found emails to avoid sending the same person two pitches from different agency listings.

what i do with the emails

each batch gets saved as a JSON file with the agency name, URL, and email. a separate script reads these batches, scans each agency's website with a custom SEO analyzer, and generates a personalized pitch email that references specific problems found on their site.

the personalization is the key differentiator. instead of "hey, we offer AI outreach services," the email says "your homepage is missing alt text on 23 images, has 4 broken links, and your page speed score is 34/100. we can fix this and find your clients' leads."

the numbers after 2 weeks

• 570+ agencies scraped across 43 countries
• 290 pitches sent via gmail SMTP
• 3 replies (all from AU/NZ/UK — english-speaking markets)
• 0 paying customers yet
• total cost: $0/month (python + gmail + linux server i already had)

the pipeline works mechanically. the conversion is the part i'm iterating on. every week the pitch gets slightly better based on what gets replies and what gets ignored.

if you want to see the landing page agencies are being sent to: https://vemtrac-outreach.pages.dev

happy to answer questions about the technical setup or share my filter list if anyone's building something similar.