74%
of requests were attacks and scanners, mostly hunting for WordPress logins on sites that don't run WordPress.
The 286,000 WordPress probes →Not all your traffic is human
Real people, search engines, AI crawlers, fake bots and attacks, sorted from what actually reaches your server. Works alongside Google Analytics, not instead of it.
Early access is free. We'll only email you about Humansift.
Recognises more than 100 bots by name. The ones marked ✓ are checked against the addresses their owners publish.
We pointed Humansift at our own five websites first
74%
of requests were attacks and scanners, mostly hunting for WordPress logins on sites that don't run WordPress.
The 286,000 WordPress probes →3.2×
as many pages loaded by ClaudeBot as by Googlebot. AI crawlers read more of our sites than search engines did.
Which AI reads your site →1 in 6
requests calling themselves Googlebot were fake, and more than half of "ChatGPT" requests.
How to spot a fake crawler →<10%
of visitors could have been people at all. Everything else was software: attacks, scripts, crawlers and bots.
The full breakdown →Try it on your site · free, no sign-up
Three checks you can run right now, on the same engine as the product.
See it for yourself
| Day | Attacks & fakes | Bots & scripts | Search & tools | AI | Possible people |
|---|---|---|---|---|---|
| 2026-10-02 | 3743 | 457 | 195 | 125 | 797 |
| 2026-10-03 | 4261 | 323 | 172 | 175 | 636 |
| 2026-10-04 | 3391 | 395 | 171 | 88 | 287 |
| 2026-10-05 | 4379 | 447 | 293 | 165 | 458 |
| 2026-10-06 | 3934 | 473 | 267 | 123 | 420 |
| 2026-10-07 | 5258 | 465 | 407 | 76 | 514 |
| 2026-10-08 | 4365 | 490 | 252 | 90 | 523 |
| 2026-10-09 | 4270 | 693 | 255 | 137 | 505 |
| 2026-10-10 | 4222 | 630 | 423 | 339 | 412 |
| 2026-10-11 | 2366 | 610 | 366 | 92 | 296 |
| 2026-10-12 | 3663 | 498 | 273 | 164 | 431 |
| 2026-10-13 | 3250 | 551 | 365 | 236 | 519 |
| 2026-10-14 | 2617 | 564 | 585 | 272 | 430 |
| 2026-10-15 | 2040 | 491 | 295 | 223 | 372 |
Daily visitors (one address and browser on one day) to our own five sites, 2–15 October 2026, from server logs. "Possible people" is an upper bound: with the tracking script, Humansift splits it into people who interacted and those who didn't.
Your SEO tools show how the web sees your site. Google Analytics shows what runs its tag. Humansift shows what actually reaches your server, and checks who sent it.
| What you can see | Google Analytics | SEO suites | Humansift |
|---|---|---|---|
| Real people, apart from automated browsers | Counts many bots as users | – | Checked by behaviour |
| Googlebot's real crawl of your pages | Never sees it | Their own crawler | Every fetch, verified |
| AI crawlers and assistants, per page | Never sees them | – | GPTBot, ClaudeBot, Perplexity… |
| Fake crawlers and attacks | Never sees them | – | With alerts |
| Keywords, backlinks, rankings | Via Search Console | Their strength | Your Search Console, side by side |
How it works
Every request, including bots that never run JavaScript. Upload a file, send them to our API, or use our Cloudflare Worker if you don't have server access.
203.0.113.9 - - "POST /xmlrpc.php" 200 66.249.66.1 - - "GET /guides/" 200 Googlebot ✓
It reports a page view and the first real interaction (a scroll, tap or keypress). Scripts that never interact don't pass as people.
<script src="https://humansift.com/t.js" defer></script>
Googlebot, Bingbot, GPTBot, ChatGPT and Perplexity are checked against the addresses their owners publish. Impostors show up as fake crawlers.
Googlebot verified "Googlebot" fake · cloud server
What you get
Visitors who scrolled, tapped or typed, per page and per source. The number your marketing should be judged on.
Which pages GPTBot, ClaudeBot and Perplexity read, and when ChatGPT fetches a page because someone asked about you.
Pages Google hasn't fetched in 30 days, crawl wasted on errors and redirects, and how fast new pages get picked up.
WordPress probes, .env hunters, login floods and card testing, with alerts when they spike.
A loud alert the moment someone ships Disallow: / or a stray noindex, before search traffic drops.
Clicks from Google, Meta and Microsoft ads split into people and bots, per campaign.
Server time and page speed measured on real visitors' requests, not one lab test.
Branded client reports, scheduled emails, and CSV feeds for Looker Studio and Google Sheets.
Alerts that matter
Humansift watches your site around the clock and emails you when something changes that analytics would only show weeks later, as a dip in a chart.
The new file has Disallow: / for every crawler. Search traffic stops within days.
Someone is trying stolen passwords on /login. Most addresses are cloud servers.
Not in OpenAI's published addresses. It asked for /.env and /config.json.
It usually fetches about 40 pages a day. Check Search Console for crawl errors.
Example alerts. Each one says what changed and what to do about it.
Who it's for
Stop guessing whether last month's jump was customers or crawlers. See the people, the pages they read and where they came from, in plain words.
Card testing and login floods never show in analytics, and bot clicks on your ads cost real money. Both are in your server logs.
One workspace per client, branded reports on a private link or a schedule, and CSV feeds that drop straight into your Looker Studio dashboards.
AI crawlers read more of our pages than Google did. Find out which ones read yours, which assistants fetch your articles for readers, and what it costs you.
Send logs by API or a Cloudflare Worker, and get verified crawler identity, attack paths and real-user response times without building a pipeline.
Humansift was built to count people, not to follow them.
Works with what you have
Questions
Something else? Join the waitlist and reply to the first email. A person reads every one.
Google Analytics counts visits that run its tag. That misses everything that doesn't run JavaScript (crawlers, scrapers, attacks) and counts automated browsers that do as users. Humansift reads what actually reaches your server, checks who sent it, and shows your Google Analytics and Search Console numbers next to it.
No. Server logs give the fullest picture, but you can also send traffic from Cloudflare with a small Worker we give you, upload log files, or start with the tracking script alone. The setup page recognises your platform (WordPress, Shopify, Wix, Squarespace, Webflow, Ghost, Next.js or Google Tag Manager) and shows the exact steps.
No. The script is under 6 KB, loads after your page, and sets no cookies. Server logs are read separately, so they add nothing to your pages.
Our script sets no cookies and we never store IP addresses: each visitor is a hash that changes every day, so nobody can be followed from one day to the next. Check your own legal requirements, but there is no tracking cookie to ask consent for.
Several signals together: the browser runs the page and interacts with it (a scroll, a tap, a keypress), it comes from a home or mobile connection rather than a data centre, it shows none of the signs of an automated browser, and it never asks for attack paths. Visitors who load a page but never interact are counted separately as probably people.
More than 100 by name: search engines, AI crawlers and assistants, SEO tools, link previews, monitors and feed readers. Googlebot, Bingbot, Applebot, GPTBot, ChatGPT, OpenAI search and Perplexity are checked against the address lists their owners publish; others by reverse DNS. Anything claiming to be one of them from the wrong address is flagged as a fake crawler.
No. SEO suites show how the web sees your site from the outside: keywords, backlinks, rankings. Humansift shows what reaches your server. They work well together.
Yes. Each client can have their own workspace, and branded client reports go out as a private link or a scheduled email, with your logo and colours. Clients can connect their Google accounts through a link you send.
Early access is free. Plans for larger sites and agencies will be announced before early access ends, and nobody is charged without choosing a plan first.
It's sorted in memory and thrown away when the result is sent back. Nothing from it is stored.
Early access is opening to a small group first. Agencies and site owners with real traffic go to the front.
Early access is free. We'll only email you about Humansift.