Hardal

Drop it in Slack. Send it to your team.

AI Visibility Playbook

AI is crawling your website. GA4 need your help.

A marketer's playbook for measuring your real AI traffic, with the tools you already run.

AI Bot Traffic · Last 24h
OpenAI0
Anthropic0
Google0
Perplexity0
ByteDance0
Humans0

Server-side only · your client-side analytics never sees this

AI Is Already On Your Site

AI is doing two things to your website right now.

// Input

Reading your content

Crawlers pull every page server-side to train models and answer queries. Your JS analytics never fires.

23,951 pages read · zero analytics events

// Result

Sending you visitors

A visitor arrives from an AI chat product. They convert 42% better, and land in Direct.

1 visitor referred · lands in Direct · no source attributed

Two Questions Your Analytics Can't Answer

Build a baseline from what you already have.

GA4's new AI Assistant channel is a start, but it misses every AI crawler and the vast majority of human traffic. This playbook gives you a first look at your real AI visibility, using the tools you already have.

01

Is AI sending you visitors, and are they converting?

A quick look via GA4.

02

Are AI engines reading your content?

A manual check of your server logs.

The Numbers You're Missing

The numbers are already in.

0.0%

of HTML web traffic is now automated. The first machine majority.

Cloudflare Radar · Jun 2026

0%

of verified bot traffic is AI crawlers (roughly 20% excluding AI-search).

Cloudflare Radar · May 2026

0%

of AI crawler requests are for training, not a live user query.

Cloudflare · 2026

+0%

growth in AI-referred US retail traffic, converting 42% better than other channels.

Adobe · Q1 2026

The crawling is huge. The traffic is tiny.

ClaudeBot reads 23,951 of your pages for every visitor it sends you. That crawling is server-side, so your analytics never sees it. Only the lone visitor shows up.

Scale of the invisible: ClaudeBot crawls vs. referrals

Bot requests

23,951

Each dot ≈ 120 requests

Human visits

1

ClaudeBot · pages crawled per visitor referred · SEOmator Q1 2026

Pages crawled for every visitor referred

ClaudeBot

23,951

pages per visit

AI Crawler

GPTBot

1,276

pages per visit

AI Crawler

Perplexity

111

pages per visit

AI Crawler

Google Search

4.9

pages per visit

Search Engine

Google search sends a visitor every ~5 pages indexed. AI crawlers take thousands. (SEOmator Q1 2026, reading Cloudflare Radar through mid-March. Ratios move fast.)

The gap you're flying blind through

The Bots

Training models devour your content and send zero traffic back.

The Humans

Visitors arrive from AI chats, convert well, then vanish into your Direct traffic.

The Blind Spot

Client-side analytics misses the first and badly undercounts the second.

CHAPTER01

GA4 AI Assistant Channel

Is AI sending you visitors?

// the easy one. Start client-side, with a channel GA4 already gives you.

See the visitors AI sends you

Since May 2026, GA4 labels AI-assistant clicks for you. No custom filters, no regex hacks. When someone clicks in from a supported AI chatbot, GA4 tags:

CHANNEL GROUPAI Assistant
MEDIUMai-assistant
CAMPAIGN(ai-assistant)

AI visitors convert 42% better than every other channel (Adobe), and the volume is growing fast. Set this up now, while the signal is small enough to act on.

GA4, step by step

  1. 1

    Open Reports → Acquisition → Traffic acquisition and read the AI Assistant row in the Default Channel Group.

  2. 2

    Add your key events as a column or comparison to see whether those visits convert.

  3. 3

    For a focused view, build an Exploration filtered to Session medium = ai-assistant with a conversion metric.

GA4 Traffic acquisition report showing the AI Assistant row in the Default Channel Group
GA4 menu labels change. The dimension values, ai-assistant, are the stable anchors. Use those.

What GA4 shows, what it hides

What to look for once it's running

  • Which pages AI visitors land on first.
  • Which sources convert. ChatGPT traffic can behave nothing like Perplexity.
  • Whether AI traffic is growing month over month.
  • Pages with high AI referral but low conversion.
  • Whether a click came from a live conversation. The page_referrer field captures chatgpt.com/c/... paths.

What it cannot see

  • ChatGPT and Claude native apps send no referrer, so those visits land in Direct. Web versions sometimes pass a referrer, but the referrer policy strips it on others. Always a partial picture.
  • Long-tail AI tools. Only assistants Google explicitly recognizes appear in the channel report.
  • Crawlers. Human clicks only.

The GA4 referrer gap

GA4's AI channel shows you a fraction of the real number. GA4only captures AI visits that arrive with a referrer header intact. In practice, most don't.

0.0%

of AI-referred traffic lands in Direct. No channel, no source, no conversion data.

Analysis of 446,000 sessions · Loamly, 2026

0%

of what GA4 calls Direct is actually AI-referred traffic.

200-site study · Attrifast, May 2026

GA4's AI Assistant channel is a floor. The real number is materially higher.

Analytics · Reports · Acquisition · Traffic acquisition
Session default channel groupSessions
Organic Search45,231
Direct12,847
Referral3,209
AI Assistantmedium = ai-assistant1,847
Paid Search921
Email456

The AI Assistant row lives here. Your analytics already has this signal.

Don't wait on Google's list

Match these referrer hostnames in your own channel group to catch custom AI traffic:

chatgpt.comchat.openai.comperplexity.aigemini.google.comclaude.aicopilot.microsoft.comdeepseek.comgrok.comyou.compoe.comphind.commeta.ai
Bing and DuckDuckGo blend AI answers with regular search results, so you cannot isolate AI-only traffic from them by hostname. A custom channel group re-buckets referral dataGA4 already holds. It does not surface new traffic.

Want to skip the setup? Hardal does this automatically.

Get early access →
CHAPTER02

Server-Side

Are crawlers reading your content?

// the part GA4 will never show you. This is your input to the AI engines.

The blind spot

GA4is an incredible tool, but it relies on JavaScript to track visitors. AI crawlers don't run JavaScript. They pull your pages directly from your server.

This means your client-side analytics has never recorded a single AI crawler. To see which AI engines are reading your content, you need to look at your server logs.

Don't worry. You don't need to learn how to query databases or write code. Your engineering team already has this information flowing through your existing infrastructure (Cloudflare, Datadog, AWS, Vercel, Splunk, etc.). You just need to know what to ask them for.

The Marketer's AI Bot Cheat Sheet

Before you ask your team for data, it helps to know what you're looking for. Not all bots do the same thing.

The TrainersGPTBot · ClaudeBot · Bytespider

Reading your whole site to build training data. Heavy traffic here means your content is baking into the next version of the model.

The SearchersOAI-SearchBot · PerplexityBot

Same job as Googlebot. Building a live index so the AI can answer questions right now.

The FetchersChatGPT-User · Claude-User

Someone just typed a question. The AI is fetching your exact page right now to build the answer.

The Copy-Paste Slack Request

Drop this exact message to your Web Ops, DevOps, or Data team to get your baseline AI visibility metrics.

# eng-analytics

Hey team! 👋

I'm doing an audit on our inbound AI visibility. Could we pull a quick report from our server logs for the last 30 days?

I'm looking for the total hit count and the top URLs crawled by these specific User-Agents:

GPTBot, ClaudeBot, PerplexityBot, ChatGPT-User, and OAI-SearchBot.

What to look for in the data

Once your team hands you the report, look for the narrative:

Heavy GPTBot but no ChatGPT-User

OpenAI is training heavily on your website, but users aren't asking about you in live chats yet.

Your Top URLs

Your most-crawled URLs are what AI treats as your most relevant content. Are your key product pages missing from this list? If so, they aren't in contention to be cited by the AI.

Pro-Tip: The Spoof Tax. Anyone can fake a user-agent and pretend to be an AI bot to scrape your website. Without cross-referencing the official IP ranges published by OpenAI and Anthropic, raw server counts often overstate real crawler activity. The Slack message above already asks about this.

//

The Complete Stack

GA4
+
Hardal

Great alone. Complete together.

GA4is excellent at what it was built for: measuring human visitors, campaigns, and conversions in the browser. Keep it. It's not going anywhere.

But GA4runs on JavaScript. AI crawlers don't execute JavaScript. Every bot training on your content, every agent reading your pricing page, invisible to GA4, visible to Hardal. You don't choose one over the other. You run both.

GA4, already doing this well

The Browser Web

Revenue, human conversion rates, campaign performance. GA4 was built for this and it does it well. No reason to replace it, this is its home turf.

Hardal, the layer GA4 can't reach

The Agentic Web

Bots training on your content. AI agents reading your pricing pages. The “Direct” traffic that's actually coming from ChatGPT. Hardal catches it all, server-side, no JavaScript required.

Keep your GA4. Add Hardal once. See everything.

Hardal AI Visibility

One report. Bots and humans. Always current. Add Hardal once, and your AI traffic, both the invisible crawlers and the human visitors, lands in one report that updates automatically as new bots ship.

Hardal Agent Breakdown: top AI agents visiting your website with traffic share and visit counts

Without Hardal

Manual chaos

Per-platform queries. A user-agent list that breaks every time a new bot ships. Spoofed bots inflating your counts.

With Hardal

One signal, all sources

Built-in IP validation. Bots and humans in one report, giving your GA4 data the complete server-side context it has been missing.

Request Early Access

See your own AI traffic.

AI Visibility is in early access. Run Hardal on your website. See your crawl traffic, the visitors AI sends you, and how both trend, in one report, updated automatically.

No spam. We'll only contact you about early access.

You have the playbook, so you are already ahead. Early-access requests from playbook readers go to the front of the queue. 🌭