How does the attribution model work?

For every page on your site we track two daily signals: (1) AI-agent UA hits (ChatGPT-User, Perplexity-User, OAI-SearchBot, Claude-User, Copilot, etc.) and (2) GA4 Direct sessions for that same path, after subtracting visible AI referrers. We then fit a per-tier regression model to find the slope: how many extra Direct sessions appear per AI-bot hit. That slope, times the bot hits, equals the modelled hidden AI sessions.

What's the difference between visible and hidden AI?

Visible AI = sessions GA4 attributed to chat.openai.com, perplexity.ai, claude.ai, and so on. Hidden AI = sessions in your Direct bucket that our model attributes to AI-agent activity. We surface both side by side so you can see total AI-driven traffic, not just the slice GA4 happens to capture.

How long until the model gives me reliable numbers?

30 days is the minimum for a stable fit per URL-depth tier (homepage, depth-2, deep pages). The dashboard shows confidence bands (using bootstrap standard errors) so you can see when the signal is converging. Most customers see meaningful estimates after the first week and tight bands by week 4.

Do I need Google Analytics 4 connected?

Yes — the model joins GA4 channel data with our server-side bot hits. You connect GA4 via OAuth, we read aggregated `runReport` data only, and tokens are encrypted at rest. We never read individual user data and never store anything beyond the aggregated daily metrics needed for the regression.

What about real-time analytics?

Beyond attribution, the dashboard shows a live activity feed (10-second polling) of every AI crawler hit, a 90-day breakdown by platform (ChatGPT, Claude, Perplexity, Gemini, Bing Copilot), top pages by AI traffic, and content-gap analysis showing pages with high AI interest but no rules configured. Email reports go out weekly with the headlines.

ANALYTICS

See the AI traffic GA4 hides from you

ChatGPT and Claude send people to your site, but GA4 logs those visits as Direct. Wrenda models the hidden slice — using the bots that ChatGPT and Claude send to fetch your pages — and shows you total AI-driven sessions, conversions, and revenue.

See Your Hidden AI Traffic See all features

Google Analytics 4Server-log signalOLS regressionBootstrap CI

The whole picture, not just the visible slice

Two analytics layers most teams don’t have: hidden-AI attribution and live crawler telemetry.

Feature 01

AI Referral Attribution

When ChatGPT cites your page, two things happen: an AI bot fetches the URL, and (sometimes much later) a human visits your site with no referrer. We measure both, fit a daily regression per URL-depth tier, and tell you exactly how many of your Direct sessions, conversions, and revenue lines came from AI.

Per-domain dashboard with 90-day window
Tiered breakdown: homepage, depth-2, deep pages
Headline picks the metric you actually have data for (revenue → conversions → sessions)
Confidence bands so you know when to trust the number
Top 20 pages table sorted by total AI-attributable score

The model in one line

Direct_sessions(t) = β₀
                       + β₁ · AI_bot_hits(t)
                       + β₂ · Organic_sessions(t)
                       + ε

hidden_AI(t) = max(0, min(direct, β₁ · hits))

// Fit per (URL-depth tier, metric).
// Min n=30 days for a stable fit.

What you see in the dashboard

Last 90 days · Direct channel breakdown1,727 sessions

Visible AIGA4 source = chat.openai.com, perplexity.ai...

312 (18%)

Hidden AIModelled from server-log AI-agent hits

847 (49%)

Direct (residual)True direct traffic — typed URL, bookmark

568 (33%)

Per-tier fit (latest day)

Tier         β₁       ±SE       R²     hidden
homepage     1.42    0.24      0.71   +89
depth-2      0.87    0.18      0.64   +312
deep         0.51    0.12      0.58   +446

Headline:   1,159 hidden AI sessions
            (95% CI: 941 – 1,377)

Live activity feed

LIVE · polling 10s

6 hits · last 90s

14:32:08ChatGPT-User/pricingoptimize

14:31:54Perplexity-User/blog/api-keys-guideoptimize

14:31:42Googlebot/featurespre-render

14:31:31Claude-User/docs/getting-startedoptimize

14:31:19GPTBot/blogoptimize

14:31:02OAI-SearchBot/use-casespass-through

12.4K

Bot hits / day

47ms

Avg latency

94%

Cache hit rate

Feature 02

Real-time crawler telemetry

Every AI crawler hit is logged with full context — user agent, path, action taken, cache status, latency. The dashboard surfaces a live feed, a 90-day platform breakdown (ChatGPT, Claude, Perplexity, Gemini, Copilot), top crawled pages, and content gaps where bots are hitting URLs you don’t have rules for.

What you get

Activity feed — 10s polling, pause/play, color-coded actions
Platform breakdown — AI bots grouped by origin (OpenAI, Anthropic, Perplexity, Google, Microsoft, You.com)
Top pages — Sortable by hits, action, cache rate
Content gaps — Bots are crawling these URLs but you have no rule — fix it in one click

Example response — every hit, structured

{
  "timestamp": "2026-05-07T14:32:08Z",
  "user_agent": "ChatGPT-User/1.0",
  "platform": "OpenAI",
  "path": "/pricing",
  "action": "optimize",
  "cached": true,
  "latency_ms": 12,
  "response_status": 200
}

WORKFLOW

From signal to attributable revenue

Four steps. No tagging changes on your site. No JS to install.

Step 01

Connect Google Analytics 4

OAuth into the GA4 property for your domain. We pull aggregated runReport data only — sessions, conversions, revenue split by channel and source. Read-only, encrypted at rest.

Step 02

Capture every AI-bot fetch

Wrenda already sees every request to your site, so we already know when ChatGPT-User, Claude-User, Perplexity-User, OAI-SearchBot, and Copilot fetch each URL. No new collection needed.

Step 03

Fit the regression nightly

A scheduled job joins server-log AI hits with GA4 Direct + Organic sessions per (date, page-path), subtracts visible AI referrers, and fits a per-tier OLS model. Outputs slope, SE, R² per metric.

Step 04

Read the dashboard

Headline metric (revenue if you have it, else conversions, else sessions), tier cards with confidence bands, daily series chart, top 20 attributed pages. Refresh button forces a manual recompute (rate-limited).

USE CASES

What teams use it for

Three flows, same data foundation.

Prove AI search ROI

CMOs need to justify investment in AI optimization. Show a hidden-AI revenue number alongside organic search to make the case for content, schema, and technical-SEO budget.

Headline metric in dollars when revenue is tracked
Quarter-over-quarter trend chart
Confidence band so finance trusts the number

Diagnose Direct-traffic spikes

When Direct traffic jumps 30% overnight, was it a campaign, a TikTok, or AI assistants? The attribution model says exactly which page paths drove the change and how much came from AI bots.

Per-page attribution scores
Daily resolution lets you map spikes to events
Compare visible vs hidden AI per URL

Find content gaps to fill

The dashboard surfaces URLs that AI bots crawl heavily but you have no content rules for. Each one is a missed opportunity — a page that AI is sending traffic to without your optimization layer.

Top crawled pages with no rule attached
One-click rule creation from the gap list
Re-measurement after rules apply

ON YOUR INBOX

You shouldn’t need to log in to know

Scheduled email reports plus configurable alerts so the dashboard comes to you.

Alerts

Usage limit thresholds (warn at 75% / 90% / 100%)
Activity drops (AI bot traffic suddenly halves)
Error rate spikes (5xx above your baseline)
Cache performance regression (hit-rate falls below target)

Email reports

Daily, weekly, or monthly cadence
Pick metrics: AI hits, hidden AI revenue, cache rate, top pages
Filter by domain or send aggregate across portfolio
Multiple recipients per schedule

API EXAMPLE

Pull the numbers anywhere

Same data the dashboard renders, available as JSON.

ai-attribution.sh
# Get the 90-day attribution summary for a domain
GET api.wrenda.ai/ai-attribution/{domain_id}/summary
Authorization: Bearer <token>

# Response
{
  "headline": {
    "metric": "revenue",
    "total_ai_attributable": 48230,
    "visible_ai":  12410,
    "hidden_ai":   35820,
    "pct_of_direct": 0.41,
    "confidence_band": { "lower": 39860, "upper": 56600 }
  },
  "by_tier": [
    { "tier": "depth1",  "beta1": 1.42, "r2": 0.71, "hidden_revenue": 4910 },
    { "tier": "depth2",  "beta1": 0.87, "r2": 0.64, "hidden_revenue": 13280 },
    { "tier": "deep",    "beta1": 0.51, "r2": 0.58, "hidden_revenue": 17630 }
  ],
  "daily_series": [...]
}

90 days

Default attribution window

Cost beyond your existing GA4 quota

Nightly

Auto-recompute cadence

FAQ

Analytics FAQs

How AI attribution works, what data we read, and when the numbers become reliable.

When ChatGPT or Claude links a user to your site, the visit usually shows up in GA4 as Direct (no referrer). It looks like the person typed your URL — but they didn't. It came from an AI assistant. The Direct channel hides this traffic. We measure how much of your Direct sessions, conversions, and revenue is actually AI-driven by correlating it with the AI-bot hits we see hitting the same page paths.

Get Started

Stop guessing. Start measuring.

Connect your GA4 property and see how much of your Direct traffic is actually AI-driven — within 30 days of installing the proxy.

Start Free Trial

14-day free trial — No credit card required