Can an AI agentbuy from you?

Every visit to your store from an AI agent — Claude, ChatGPT, Perplexity, Gemini, and emerging agent surfaces. We classify each session by platform, build the funnel (landing → product → cart → checkout), surface where each platform drops off, and on demand run a real Claude agent through the same path on your live site to capture exactly what it tried and where it gave up. The scanner is the free entry point; the dashboard and replay are where teams keep working.

Hotjar, Contentsquare, and FullStory record sessions from human visitors. They cannot see what happens when an AI agent (Claude Desktop, ChatGPT, Operator) tries to use your site on behalf of a customer. Agents do not show up as human sessions, do not have user IDs, and often fail at structural problems your human-tested checkout never surfaces. Serge fills exactly that gap.

GEO tools measure whether ChatGPT mentions your brand in its answers. They sit upstream of the customer journey. Serge sits where the customer actually arrives — when an agent lands on your site and tries to find a product and add it to a cart, GEO tools cannot see what happens next. Serge can. The two are complementary, not competitive.

A lightweight JavaScript tag (under 6 KB gzipped) you add to your site. It detects sessions using behavioral signals, classifies agent traffic by platform (Claude, ChatGPT, Perplexity, Gemini, etc.), and sends structured event data to your Serge dashboard. It is first-party, avoids personal information, supports a no-cookie setup, and is designed for teams that need a low-friction install.

Click "Replay" on any flagged session in your dashboard. Serge spins up a real Claude agent in a managed headless browser, gives it the same task the original visitor seemed to be doing, and runs it against your live site. We capture every action, every request, every screenshot, and the inferred reasoning at each step — then hand you a shareable session URL you can forward to your engineers. It is not a recording of the original visitor; it is a fresh agent run that reproduces the failure mode.

No. The snippet only sees publicly available browser signals — no form values, no auth tokens, no PII. The active replay runs against the same public URLs anyone with a browser could visit. No credentials needed, no private data accessed. Consent requirements still depend on your setup and jurisdiction.

Commerce product, engineering, and front-end teams responsible for PDP, cart, checkout, or conversion. The product is currently shaped around stores where agent-assisted research already matters and where someone on the team feels the pain of unexplained "direct" traffic or silent conversion failures.

It is the starting point — a Product Director can run it in thirty seconds, screenshot the result, and forward it to the team. The scanner gives a single score and a fix list so the internal conversation can start. The dashboard, funnels, and replay are the paid product. They run on the same core logic, but the scan is the easiest way to get in.

Tier 1 signals (deterministic — known browser flags, DOM artifacts, signed RFC 9421 identity headers) have ~0% false positive rate. Tier 2 signals (heuristic — UA mismatches, CDP leaks) have 1-2%. Tier 3 signals (behavioral composite) are scored client-side and only the classification result is sent to the server. Combined, the system reaches >95% precision. Specific platforms (Claude, GPT, Perplexity, Gemini) are identified by signatures we maintain centrally.

Yes — we have to, to build the funnel and surface the trends. Data retention is tier-based: 30 days on Launch, 90 days on Grow, 12 months on Scale, custom on Enterprise. Anonymized benchmark statistics may be derived from cohort data, but raw events stay scoped to your workspace.

Most stores that already get meaningful organic or AI-assisted traffic see their first identified agent session within an hour of installing the snippet. The first daily briefing fires the morning after we have at least 24 hours of data.

The scanner produces a deterministic 0-100 snapshot for product findability and the next meaningful step — basically, how easy is it for an agent to find a product and keep moving. Read /methodology for the current scoring model and how it is evolving. The score is the entry point; the dashboard data is what actually moves the needle.