Separate AI bot hits from customer traffic

A server can record a search crawler fetching a page. It can also record a page fetched after a person asks an assistant a question. A web analytics tool may later record a tagged referral session. The site can record an action during that session, while a CRM, scheduler, or order system records the business result.

Those records answer different questions. A crawler request shows machine access. A user-triggered fetch shows that a person caused an assistant to retrieve a page. A referral session shows that a visit reached the site and was attributed to a source. An on-site action shows what happened during a visit. A business outcome shows what the business system confirmed.

Keep the totals separate first. Connect them only when a shared identifier or another documented method supports the join. Without that connection, five rows are more accurate than one funnel.

Terms used in the worksheet

• An HTTP request is one attempt to fetch a web address or resource from a server.
• A content delivery network (CDN) is infrastructure that serves or filters site requests before they reach the site's main server. A server or CDN log is the stored request record.
• A user agent is identifying text supplied with a request. Treat it as classification evidence, then record any network or vendor verification used to support the identity.
• A referrer is the page or service recorded as sending a visit. A UTM source is a source label carried in a link parameter, such as utm_source=chatgpt.com.
• An identifier is a value used to connect records. A session identifier can connect events within one measured visit; a business-system identifier can connect a submitted form to a lead, booking, or order when policy allows it.
• Scope states what an analytics value applies to, such as a first user, one session, or one event.
• A customer relationship management system (CRM) stores and manages lead or customer records. A system of record is the system the team accepts as authoritative for a named result.

Server and CDN logs count HTTP requests. A page can be requested more than once, retried, redirected, blocked, or requested alongside other resources. Preserve the path, response status, timestamp, user agent, and available verification evidence before reducing the log to a class total.

A bot dashboard applies the vendor's classification rules. Cloudflare, for example, distinguishes behaviors such as Search, Agent, Training, and Transact, and documents several bot-verification methods.[6] That category can help classify a request. It does not create a web session or customer record.

Analytics reports count according to their own collection and attribution rules. Google Analytics distinguishes user-, session-, and event-scoped acquisition dimensions.[5] A session-source report and a first-user-source report can therefore show different totals without either being a server-log count. Consent choices, tag loading, redirects, cross-domain movement, filters, and date boundaries can add more gaps.

Business systems answer another question. A CRM may confirm a qualified lead after the original session ends. A scheduler may own the booking state. An order system may own payment completion. Record the identifier and join rule that connect these systems. If the connection is absent, report the outcome separately.

Mismatch investigation checklist

1. Confirm both reports use the same inclusive dates and timezone.
2. Name the unit in each report: request, unique path, session, user, event, lead, booking, order, or revenue amount.
3. Save the exact filter, user-agent rule, bot category, source dimension, event name, and status rule.
4. Check redirects, response codes, consent behavior, cross-domain steps, internal traffic, monitoring, and known spam or test traffic.
5. Confirm whether the source dimension is first-user, session, or event scoped when the analytics tool makes that distinction.
6. Check whether a shared identifier exists and what percentage of eligible records it covers.
7. Leave the mismatch unresolved when the available evidence cannot settle it.

This worksheet records what each measurement system observed during a date range. If the review instead reveals that a service detail, location, price, policy, or other public fact is missing or inconsistent, use the AI Search Source Map. That worksheet traces one claim across owned pages, outside sources, and a dated assistant answer. It does not replace this event classification.

For the field-note explanation behind this worksheet, read OpenAI reported a 1,223% jump in search and bot hits. Do not call it customer traffic.

• □Every total has a named source system, date range, timezone, unit, and filter.
• □Search crawler requests and user-triggered fetches remain separate rows.
• □Referral sessions use the analytics tool's documented source and scope fields.
• □On-site actions name the actual event or completed interaction.
• □Business outcomes come from the system that confirms them.
• □A join names the shared identifier or documented method and reports coverage.
• □Missing identifiers and unresolved classifications remain visible.
• □The report does not sum requests, sessions, actions, and outcomes.
• □The report does not claim causality, incrementality, ranking, sales, or revenue without separate evidence.
• □Shared files exclude unnecessary customer data, full IP addresses, credentials, payment data, and private message content.

1. [1] Overview of OpenAI Crawlers — OpenAI; accessed 2026-09-04.
2. [2] Publishers and Developers FAQ — OpenAI; accessed 2026-09-04.
3. [3] ATV Big Air Tour case study — OpenAI; published 2026-09-02; vendor-published customer result.
4. [4] Traffic-source dimensions — Google Analytics; accessed 2026-09-04.
5. [5] Scopes of traffic-source dimensions — Google Analytics; accessed 2026-09-04.
6. [6] Verified bots — Cloudflare; last updated 2026-07-01; accessed 2026-09-04.
7. [7] OpenAI Bot Hits vs ChatGPT Referral Traffic — BaristaLabs field note; accessed 2026-09-04.
8. [8] AI Search Source Map — BaristaLabs resource; accessed 2026-09-04.

The OpenAI ATV Big Air Tour case is a vendor-published customer result, not an independent audit. Its combined bot-hit values are not reused in the blank worksheet or constructed example.

Bring one server or bot report, one referral report, and the business action you want to verify. BaristaLabs can help classify the evidence, inspect the joins and gaps, and define the next measurement change through AI-assisted website development.