Fathom Adds AI Traffic Analytics, Luna and Historical Data Cleanup

In October 2026, Fathom Analytics introduced three updates to its web analytics platform: a dedicated report for visits from AI services, an intelligent assistant called Luna for analyzing statistics through natural language, and a tool for removing unwanted traffic from historical data.

The updates were published on October 2, 6, and 7, respectively. They address three tasks: measuring referrals from generative search engines, making analytics more accessible, and improving data quality by excluding test and invalid visits.

A Dedicated AI Traffic Report for Referrals From AI Services

On October 2, Fathom added an AI tab to the Referrers section. Visits from AI assistants now appear as a separate category instead of being distributed exclusively among Organic, Paid/Campaign, and other traditional sources.

The system recognizes referrals from ten services: ChatGPT, Perplexity, Claude, Gemini, Microsoft Copilot, DeepSeek, Grok, Meta AI, You.com, and Poe. For each source, it displays visitor and pageview counts. Users can select an individual AI service and filter the entire analytics dashboard, including visited pages and events.

For example, a marketer can see how many visitors came from ChatGPT, which content they opened, and how many desired actions they completed. The new AI category is also available in the overall Sources classification alongside Direct, Organic, Social, and Paid/Campaign.

How Fathom Identifies AI Traffic Sources

Fathom uses two main signals to identify referrals from AI services. The first is the HTTP Referrer passed by the browser when a visitor follows a link from the relevant site. The second is the utm_source parameter when its value identifies a recognized AI service.

For example, ChatGPT referrals can be identified through the chatgpt.com and chat.openai.com domains or through utm_source=chatgpt.com. This matters because mobile and desktop AI applications do not always pass a Referrer when opening an external link. Fathom groups recognized visits from the same service regardless of the method used to identify their source.

Paid referrals from AI services are not mixed with regular AI Referral Traffic. If a link contains utm_medium or utm_campaign, the visit remains in the Paid/Campaign category. This helps separate organic referrals resulting from AI assistant recommendations from paid acquisition when ad links are tagged correctly.

AI Traffic Is Available for Previous Periods

The new classification applies to the entire history of data already collected by Fathom. After the AI tab appears, users do not need to reinstall the analytics script or wait for new data to accumulate. They can select Last 365 Days or All Time and examine AI referrals from earlier periods.

Historical visits are automatically redistributed among source categories. As a result, some Organic and Paid/Campaign figures may decline because part of that traffic is now counted as AI. Total visitor counts do not change: this is a new classification of existing data, not an addition or deletion of visits.

Marketing reports therefore need to account for changes in source distribution when comparing old and new exports. For example, a decline in Organic after the AI tab launches does not necessarily indicate an actual drop in search traffic. Some of the change may reflect visits moved into the new category.

Filtering AI Traffic and Using the API

The new classification is not limited to standard reports. Advanced Filters now include Referrer Source = AI and AI Source, which allow users to select a specific AI service. For example, they can filter visits from Perplexity and examine landing pages, countries, devices, and completed events.

Fathom also updated data exports: standard report downloads now include AI.csv. Custom exports can use the AI Source dimension. The API provides an ai_source parameter for grouping and filtering pageviews and events by AI service; the existing referrer_source dimension also returns AI for qualifying visits.

This allows companies to include AI Referral Traffic in their own BI dashboards and acquisition performance reports.

Limitations of AI Traffic Measurement

Fathom explicitly states that the new feature measures website visits from AI services, not brand visibility within AI answers. It does not show how often ChatGPT or Perplexity mentioned a company, which prompts surfaced its content, or which competitors appeared alongside it. Those tasks require separate AI Search Visibility Monitoring tools.

Referrals cannot always be identified correctly. If an AI application sends no Referrer and the link lacks a recognized utm_source, the visit is classified as Direct. AI Traffic figures therefore represent detected referrals and do not guarantee full coverage of everyone who reached the site after interacting with AI search.

This limitation should be considered when comparing AI traffic with organic search and evaluating the contribution of GEO optimization to website visits.

Luna: An AI Assistant for Web Analytics Questions

On October 6, Fathom launched Luna, an integrated AI assistant accessed through a button in the lower-right corner of the analytics interface. Luna answers questions about using Fathom, configuring analytics, and metrics from connected websites.

For example, a user can ask how to track button clicks, why no visitors appear in a report, or which plan their account uses. The assistant consults Fathom's documentation and current support knowledge base.

For analytics questions, Luna can access account statistics only after the user grants separate permission. If an answer requires website data, Luna asks the user to confirm read access with the Allow button. Until approval, those metrics are not shared with the assistant.

Analyzing Statistics With Natural-Language Requests

After access is granted, Luna can answer questions about visitors, traffic trends, and event performance. For example, a user might ask: 'How many visitors came to the site last month, and how did that compare with the previous month?' The assistant retrieves the relevant data and can present results in a table or chart.

More detailed requests are supported, including source comparisons, page analysis, event analysis, and date-based breakdowns. For example, a marketer can ask which sources generated the most registrations in the past week. Luna finds the corresponding event and breaks it down by Referrer, UTM Source, or landing page.

Queries can contain two conditions at once. For instance, a user can retrieve daily registration counts from chatgpt.com for a selected week. Within a single answer, Luna can make up to ten statistical queries. Previously retrieved data can be reused in subsequent follow-up questions in the same conversation.

Realtime Analysis and Cross-Site Comparisons

Luna can also work with the Realtime report. Users can request current activity, including the number of visitors on the site, the pages they are viewing, and their traffic sources. This uses Realtime statistics for the last five minutes.

For historical data, Luna supports daily and monthly breakdowns covering up to 12 months per request. It can also compare multiple websites when the Fathom plan grants access to the relevant properties in the account.

For example, a company operating several online stores can request a comparison of visitors and pageviews across sites. Luna can generate a results table suitable for subsequent CSV export.

This does not mean unrestricted access to data for any period or every API dimension. The assistant uses a defined set of supported statistical requests.

Luna Permissions and Escalation to Support

Luna's permission to read analytics applies only within a particular conversation. When a new chat begins, the user must approve access to statistical data again.

If the user does not want to share metrics with the assistant, Luna can instead provide a link to the relevant dashboard view with dates and filters already selected. The user can then inspect the statistics independently.

Conversations are saved in Past Chats for later review. The Escalate to a human command sends a conversation to a Fathom support specialist, who receives the chat history and responds by email.

Fathom specifically warns that the AI assistant can make mistakes. Analytics conclusions should therefore be checked against the original reports, especially before business decisions are made.

Remove Unwanted Visits: Cleaning Historical Data

On October 7, Fathom introduced Remove Unwanted Visits, a tool for excluding unwanted visits from historical statistics. It addresses situations where employee test visits, staging-site activity, unwanted bot traffic, or invalid referrals have entered analytics reports.

Previously, those visits could continue to distort reports after testing ended. Site owners can now select a date range, apply filters, preview matching visits, and exclude them from analytics.

For example, a company can remove visits to a test site that accidentally used the tracking code of its main domain. Another use case is excluding unwanted visits to a specific page by automated scanners. Unlike an ordinary dashboard filter, the new operation changes the set of visits counted in subsequent reports.

Previewing Visits Before Removal

To begin cleanup, a site owner opens Settings, selects the site, and navigates to Manage → Data removals. The Create data removal command opens the dashboard in a special mode for selecting a period and configuring filters.

For example, a user can select visits to a particular page from a specific country or Referrer. After clicking Next, Fathom displays a preview showing the number of matching visitors and pageviews, along with the main pages, countries, and sources associated with those visits.

This lets the owner check whether legitimate visitors were included. If the selection is too broad, the operation can be canceled and the filters refined. After confirmation, Fathom recalculates the data. Longer periods are processed progressively, with a notification when processing finishes.

How Removal Affects Reports and Events

Remove Unwanted Visits excludes entire visits rather than individual pageviews. When a visit matches the chosen filters, Fathom excludes the corresponding visitor's activity for that entire day within the selected period. This includes other pages viewed that day and events completed during the visit.

For example, if a test user visits /pricing, then browses the catalog and completes a registration event, removing the matching visit excludes all associated activity. The preview can therefore show more pageviews than the original dashboard filters suggest.

Excluded data stops counting toward Fathom's interface, exports, email reports, and API. The associated pageviews also stop contributing to the service's usage volume.

The tool works with visits, not individual events. It cannot selectively delete one invalid conversion while retaining all other activity from the same visit.

Restoring Data With Undo

Despite its name, Remove Unwanted Visits does not irreversibly destroy excluded visits. Fathom provides an Undo option in the Data removals section. If the owner discovers that real visitors were excluded, Undo restores their data and Fathom recalculates the reports.

The previously excluded visits, pageviews, and events then return to the statistics. This matters for historical analytics, where an accidental removal could distort performance metrics. Nevertheless, the ability to restore data does not eliminate the need to check filters carefully before applying them.

Only one removal operation can run at a time for each site. At the account level, there is a limit of 30 removal or restoration operations per day.

Data Removal Limitations

Remove Unwanted Visits has several restrictions. First, only the site owner can use it. People who have shared dashboard access cannot remove statistics themselves.

Second, data previously imported from Google Analytics or Gauges cannot be removed. Third, Referrer Source and AI Source filters cannot be used to select visits for removal. Cleanup must instead use other criteria, such as pages, countries, or specific Referrer Hostnames.

The operation is also unavailable while viewing Google Search Console data. Removal covers visits that occurred before the selection was previewed; new visits arriving afterward remain in reports.

To keep unwanted traffic from appearing again, Fathom recommends configuring its Firewall to block the relevant sources, countries, or IP addresses. Data removals thus correct historical statistics, while the Firewall prevents unwanted visits from being collected in the future.

Fathom Analytics: AI Referrals and Better Web Analytics Data

Fathom Analytics is a privacy-focused web analytics platform designed to simplify the measurement of website visits, traffic sources, and user events. Its October updates expand the platform in three areas.

AI Traffic helps measure detected referrals from generative search engines and AI assistants, including their contribution to tracked events. Luna offers a natural-language interface for analytics queries, with separate permission to read statistics. Remove Unwanted Visits helps clean historical data of test and unwanted visits without irreversible deletion.

For ecommerce businesses, these features can help analyze new customer acquisition channels and validate conversion data quality. Service companies can use them to investigate lead sources and exclude test traffic. B2B companies can evaluate AI-search referrals to product, educational, and lead-generation pages.

Fathom does not replace specialized GEO platforms for monitoring brand mentions in AI answers, and Luna does not replace source-data verification or a comprehensive marketing attribution system.

Sources

1. Fathom — AI Traffic Tab, October 2026
2. Fathom — Luna AI Assistant, October 2026
3. Fathom — Remove Unwanted Visits, October 2026
4. Fathom Documentation — Sources
5. Fathom Documentation — Data Removals