GA4 Medium Values Explained: AI Assistant, Native Traffic Classification, and Default Channel Rules

Google Analytics now gives recognized AI-assistant referrals their own acquisition classification. Since May 13, 2026, a visit from a recognized AI assistant can be assigned the native ai-assistant medium, the (ai-assistant) campaign, and the AI Assistants channel in GA4’s Default Channel Group. This guide explains what changed, what the Medium dimension actually means, which values GA4 creates automatically, which values come from advertising integrations or utm_medium, and why a medium should never be interpreted as if it were the final channel label.

Practical rule: use Session source / medium to inspect the acquisition signal that started a session, and use Session default channel group to understand how GA4 classified that signal. The two answer different questions. A row such as facebook.com / referral can still belong to Organic Social, while chatgpt.com / ai-assistant is now explicitly classified as AI Assistants.

Executive summary

Google introduced native AI-assistant traffic measurement in GA4 on May 13, 2026. According to Google’s Google Analytics release notes, when the referrer matches a recognized AI assistant, Analytics can automatically assign ai-assistant as the medium, place the traffic in the AI Assistants channel, and assign (ai-assistant) as the campaign. Google names services such as ChatGPT, Gemini, and Claude in the launch notice, while the current Default Channel Group documentation also describes AI Assistant traffic as coming from sources such as ChatGPT, Gemini, DeepSeek, Copilot, or Grok.

This is more important than simply adding one more row to Acquisition reports. Before the change, a recognizable AI referrer commonly behaved like an ordinary website referrer and could appear under Referral, while an AI click without usable referral information could fall into Direct. The new classification tells GA4 that a recognized AI referrer is a distinct acquisition type rather than a generic referral.

However, GA4 does not have a closed official list of “native mediums” that every property must use. Medium is a traffic-source text dimension. Its value can be produced in several ways: GA4 can infer it from referral information, an advertising integration can populate it through auto-tagging, or a marketer can provide it manually through utm_medium or campaign fields. The Default Channel Group then applies Google’s rules to source, medium, campaign, source platform, ad-network, and other available signals.

That distinction explains several common surprises:

• m.facebook.com / referral can be classified as Organic Social because the source is recognized as social, even though the medium is still referral.
• A manually tagged paid campaign can use a medium such as paid_social, yet GA4 determines the final paid channel from both the source and the paid-medium pattern.
• Google Ads traffic is not classified from cpc alone. For integrated traffic, GA4 can use source platform, ad network type, and campaign type.
• chatgpt.com / ai-assistant now has a purpose-built path into AI Assistants.
• (direct) / (none), (not set), Unassigned, and (data not available) describe different states and should not be treated as synonyms.

There is also a second 2026 change worth separating from the new medium. On June 11, Google added the new Source Group dimension and updated Source Platform classifications. Google says Source Group normalizes multiple source strings into a cleaner platform-level value, is populated retroactively, and includes built-in grouping for emerging sources such as ChatGPT and Perplexity. This does not replace Medium or the AI Assistants channel; it gives analysts another layer for normalizing where traffic came from.

Medium is not the same thing as a channel

Google defines Medium as the method by which traffic reaches a website or app. Common examples include organic, cpc, referral, social, email, and none. That is a useful definition, but it is easy to overread. Medium is one input in a larger acquisition model.

Dimension	What it answers	Typical example
Source	Which website, platform, app, or source sent the traffic?	google, facebook.com, chatgpt.com
Medium	What type or method of acquisition was recorded?	organic, cpc, referral, email, ai-assistant
Campaign	Which marketing initiative or campaign was associated with the traffic?	summer_sale, a Google Ads campaign name, or (ai-assistant)
Default channel group	Which broad rule-based marketing channel does GA4 assign?	Organic Search, Paid Social, Referral, AI Assistants
Source Platform	Which buying, publishing, or marketing platform family does GA4 identify?	Manual, Google Ads, DV360, and other recognized platform values
Source Group	Which normalized source identity should represent several messy source strings?	A normalized platform/source label instead of several variations of the same source

The practical consequence is simple: do not build reporting logic from Medium alone when GA4 already has a channel or source-platform dimension that answers the business question more directly. Medium is still essential for debugging UTMs and attribution, but it is not a universal channel taxonomy.

How GA4 populates the Medium dimension

Current GA4 documentation effectively gives analysts three main paths into the Medium dimension.

1. Automatic processing of referrers and direct traffic

When a user arrives without manually tagged campaign parameters or a usable advertising click identifier, GA4 can use the page’s referral information. Google’s campaigns and traffic-source documentation explains that the document referrer can be used to set traffic-source dimensions. Known search engines are processed as organic search, ordinary referring sites can be processed as referrals, and a visit with no usable referral information can become Direct.

This is the layer where values such as organic, referral, and (none) commonly originate. Since May 2026, ai-assistant has joined that automatic classification logic for recognized AI-assistant referrers.

2. Advertising integrations and auto-tagging

When GA4 is integrated with advertising platforms, acquisition data can be populated from platform-specific identifiers and metadata instead of relying only on the visible landing-page referrer. Google recommends auto-tagging for supported advertising integrations because it unlocks more dimensions than manual tagging. For Google Ads, current documentation shows google / cpc as the cross-channel source/medium value for auto-tagged traffic in common scenarios.

Importantly, integrated advertising traffic can be classified using more than the medium. The current Default Channel Group rules use source platform, Google Ads ad network type, campaign type, DV360 creative format, CM360 creative format, and SA360 engine-account type where those signals are available. This is why a single medium label is not sufficient to reconstruct every paid channel.

3. Manual tagging with UTM parameters

With manual tagging, utm_medium maps into GA4’s manual-medium dimensions and can also contribute to cross-channel acquisition reporting. Google’s URL builder documentation uses examples such as utm_medium=email and utm_medium=referral.

A manually tagged URL might look like this:

https://example.com/landing?utm_source=facebook&utm_medium=paid_social&utm_campaign=summer_sale

GA4 preserves the campaign values you send, but its Default Channel Group applies Google’s own classification rules afterward. That is why arbitrary UTM naming can produce fragmented Medium reports or an Unassigned channel even when the URL technically contains a valid utm_medium.

For a practical naming standard, see the metricfixer guide Manual UTM Tagging in GA4: Recommended Values That Preserve Default Channels.

What the new ai-assistant medium actually means

The new AI classification is unusually explicit compared with many older source rules. Google says the ai-assistant medium is assigned when the referrer matches a recognized AI assistant. The same traffic is assigned to the AI Assistants channel and receives the campaign value (ai-assistant).

This means a row such as chatgpt.com / ai-assistant should be read as measured traffic attributed to users arriving from ChatGPT. It is not a count of how many times an AI crawler fetched the site while preparing answers. Server-side crawler requests belong in server, CDN, WAF, or bot-management logs; GA4’s acquisition reports are not a crawler log.

The new channel also does not expose the user’s prompt. The documented traffic-source fields identify the referring source, medium, campaign, and broader channel classification. They do not provide the conversation text or query that caused the assistant to recommend the page.

Why the AI Assistants channel still does not equal all AI-driven traffic

The native AI channel depends on a recognizable acquisition signal. If a click reaches the site without usable referral information and without another campaign identifier, GA4 can only work with what the browser and measurement implementation provide. Google’s Direct documentation states that (direct) / (none) is used when traffic does not have a clear referral source, and it lists missing campaign information, redirects, and other technical causes that can remove attribution signals.

Therefore, the AI Assistants channel should be treated as a measurable subset of AI referral traffic, not a census of every visit influenced by an AI answer. A user can learn about a brand in an assistant and later type the URL, search for the brand, open a copied link that has lost referral context, or return in another session. GA4 cannot reconstruct that earlier AI influence from the final visit alone.

This is also why an increase in Direct traffic cannot be automatically relabeled as “dark AI.” Some of it may be unidentifiable AI referral traffic, but Direct also contains genuine typed URLs, bookmarks, offline-document clicks, stripped campaign data, blockers, and other acquisition gaps. The correct response is to improve evidence, not to rename the whole bucket.

Google AI Overviews and AI Mode are not in the AI Assistants channel

One of the most important details in Google’s current channel documentation is that Google AI Overviews and AI Mode are explicitly excluded from AI Assistants. Google places non-ad traffic from these Search experiences inside Organic Search.

That creates two different GA4 measurement categories for experiences that both use generative AI:

User discovery path	GA4 treatment	What analysts should infer
ChatGPT, Claude, Gemini assistant, Copilot, Grok, or another recognized external assistant	Can be classified as ai-assistant and AI Assistants when the referrer is recognized	External AI-assistant referral traffic
Google AI Overview or Google AI Mode result	Organic Search	Part of Google Search traffic, not the AI Assistants channel

So a dashboard labelled “AI traffic” that uses only the AI Assistants Default Channel Group will not represent all traffic originating from AI-generated interfaces. It specifically represents the channel definition Google has chosen for external recognized AI assistants.

GA4 medium values: automatic values and recognized channel patterns

The phrase “native medium” is useful informally, but it can be misleading if it suggests GA4 has a fixed dropdown of allowed media. A better model is to separate values GA4 can create automatically from values and patterns the Default Channel Group knows how to classify.

Common medium values GA4 can produce automatically

Medium value	Typical origin	Typical interpretation
ai-assistant	Recognized AI-assistant referrer	New native AI referral classification introduced in May 2026
organic	Known search-engine referral or other supported attribution processing	Unpaid search acquisition
referral	Recognized external referring site when no higher-priority campaign signal takes over	Referral method; final channel can still be Social, Video, or Shopping based on source classification
(none)	No identifiable source for a Direct visit	Used with (direct) for Direct traffic
cpc	Common cross-channel medium for Google Ads auto-tagged traffic	Paid click traffic; final channel can depend on richer advertising metadata

These are not the only values that can appear. Manual UTMs, imports, app campaign fields, and advertising integrations can produce many other media. The point is that the most familiar values are not all created in the same way.

Medium values and patterns recognized by the current Default Channel Group

Google publishes the current Default Channel Group rules. The table below focuses on the medium-related part of those rules for manual or cross-channel traffic. Source lists and integration metadata can still change the final result.

Medium value or pattern	Default-channel use	Important limitation
ai-assistant exactly	AI Assistants	GA4 sets this automatically when a recognized AI referrer matches. Do not use it casually as a custom campaign label.
(none) or (not set) with source (direct)	Direct	Direct requires the source/medium combination, not the medium alone.
organic exactly	Organic Search	Known search sources can also qualify even without this exact medium.
referral, app, link	Referral	A recognized social, video, or shopping source can be classified into its organic channel instead.
Paid-pattern media matching ^(.*cp.*|ppc|retargeting|paid.*)$	Paid Search, Paid Social, Paid Shopping, Paid Video, or Paid Other	The source category determines the specialized paid channel; otherwise the traffic can fall into Paid Other.
display, banner, expandable, interstitial, cpm	Display	These are explicit manual-traffic medium matches.
social, social-network, social-media, sm, social network, social media	Organic Social	Do not use an organic-social medium for paid social just because it contains the word “social.”
Any medium matching a pattern containing video	Organic Video	Recognized video sources can also qualify; paid video uses the paid-pattern logic plus source/platform information.
email, e-mail, e_mail, e mail	Email	Google also checks equivalent source values.
affiliate	Affiliates	Exact value in the published manual-traffic rule.
audio	Audio	Exact value in the manual rule; integrated platforms can also use creative-format metadata.
sms	SMS	Google also recognizes source sms.
Medium ending in push, or containing mobile or notification	Mobile Push Notifications	Source firebase is another qualifying rule.

There is no dedicated “Cross-network medium” requirement in the manual rules. Google can classify Cross-network from campaign information, while integrated Google Ads traffic uses platform and campaign metadata. Likewise, Organic Shopping can be identified from a recognized shopping source or a shopping-like campaign name without requiring a special organic-shopping medium.

Why source / medium and channel can look inconsistent

The apparent inconsistencies are usually the result of GA4 doing more classification than the Source / Medium row reveals.

Source / medium example	Likely default channel	Why
google / organic	Organic Search	Google is a recognized search source and the medium is organic.
facebook.com / referral	Organic Social	Facebook is on GA4’s social-source list even though the medium can remain referral.
youtube.com / referral	Organic Video	YouTube is recognized as a video source.
amazon.com / referral	Organic Shopping	A recognized shopping source can qualify without a dedicated shopping medium.
facebook / paid_social	Paid Social	The source is social and the medium matches Google’s paid pattern.
newsletter / email	Email	The medium matches the Email rule.
chatgpt.com / ai-assistant	AI Assistants	The new native medium exactly matches the AI Assistant rule.
(direct) / (none)	Direct	No clear referral source is available.

This is why acquisition audits should compare at least two columns together: Session source / medium and Session default channel group. When paid-media integrations are involved, add Source Platform and the available platform-specific campaign dimensions as well.

UTM medium best practices under the current rules

The fact that GA4 accepts almost any text in utm_medium does not mean every value is equally useful. A technically valid custom medium can still fragment reporting or fail to match a Default Channel Group rule.

• Use a controlled lowercase taxonomy. Channel matching is not case-sensitive, but the raw Medium dimension can still become harder to read when teams mix values such as Email, email, and EMAIL.
• Use values that align with Google’s published rules when you want native channel reporting. For example, email, affiliate, sms, and properly designed paid-media values are easier to maintain than arbitrary labels.
• Do not use newsletter as a replacement for email unless another source rule deliberately classifies it. Put newsletter detail in utm_campaign, utm_content, or another campaign field instead.
• Do not manually tag ordinary internal links. Internal UTMs can overwrite acquisition context and create artificial campaign changes in event-level attribution data.
• Keep Google Ads auto-tagging enabled where supported. Google’s own documentation recommends platform integrations and auto-tagging because they provide richer dimensions than manual UTMs alone.
• Do not force utm_medium=ai-assistant on unrelated traffic. The current default rule will classify an exact ai-assistant medium into the AI Assistants channel, so using it as a generic label can contaminate the native category.

If a property already contains a large number of non-standard values, first inventory the actual Session source / medium pairs before changing naming conventions. A clean taxonomy should reduce future fragmentation without erasing the meaning of historical data.

First user medium, Session medium, and event-scoped Medium

Even a perfectly classified medium can look “wrong” if the wrong scope is being inspected. Google separates traffic-source dimensions into user, session, and event scopes.

First user medium

First user medium describes how the user was first acquired. If a person first discovers the site through a recognized AI assistant and later returns through Direct, their First user medium can remain ai-assistant. This dimension belongs naturally in the User acquisition report.

Session medium

Session medium describes how the current session began. This is usually the best dimension for operational traffic analysis because it answers, “What brought this visit to the site?” It belongs in the Traffic acquisition report.

Event-scoped Medium

The unprefixed Medium dimension is used in event-level attribution contexts. Google notes that for non-key events, event-scoped source and medium can be (not set). It should not be used as a substitute for Session medium when the objective is to count visits by acquisition source.

For a deeper explanation of acquisition and conversion timing, see GA4 Conversion Attribution Models Report: Interaction Time vs Conversion Time.

Special values that are often mistaken for mediums

GA4 uses several placeholders and channel labels that look similar in a table but mean different things.

Value	Where it appears	What it means
(none)	Medium	The normal medium paired with (direct) when no clear source is identified.
(not set)	Many dimensions	No value was received or populated for that dimension at the requested scope. For Session source / medium, a missing session_start is one documented cause.
Unassigned	Channel group	GA4 received traffic-source information but it did not match any channel rule in the channel group being viewed.
(data not available)	Source and medium traffic-source dimensions	Google has information but has not finished processing it yet in certain reporting situations; some event-level daily values may later update.
(other)	Aggregated reporting row	A cardinality aggregation row, not a traffic source.

These distinctions matter during debugging. A large Direct row points to missing or genuinely absent source information. Unassigned means the source information exists but does not fit the channel rules. (not set) means the requested dimension is missing. (data not available) can be a processing state rather than a tagging defect.

For a diagnostic workflow, see Why GA4 Shows (direct), (not set), and Unassigned: A Diagnostic Decision Tree.

The other 2026 change: Source Group and updated Source Platform

On June 11, 2026, Google introduced Source Group as a new dimension concept. The goal is to consolidate multiple source strings that represent the same platform into one cleaner reporting identity. Google’s launch example is the problem of seeing several Facebook or Instagram naming variations instead of one normalized source.

Google also says the update is designed to be future-proof for emerging traffic sources and specifically cites ChatGPT with OpenAI and Perplexity as examples. Unlike the AI-assistant medium launch note, the Source Group announcement explicitly says the new grouping is populated retroactively.

This gives analysts four useful layers for AI traffic:

• Source: the detailed referring source that GA4 recorded.
• Source Group: a normalized source identity where Google has a grouping rule.
• Source Platform: the broader platform family.
• Medium / channel: the acquisition method and marketing-channel classification.

These should be used together rather than treated as competing dimensions. Source Group is particularly useful when one platform can generate several source strings over time, while ai-assistant answers the separate question of how that traffic was classified.

How to analyze AI traffic before the native channel existed

Before May 2026, analysts often built a custom channel around known AI referral sources. Google still documents this approach in its Custom Channel Groups guide, with examples based on ChatGPT, Gemini, Microsoft Copilot, Claude, and Perplexity.

Custom channel groups remain useful for history because Google says they can be applied to reports retroactively. The practical approach is:

1. Create or copy a custom channel group.
2. Add an AI-assistant channel based primarily on Source values for the AI platforms you want to include.
3. Place the AI rule above Referral so matching historical referrals are not captured by the more generic referral rule first.
4. Validate the resulting source list against actual historical Session source / medium rows.
5. Keep the native AI Assistants channel as the forward-looking default view, while the custom group provides a consistent historical comparison.

Do not assume that every old chatgpt.com / referral session will be rewritten into chatgpt.com / ai-assistant. Google explicitly states that Source Group is retroactive, but the May 13 AI-assistant release note does not make the same statement about backfilling the Medium dimension. For historical reporting, a retroactive custom channel group is therefore the safer method.

A practical GA4 report for AI-assistant traffic

For most websites, the most useful starting report is not simply “how many AI visits did we get?” but “which assistants sent traffic, where did that traffic land, and did it create valuable sessions or conversions?”

In Reports → Acquisition → Traffic acquisition, start with Session source / medium and compare it with Session default channel group. Then use an Exploration or customized report to add:

• Session source / medium;
• Session default channel group;
• Source Group where available;
• Landing page + query string;
• Sessions;
• Engaged sessions or engagement rate;
• Key events;
• Total revenue for ecommerce properties;
• New users if acquisition quality matters.

For AI analysis, filter the native channel to AI Assistants, then break down by source. This shows whether ChatGPT, Claude, Gemini, Perplexity, Copilot, or another recognized assistant is actually sending measurable visitors and which landing pages receive them.

Also compare the same landing pages against Organic Search. The comparison is useful because a page can receive traffic from both traditional search and AI recommendations, but the acquisition mechanics are different. Do not compare only volume; compare engagement, key-event rate, revenue, and the type of landing page.

What GA4 cannot tell you about AI visibility

The new channel improves referral attribution, but it does not turn GA4 into an AI visibility platform.

• GA4 does not tell you how many times an assistant mentioned or cited your brand when no click occurred.
• GA4 does not expose the user’s prompt through the documented traffic-source fields.
• GA4 cannot recover an AI origin after referral information is lost unless another measurable campaign or platform signal survives.
• GA4 cannot tell whether a later Direct or branded-search visit was influenced by an earlier AI answer that never generated a measurable click.
• GA4 acquisition reports are not a substitute for server logs when you need to study AI crawlers, fetchers, or agents requesting pages directly.

The correct interpretation is therefore: GA4 measures attributable site visits and downstream behavior; it does not measure the full upstream visibility of your content inside AI systems.

GA4 medium audit checklist

• [ ] Review Session source / medium rather than an unscoped Medium dimension when the question is session acquisition.
• [ ] Compare Session source / medium with Session default channel group.
• [ ] Check whether recognized AI referrals now appear with ai-assistant.
• [ ] Separate Google AI Overviews and AI Mode from external AI assistants; Google classifies the former as Organic Search.
• [ ] Inventory manually tagged utm_medium values and compare them with Google’s current Default Channel Group rules.
• [ ] Standardize manual media in lowercase and remove redundant synonyms.
• [ ] Keep supported advertising auto-tagging enabled and avoid replacing rich platform attribution with manual UTMs alone.
• [ ] Investigate unusual (direct) / (none), (not set), Unassigned, and (data not available) separately.
• [ ] Use Source Group to reduce source-name fragmentation where the dimension is available.
• [ ] Use a retroactive custom channel group for historical AI referrals rather than assuming the new ai-assistant medium was backfilled.
• [ ] Analyze AI traffic by landing page, engagement, key events, and revenue rather than by session count alone.

Bottom line

The arrival of ai-assistant is a meaningful change because Google has promoted external AI assistants from a special-case referral analysis into its native traffic-source model. A recognized AI referral can now arrive with its own medium, campaign label, and Default Channel Group, making standard GA4 reports much easier to interpret.

But the larger lesson is about GA4 acquisition architecture. Medium is not a channel, and channel is not a source. Source identifies where traffic came from. Medium describes how the acquisition was recorded. Default Channel Group applies Google’s classification rules. Source Platform and the new Source Group add platform-level normalization. User, session, and event scopes then decide which acquisition moment the report is actually describing.

Once those layers are separated, rows such as facebook.com / referral, google / cpc, (direct) / (none), and chatgpt.com / ai-assistant stop looking inconsistent. They are different observations at different layers of the same acquisition system.

Related metricfixer guides

• Manual UTM Tagging in GA4: Recommended Values That Preserve Default Channels
• Why GA4 Shows (direct), (not set), and Unassigned: A Diagnostic Decision Tree
• Why Google Ads Clicks Do Not Match GA4 Users, Sessions, or Page Views
• GA4 Conversion Attribution Models Report: Interaction Time vs Conversion Time

Methodology and sources

This article is based primarily on current Google Analytics documentation and release notes reviewed in August 2026. The review separates documented platform behavior from practical interpretation. Google’s published Default Channel Group rules were used for the medium-pattern table; Google’s traffic-source documentation was used for source, medium, scope, manual-tagging, auto-tagging, Direct, and placeholder behavior; and the 2026 release notes were used for the AI Assistants and Source Group updates. Recommendations about UTM governance, historical reporting, and audit workflow are operational guidance derived from those documented rules and common analytics-debugging practice.

• Google Analytics: What’s new in Google Analytics
• Google Analytics: Default channel group
• Google Analytics: Traffic-source dimensions
• Google Analytics: About traffic-source dimensions
• Google Analytics: Scopes of traffic-source dimensions
• Google Analytics: Campaigns and traffic sources
• Google Analytics: Traffic-source dimensions, manual tagging, and auto-tagging
• Google Analytics: URL builders and UTM campaign parameters
• Google Analytics: Custom channel groups
• Google Analytics: Understand (direct) / (none) traffic
• Google Analytics: Tagging best practices to avoid Unassigned, (not set), and Direct issues
• Google Analytics: What the value (not set) means in reports
• Google Analytics: What (data not available) means in reports
• Google Analytics: About the Source Platform dimension

This article is for analytics, measurement, and reporting guidance only. metricfixer is not affiliated with Google, Google Analytics, OpenAI, ChatGPT, Anthropic, Claude, Microsoft, Copilot, Perplexity, xAI, Grok, DeepSeek, or other platforms mentioned in the article. GA4 traffic-source rules, source lists, advertising integrations, referrer behavior, and reporting interfaces can change after publication. Validate important attribution decisions against current platform documentation and your own collected data.