Share of answer is the operating metric for this practice.

How we count it
Request an audit

Insight · 13 min

AI search shows up in your pipeline under other names

AI search attribution is the work of connecting what ChatGPT, Perplexity, Gemini, Copilot, Claude, and Google’s AI features say about you to visits, leads, and pipeline. Only part of that path is visible. Some clicks arrive with a referrer or a tag, and many do not. Many buyers never click, then turn up later as direct visits, branded searches, or a sales call. As of November 2026, GA4 has a default AI Assistant channel. It still undercounts, so B2B teams need four sources read together: a custom AI channel, a self-reported “how did you hear about us” field, CRM capture, and branded search trends.

In short

  • Referral data counts clicks that carry a referrer or tag. Answers with no click, app traffic, and stripped referrers land in Direct, Unassigned, or nowhere.
  • GA4 added an AI Assistant default channel on May 13, 2026. It excludes AI Overviews and AI Mode, and Google does not publish its full source list. Build a custom channel group anyway, and place the AI rule above Referral.
  • A free-text “How did you hear about us?” field catches influence that analytics misses. Store the answer on the contact and carry it to the opportunity.
  • Branded search growth in Search Console is a proxy for AI exposure. It is not proof.
  • Conversion-rate studies of AI traffic disagree, from parity to 3x and higher. Do not put a multiplier in a forecast.
  • Report exposure, visits, and pipeline as separate rows, and label each one observed or inferred.

Why does AI search traffic show up as direct?

AI traffic shows up as direct when the visit arrives with no referrer and no campaign tag. Analytics tools name a source from those two signals. When both are missing, GA4 files the session under Direct.

Four things strip the signals.

  1. Apps. A link opened from a mobile or desktop assistant app often carries no referrer. The visit looks like someone typed the URL.
  2. Link attributes and policy. A link marked rel="noreferrer" sends no referrer. Without any policy set, browsers default to strict-origin-when-cross-origin, which sends only the origin (for example https://chatgpt.com/), never the conversation URL. You can learn which assistant sent a visit. You cannot learn the prompt.
  3. No click at all. The buyer reads the answer, remembers the name, and comes back later through a bookmark, a typed URL, or a Google search for the brand. That visit is direct or branded organic. The answer did the work, and analytics gives it no credit.
  4. Copy and paste. A URL copied out of an answer and pasted into a new tab has no referrer.

The third case is the largest in B2B, and no tag can fix it. That is why AI search attribution cannot rest on analytics alone.

What does each AI engine pass to analytics?

Only OpenAI documents its referral tagging. The behavior of the other engines described below comes from observed traffic and practitioner compilations, not vendor documentation, and it changes. Check your own logs.

EngineWhat arrives, when it arrivesDocumented by the vendor?Common failure
ChatGPTutm_source=chatgpt.com on referral URLs, plus a chatgpt.com referrer from the web appYes: OpenAI’s publisher FAQ says ChatGPT “automatically includes the UTM parameter utm_source=chatgpt.com in referral URLs”App visits arrive as direct. A tag with no utm_medium can land in Unassigned
Perplexityperplexity.ai referrer from the webNoApp visits lose the referrer some of the time
Geminigemini.google.com referrer from the webNoApp visits largely arrive as direct
Microsoft Copilotcopilot.microsoft.com referrer from the webNoEdge and Windows surfaces can send other hostnames or none
Claudeclaude.ai referrer, inconsistentlyNoDesktop and mobile app visits arrive as direct
Google AI Overviews and AI ModeGoogle search referrerGoogle counts these clicks in Search ConsoleIndistinguishable from organic search in GA4

The reliability notes for Perplexity, Gemini, and Claude follow a May 2026 reference table compiled by SearchPilot from published research. Treat them as reported, not measured by us.

Tags can also disappear for a while. On October 14, 2026, Search Engine Roundtable reported that ChatGPT links had lost both the UTM tag and the referrer. The source of the report later described it as a bug, and tagging returned. Annotate gaps like this in your report. Otherwise a tracking outage reads as a demand drop.

Does GA4’s default channel group include an AI channel?

Yes. As of November 2026, GA4’s default channel group includes a channel named AI Assistant. Google’s “What’s new” page dates it to May 13, 2026, and says you “can now identify how users are discovering your site through chatbots like ChatGPT, Gemini, and Claude.”

Google’s default channel group help page defines it this way:

  • “AI Assistant is the channel by which users arrive at your site from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok.”
  • A session qualifies when the medium exactly matches ai-assistant, or when the referrer matches Google’s list of AI assistants. GA4 then sets the medium to ai-assistant and the campaign to (ai-assistant).
  • It “excludes Google’s AI Overviews and AI Mode.” Those clicks stay in Organic Search, which Google defines as including “Google’s AI Overviews and AI Mode.”

Three limits matter.

  1. Google does not publish the referrer list. The channel page names five assistants. The “What’s new” entry names Claude, which the channel page does not. Neither names Perplexity. Perplexity appears only in a separate June 11, 2026 feature, the Source Group dimension.
  2. It cannot see visits that arrive with no referrer and no tag.
  3. It puts Google’s own AI surfaces in Organic Search, so they never appear in the AI number.

If you have read that GA4 has no AI channel, that was true before May 2026.

How do you set up an AI channel in GA4?

Build a custom channel group that copies the default group, add an AI channel matched on source, and move it above Referral. You keep a list you control, and the default AI Assistant channel stays in place for comparison.

  1. In Admin, open Data display → Channel groups, and copy the default channel group. Standard properties allow two custom groups. Analytics 360 allows five.

  2. Add a channel named “AI assistants”. Set the condition to Source matches regex, or Medium exactly matches ai-assistant, using an OR.

  3. Use a regex anchored to exact hostnames. A starting point:

    ^(chatgpt\.com|chat\.openai\.com|(www\.)?perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai|chat\.deepseek\.com|grok\.com)$

  4. Reorder. Google’s help page says traffic “is included in the first channel whose definition it matches given the current order of channels in the group.” An AI rule below Referral never fires, because Referral catches the session first.

  5. Save, then compare the custom channel with the default AI Assistant channel for the same dates. The gap shows what Google’s list misses.

Keep the regex narrow. Google’s own help-page example opens with ^.*ai|. That alternative matches any source containing the letters “ai”, including mail.google.com and most email platforms. Bare tokens like gpt cause the same problem. Review the list each quarter, and add any new hostname you see in your referral report. Google says custom channel groups apply to reports retroactively, so you can fix a rule and reread past data.

How do you capture self-reported attribution?

Add an open text field, “How did you hear about us?”, to the demo, contact, and trial forms. Self-report is the only source that records an answer the buyer read but never clicked.

  • Use free text, not a dropdown. A dropdown limits the answer to what you expected. Buyers write “ChatGPT recommended you” or “Perplexity listed three vendors” when they are given room.
  • Make it optional, or keep it short. A required field gets junk answers.
  • Code the answers weekly. Use a small fixed set: AI assistant (with the engine if named), search, peer or community, review site, event, outbound, other. Keep the raw text beside the code.
  • Store it on the contact, then copy it to the opportunity. A field that lives only in the form tool never reaches a pipeline report.
  • Ask sales to confirm on the first call. Add one discovery question: “What did you read or ask before you contacted us?” Log the answer in the same field.

Self-report has known biases. Buyers remember the last thing they did, not everything they did, and some will not answer. Report it as a share of responses, with the response rate beside it. Amsive’s 2025 study of LLM conversion recommends this same open-ended question to add context that last-touch analytics cannot.

How should the CRM record AI influence?

The CRM should hold the first-touch source, the self-reported source, and the analytics channel side by side, and it should not merge them. Each answers a different question.

CRM fieldComes fromWhat it tells you
Original sourceAnalytics or marketing automation, first sessionWhere the first tracked visit came from
Latest source before conversionAnalytics, last sessionWhat the buyer clicked before the form
Self-reported sourceForm field, sales discoveryWhat the buyer says influenced them
AI touch (yes or no)Any AI channel session, or self-report naming an assistantWhether AI appears anywhere in the record
First landing pageAnalyticsWhich page the AI answer or search sent them to

Capture UTM and referrer values in hidden form fields at the first visit, and keep them in a first-party cookie or local storage, so they survive across sessions to the form fill. Without this, a buyer who first came from Perplexity and converted three weeks later from a direct visit gets recorded as direct.

Is branded search a proxy for AI visibility?

Branded search growth is a useful proxy for AI exposure, and only a proxy. When buyers meet your name in an answer and then search for it, branded impressions rise. Many other things raise them too: events, ads, press, outbound, and seasonality.

Search Console has made the split easier. Google added a branded queries filter to the Performance report in November 2025, and expanded it to all eligible sites on March 11, 2026. Google says the classification is “determined by an internal, AI-assisted system” and may misidentify some queries. Only top-level properties with enough volume qualify. For other properties, filter queries with a regex on your brand names and product names.

Read branded impressions next to your AI visibility panel. A branded lift that follows a measured rise in share of answer, with no campaign in the same window, is stronger evidence than either number alone. It is still an association. Write down every other change in the window.

What lookback window fits B2B?

Use the longest lookback window GA4 offers, and pull pipeline from the CRM, not from GA4. B2B buying cycles often run longer than GA4 can see.

Google’s attribution settings allow 7 or 30 days for acquisition key events, and 30, 60, or 90 days for all other key events, with 90 as the default. Google notes the chosen window also applies to session attribution. Set other key events to 90 days.

Ninety days is still short for a six-month enterprise deal. Join analytics sessions to CRM contacts through the hidden form fields above, and measure AI touches across the full opportunity history in the CRM. Report pipeline by the quarter the opportunity was created, and look back at least two quarters before you judge a change.

Does AI search traffic convert better?

The evidence is contested. Published studies range from parity to several times the organic rate, and each comes from a vendor or agency with an interest in the result.

StudySampleFindingLimits
Amsive, Sept 202554 sites, six months of GA4 dataLLM mean conversion 4.87% vs organic 4.60%. Paired t-test p = 0.794, not significant. B2B LLM 2.17% vs organic 1.16% in the full set; not significant in the thresholded subset (p = 0.705)LLM traffic averaged 0.24% of sessions. Last-touch attribution. Macro conversions only
Siege Media, June 202678 sites, January to May 2026, five industriesMedian AI-to-organic conversion ratio 1.26x. B2B SaaS 1.05x. AI matched or beat traditional traffic on 72% of sitesRatios above 3x grouped. GA4 key event rate. Category-level results only
Microsoft Clarity, Nov 20251,277 mostly publisher and news domains, one monthLLM sign-up rate 1.66% vs search 0.15%. Headline “3x the rate of other channels”Publisher sites, not B2B. One month. Conversion is Clarity “smart events”

Siege Media’s own title asks “Are they really 7x better?” and its answer, in its data, is no. The defensible reading is that AI referral traffic converts at least as well as organic on most sites, from a small base. For B2B, the difference from organic is small and not statistically secure. For more on reconciling numbers like these, see AI search statistics, read by method.

Two cautions apply to any conversion rate you compute yourself. The AI click is often late in the journey, after the buyer has already chosen a shortlist, so the rate looks high. And a few dozen sessions a month give a rate with a wide margin. The share of answer guide shows how to put an interval on a proportion.

How do you report AI search to leadership?

Report a funnel with three stages, exposure, visits, and pipeline, and mark each number as observed or inferred. Leadership will want one number. Give them three, and say which ones are counted and which are estimated.

StageMetricSourceObserved or inferredThis quarterLast quarter
ExposureShare of answer, commercial queries, with intervalAI visibility panelObserved (sampled)
ExposureCopilot and Bing AI citationsBing Webmaster Tools AI PerformanceObserved (first-party, citations only)
ExposureGenerative AI impressionsSearch ConsoleObserved (first-party, impressions only)
ExposureBranded search impressionsSearch Console branded filterObserved. Its link to AI is inferred
VisitsAI assistant sessions, by engineGA4 custom channelObserved (undercount)
VisitsKey events from AI sessionsGA4Observed (undercount)
PipelineLeads self-reporting an AI assistantForm field and CRMObserved (self-report, state response rate)
PipelineOpportunities with any AI touch, and their valueCRMObserved
PipelineShare of direct and branded growth attributed to AIAnalysisInferred. State the method

Keep the observed rows above the inferred rows, and never add them together. Put the concurrent changes underneath: launches, campaigns, model updates, tracking gaps. The platform keeps these numbers in one record with the panel that feeds them.

Worked example (illustrative)

The numbers below are invented to show the method. They do not describe a client.

A B2B software company sells to IT operations teams, with a 90-to-180-day sales cycle. In one quarter:

  • The panel shows its share of answer on 120 commercial queries rose from 14% to 22% across ChatGPT and Perplexity. The intervals do not overlap.
  • The GA4 custom AI channel recorded 610 sessions and 19 demo requests. The default AI Assistant channel recorded 470 sessions. The gap was Perplexity and Claude.
  • 212 demo requests came in. 141 answered the free-text field. 23 of those named an AI assistant, which is 16% of responses.
  • Of those 23, only 9 had any AI session in their analytics history. The other 14 arrived as direct or branded search.
  • The CRM showed 31 opportunities with an AI touch, from either source, worth a stated pipeline value.
  • Branded impressions rose 18%. A trade show fell in the same quarter.

The report to leadership states the observed numbers as counted. It says self-report found more than twice as many AI-influenced leads as analytics did. It does not claim the 18% branded lift came from AI, because the trade show sits in the same window. It sets the next quarter’s question: does the branded lift hold after the event effect fades?

Common mistakes

  • Reading the AI channel as the total. It counts tagged and referred clicks only. Report it as a floor.
  • Placing the AI rule below Referral. GA4 assigns the first match. Move the AI channel up.
  • Using a loose regex. .*ai matches mail providers. Anchor exact hostnames.
  • A dropdown instead of free text. It records your guesses, not the buyer’s words.
  • Leaving self-report in the form tool. Map it to the contact and the opportunity, or it never reaches pipeline.
  • Forecasting with a conversion multiplier. Published multiples disagree. Use your own rate, with its interval.
  • Summing observed and inferred numbers. Keep them in separate rows and label them.
  • Counting AI Overviews as AI assistant traffic. GA4 puts them in Organic Search. Use Search Console for them.

For the measurement method that produces the exposure rows, see measuring AI search visibility. For how the three surfaces fit together, see SEO, AEO, and GEO, defined and compared.

Sources

Bring us the domain.

We will tell you where you are absent, which page is in the way, and what has to be true before a citation is possible.

Request an audit