AI search can introduce a prospect to your business. The difficult part is proving what happened next
No single report shows the complete journey from an AI citation to a qualified lead. A reliable attribution setup connects platform visibility, website analytics, form activity, CRM outcomes and one short self-reported attribution question.
OpenAI makes part of that journey unusually straightforward. Links from ChatGPT search automatically include `utm_source=chatgpt.com`, according to its publisher documentation. Google reports visibility and clicks from AI-powered search experiences inside Search Console. Neither system, however, can tell you whether the resulting visitor completed a form, attended a sales call or generated revenue.
That connection has to be built after the click.
Quick answer: Track AI-generated demand at three levels. Use Search Console and AI visibility data to measure exposure, GA4 to identify attributable sessions, and form plus CRM data to connect those sessions with business outcomes. Add self-reported attribution because some AI-assisted journeys leave no reliable referral signal.
Can you track leads from ChatGPT?
Yes. When someone clicks a cited link from ChatGPT search, OpenAI adds `utm_source=chatgpt.com` to the destination URL. Analytics software can use that parameter to identify the session, while a form can retain it and attach it to the eventual submission.
The simplest attributable journey looks like this:
1. A prospect asks ChatGPT for a product or service recommendation.
2. ChatGPT cites or recommends a relevant page.
3. The prospect follows the link containing `utm_source=chatgpt.com`.
4. GA4 records the session source.
5. A hidden field stores the same source with the form submission.
6. The submission enters a CRM with its acquisition data intact.
This gives marketing and sales teams a traceable path from AI referral to lead. It does not capture every person influenced by ChatGPT. Someone may read a recommendation, close the conversation and later type the company name into Google or visit the website directly. Analytics will see the later visit, not the earlier influence.
That distinction matters. Referral attribution measures identifiable traffic. It does not measure every recommendation that helped create demand.
Does Google Analytics identify AI traffic automatically?
GA4 can identify visits when the source reaches the website through a referrer or UTM parameter. Its traffic-source dimensions include source, medium and campaign, which can be inspected at user, session and event scope.
Start by looking for sources such as:
– `chatgpt.com`
– `perplexity.ai`
– `copilot.microsoft.com`
– `gemini.google.com`
– Other AI assistants that appear in your referral data
Do not assume that every AI-influenced visit will appear under one of those names. Google defines `(direct) / (none)` as traffic for which Analytics has no clear referral information. Missing parameters, redirects, privacy controls and later direct visits can all break the chain.
This creates two broad categories:
| Category | What you can observe | What remains uncertain |
|, |, |, |
| Direct AI referral | Source, landing page, session behaviour and form events | Whether AI was the only influence |
| AI-assisted visit | A later branded, organic or direct session | Which assistant influenced the visitor |
The first category can be measured through analytics. The second requires supporting evidence from the respondent or CRM.
Can Search Console show traffic from AI Overviews and AI Mode?
Google Search Console reports how a site performs inside Google’s generative search features. Its Generative AI performance report includes AI Overviews and AI Mode, with data that can be grouped by page, country, device and date.
Google also counts external link clicks from AI Overviews and AI Mode as Search clicks. This makes Search Console useful for measuring whether pages are appearing and receiving engagement inside Google’s AI-powered results.
Search Console and GA4 answer different questions:
| Tool | Primary question |
|, |, |
| Search Console | Did Google show or link to our content in generative search? |
| GA4 | What did identifiable visitors do after reaching the website? |
| Form analytics | Did they start, abandon or submit the form? |
| CRM | Did the submission become a qualified opportunity or customer? |
Treating one of these tools as the complete attribution system creates predictable blind spots. Search Console measures search performance, not lead quality. Form analytics measures the conversion experience, not the entire discovery process. CRM data measures commercial progression, but only after acquisition details reach the record.
The useful picture appears when those datasets share a common identifier or source value.
How do you preserve the traffic source in a lead form?
Use hidden fields to store acquisition parameters with the submission. A hidden field is invisible to the respondent but records a value supplied in the form URL, such as `utm_source`, `utm_medium`, `utm_campaign`, `ref` or a custom channel label.
In Rowform, hidden fields can capture URL parameters automatically. If a form URL contains:
“`
https://app.rowform.io/f/example?utm_source=chatgpt.com&utm_medium=referral
“`
the corresponding hidden fields can preserve `chatgpt.com` and `referral` alongside the respondent’s answers.
Rowform includes hidden-field values in its Results view, CSV exports, Google Sheets integration, notifications and webhook payloads. This allows the acquisition source to move with the lead instead of remaining isolated inside an analytics report.
A practical field set is:
| Field | Example value | Purpose |
|, |, |, |
| `utm_source` | `chatgpt.com` | Identifies the referring platform |
| `utm_medium` | `referral` | Classifies the traffic channel |
| `utm_campaign` | `ai_search` | Groups AI acquisition reporting |
| `landing_page` | `/pricing` | Records the first relevant page |
| `original_source` | `chatgpt.com` | Preserves first-touch acquisition |
| `latest_source` | `google` | Records the most recent known touch |
Only parameters already present in the URL can be collected directly by a hidden field. If the form sits on a different page or domain from the original landing page, the website must preserve those values as the visitor moves between pages.
That can be handled through first-party storage, a tag-management setup or application code. The exact implementation depends on how the website and form are embedded.
Which form events should you send to GA4?
Track enough events to distinguish traffic quality from form friction. A submission event alone tells you how many people finished. It cannot explain whether AI visitors failed to begin, abandoned a particular question or completed the form but produced low-quality leads.
Rowform can send five events to the Google Tag Manager data layer:
– `rowform_form_view`
– `rowform_form_start`
– `rowform_question_view`
– `rowform_question_answer`
– `rowform_form_submit`
These events can be routed to GA4 through GTM. Each event includes form metadata, while question events identify the question type and position. Answer values are not sent to the data layer.
This supports a useful acquisition funnel:
“`
AI-attributed session
→ landing-page engagement
→ form view
→ form start
→ question progression
→ form submission
→ qualified lead
→ opportunity
→ customer
“`
Once the events are available in GA4, compare AI-referred sessions against other channels using the same definitions. Useful measures include:
– Landing-page-to-form-view rate
– Form-view-to-start rate
– Form completion rate
– Median completion time
– Question-level drop-off
– Qualified-lead rate
– Meeting-booked rate
– Opportunity rate
– Customer conversion rate
Avoid judging the channel on submission volume alone. A small source may generate fewer forms but a higher proportion of relevant prospects. Another source may produce high completion and poor commercial fit.
What should you ask on a form to measure AI influence?
Ask one short self-reported attribution question: “How did you first hear about us?” Include options for ChatGPT or another AI assistant, Google, social media, a recommendation and an open-ended alternative.
For example:
– Google search
– ChatGPT or another AI assistant
– LinkedIn or another social network
– Recommended by a person
– Podcast, newsletter or community
– Other
The wording should focus on first discovery rather than the most recent click. “How did you get here today?” often produces a last-touch answer. “How did you first hear about us?” is more likely to reveal the channel that introduced the brand.
An optional follow-up can ask which assistant or source the respondent used. Conditional logic keeps that question hidden from everyone else.
Self-reported attribution is not a replacement for analytics. Respondents forget, simplify or credit the touchpoint they remember most clearly. It still reveals influence that click-based tracking cannot observe.
Use the two sources together:
| Analytics says | Respondent says | Likely interpretation |
|, |, |, |
| `chatgpt.com` | ChatGPT | Direct AI referral with corroboration |
| Direct | ChatGPT | AI-assisted visit without retained click data |
| Google organic | ChatGPT | AI created awareness; Google completed navigation |
| `chatgpt.com` | Google | Multiple touches or imperfect recall |
| Direct | Unknown | Unattributed demand |
Disagreement is useful. It shows where a single-touch model would have assigned too much certainty.
How do you connect a form submission to the CRM?
Send the form response, hidden attribution fields and submission metadata into the same lead or contact record. The record should preserve the original source even when later sessions or campaigns create new touchpoints.
Rowform webhooks send submission data to an external service in real time. The payload can include the form ID, answers, hidden fields, timestamp and response metadata. A webhook can pass that information to a CRM, a data warehouse or an automation platform.
The minimum useful CRM record contains:
– Original source
– Latest known source
– Landing page
– Self-reported source
– Form name
– Submission date
– Lead status
– Qualification result
– Opportunity value or revenue, where applicable
The source field should not be overwritten every time the person returns. Store first-touch and latest-touch values separately. Otherwise, a prospect discovered through ChatGPT and later returning through Google will become an organic-search lead by default.
For B2B companies, attribution should continue beyond lead creation. A channel that generates ten submissions and no qualified opportunities has performed differently from one that generates three submissions and two sales conversations.
How should you measure AI search lead quality?
Compare AI-referred and AI-influenced leads against the same downstream milestones used for other channels. Completion rate describes the form. Qualification and revenue describe the acquisition source.
A compact reporting model can use five stages:
| Stage | Recommended measure |
|, |, |
| Visibility | Generative-search impressions, citations or mentions |
| Visit | Sessions and engaged sessions from identifiable AI sources |
| Form | Views, starts, submissions, completion and drop-off |
| Lead | MQL rate, accepted-lead rate and meeting rate |
| Revenue | Opportunities, pipeline value and customers |
This separation prevents a common attribution error: treating visibility as traffic or traffic as revenue.
The Stack House applies this distinction when analysing AI search visibility for B2B companies. Its approach connects model mentions and citations with Search Console, website analytics and conversion data, rather than treating a visibility score as proof of commercial impact.
That framework is especially useful when direct AI referrals remain small. A company may see growing generative-search exposure before referral sessions or qualified leads become large enough to evaluate reliably. Each stage provides evidence, but they are not interchangeable.
What is the best AI lead attribution setup for a small team?
A small team does not need a custom attribution platform. It needs consistent naming, source preservation and one reporting destination.
A practical setup is:
1. Search Console for Google AI visibility and search clicks.
2. GA4 for attributable sessions and on-site behaviour.
3. Google Tag Manager for Rowform interaction events.
4. Rowform hidden fields for URL parameters attached to submissions.
5. One self-reported question for otherwise invisible influence.
6. A webhook or native integration to move responses into the CRM.
7. Separate first-touch and latest-touch fields so later visits do not erase discovery.
Start with `rowform_form_view`, `rowform_form_start` and `rowform_form_submit` before adding question-level instrumentation. Add complexity only when it answers a decision the team is prepared to make.
For example, question-level drop-off matters when you can rewrite or reorder the question. Multi-touch modelling matters when the business has enough qualified volume to compare channels. A twelve-field attribution system with fifteen monthly leads will mostly produce tidy columns.
What are the most common AI attribution mistakes?
The most common mistakes come from expecting one tool or one touchpoint to explain the entire buying process.
Treating direct traffic as genuine direct discovery
Direct traffic means the source is unavailable. It does not prove that the visitor independently remembered and typed the URL. AI-assisted journeys, shared links and untagged documents can all finish in the same bucket.
Measuring submissions without lead quality
Forms are conversion points, but a completed form is not automatically a commercially useful outcome. Qualification, meeting and opportunity data need to return to the acquisition report.
Overwriting the original source
The latest session is easier to observe and often receives the credit. Preserve the first known source before updating any subsequent-touch field.
Asking an attribution question too early
An attribution question is useful, but it should not be the first demand placed on the respondent. Establish the form’s purpose, collect the necessary information and place attribution near the end unless routing depends on it.
Creating an “AI traffic” channel without documenting its rules
List the hostnames, UTM values and classification rules included in the channel. AI products change domains and referral behaviour. An undocumented regex becomes unreliable reporting surprisingly quickly.
Combining visibility and conversion into one score
Being cited, receiving a visit and generating a customer are related events with different denominators. Report them as a funnel rather than compressing them into a proprietary percentage nobody can audit.
Frequently asked questions
Does ChatGPT add UTM parameters to links?
OpenAI states that ChatGPT search automatically adds `utm_source=chatgpt.com` to referral URLs. The parameter can be read by analytics software and preserved in a form hidden field when it remains in the destination URL.
Why does an AI-generated lead appear as direct traffic?
The user may have copied the company name, returned later, used another device or arrived through a link that did not preserve referral information. GA4 assigns `(direct) / (none)` when it has no clear source, so direct traffic should be treated as unknown rather than literally direct.
Can forms identify whether ChatGPT influenced a lead?
A form can preserve a direct referral source and ask the respondent how they first discovered the company. The combination identifies both observable clicks and some otherwise invisible AI influence, although self-reported answers remain imperfect.
Should AI traffic use its own GA4 channel group?
A custom channel group can make recurring analysis easier once naming rules are documented. Keep the original source and medium available so the combined channel can still be audited by platform.
Is a higher AI referral conversion rate proof that AI traffic is better?
No. Small samples, branded queries, landing-page choice and audience intent can distort the comparison. Evaluate qualification and revenue over a meaningful period before changing acquisition priorities.
Conclusion
AI search attribution works best as a chain of evidence. Platform reports show visibility, analytics records identifiable visits, form data captures conversion and the CRM reveals commercial value.
For ChatGPT referrals, preserve `utm_source=chatgpt.com` in both analytics and the form submission. For journeys without a reliable click signal, add a short self-reported attribution question. Then keep the original source attached to the lead as it moves through qualification, sales and revenue.
The objective is not to assign perfect credit to every interaction. It is to distinguish measurable AI referrals, probable AI influence and genuinely unknown demand without pretending they are the same thing.
Rowform gives you the practical form layer for this workflow: hidden fields to preserve acquisition data, GTM events to measure interaction, conditional logic for self-reported attribution and webhooks to move each submission into your CRM.
Ready to connect your forms to the full customer journey? Create your first Rowform