You attribute SaaS pipeline to ChatGPT and AI search by asking buyers where they heard of you. Then you store the answer in dedicated CRM fields and read it beside your referrer data. A GA4 channel for AI traffic can't do the job alone.
Much AI influence never arrives as a click. Engines name vendors in their answers, and the buyer turns up later as a branded search, a direct visit, or a demo request.
Key Takeaways
- Referrer data sees only a small share of AI influence, because most buyers arrive later through search or a typed URL.
- Self-reported attribution asks the buyer. In one published company comparison, a survey credited AI search with roughly 9% of conversions, against roughly 1% in GA4.
- Ask differently by path: one optional picklist question in self-serve signup, one required free-text field on the demo form.
- Store the answer in dedicated CRM fields, copy it to the deal, and never overwrite your system's original source.
- Report AI-sourced pipeline beside referrer data and share of voice, not instead of them.
Why Doesn't Referrer Data Show the Pipeline ChatGPT Creates?
Referrer data misses much of the pipeline ChatGPT creates, because most AI influence never arrives as an AI referral. A buyer reads an answer, then searches your name or types your URL days later. Visits from AI apps often carry no referrer at all, so GA4 files them as Direct.
When a click does come from an AI engine, GA4 can sort it into its own channel. We cover that setup in our guide on how to track AI referral traffic in GA4. A channel only counts clicks, though, and it can't see the buyer who never clicked.
Picture one SaaS buying path. A buyer asks ChatGPT for SOC 2 compliance tools, and the answer names a handful of vendors. The buyer notes two names, closes the tab, and searches one of them on Google later. Your analytics record a branded search visit. ChatGPT, which did the introducing, gets no credit.
That branded visit looks like proof your brand is strong, and it may be. But it also hides where the demand started. Our guide to whether AI Overviews are cutting into your SaaS pipeline explains this branded-search trap in more detail. Brand demand and AI demand end up in the same bucket.
Engines Name You Without Linking
There's a second reason the click never comes. AI engines often name a vendor without linking to its site. Our pillar guide to GEO for B2B SaaS covers our audit of 25 software-buying prompts across ChatGPT, Perplexity, and Google AI Overviews. In that audit, 58% to 80% of named vendors were recommended without their own site cited, depending on the engine. A name with no link sends no referral, however persuasive the answer was. The buyer still has to find you some other way.
So what does this mean for measurement? A report built only on referrers will undercount AI's part in your pipeline. We'd also expect the undercount to be largest for buyers who researched the most before they contacted you. That makes the gap skewed, and a skewed gap is harder to correct than a random one.
A 2026 Similarweb panel study puts numbers on this. It found that 55.9% of AI-influenced traffic arrived via search. PPC Land's write-up of the report adds that the AI channel itself accounted for only 8.8% of AI-influenced visits. The panel covered US desktop users and consumer brands in finance, travel, and beauty, not SaaS. Treat it as a sign of the pattern, not a SaaS benchmark.
How Should You Ask Buyers Where They Found You?
Self-reported attribution means the buyer tells you how they found you, in a form field or on a call. Ask for it on every conversion path, but not the same way on each. A demo request and a self-serve signup reach different buyers at different moments, so they need different questions.
On Demo Request Forms, Ask an Open Question
On a demo request form, add a required open text field: "How did you hear about us?" Buyers asking for a demo are motivated, so a short typed answer is a small ask. Free text catches detail a picklist can't, such as the engine and sometimes the prompt: "asked ChatGPT for SOC 2 tools." Keep any picklist off this form, or place it second. A list of options invites the first plausible click.
Don't add more required fields. One open question is easier to answer well than several dropdowns, and your operations team can sort the text into clean categories each week.
On Self-Serve Signup, Ask One Optional Question
In a self-serve signup, every extra field can cost you activation. So ask one optional single-select question after signup, inside onboarding, where skipping it costs the user nothing. Separate the options by engine:
- ChatGPT
- Perplexity
- Google AI Overview or AI Mode
- Gemini
- Claude
- Microsoft Copilot
- Google search
- A review site such as G2 or Capterra
- A colleague or friend
- Social media
- Podcast or event
- Other, with a text box
Keep each engine as its own option, never one "AI" bucket. Each engine needs its own playbook, so a single AI total can't tell you where to work next.
On Both Paths, Capture the Referrer Too
On both paths, capture the technical trail as well. Add hidden fields that store UTM parameters, the landing page, and the referrer. That record won't show the answer that named you, but it shows the route the buyer took afterward. The GA4 guide linked above lists the AI hostnames worth watching.
Self-reports have one real weakness: memory. Buyers misremember, and "Google" can mean they searched your name after ChatGPT suggested you. That's why the free-text answer matters on the demo form. It gives the buyer room to tell the whole story, and it gives you something to check against the referrer.
What Does Self-Reported Data Catch That GA4 Misses?
Self-reported data catches influence that never arrives as a click: the answer that named you, or the shortlist a buyer built in a chatbot. It won't give you session paths or landing pages. Referrer data does, so you need both views.
Think about what each tool is built to see. GA4 records sessions: where a visit came from, which page it landed on, and what happened next. It has no field for "a chatbot recommended you last week." A form answer can hold exactly that. It's cruder, but it measures the thing you're trying to credit.
The two views also check each other. If self-reported AI answers rise while AI referral sessions stay flat, buyers may be finding you through answers that don't link. If both rise together, the engines may be linking to you as well as naming you. So use a simple rule: the buyer's answer tells you the source, and the referrer tells you the path. The CRM section below applies it to the cases where they disagree.
Graphite's analysis of n8n shows how far apart the two views can sit. According to GA4, answer engine optimization, or AEO, accounted for roughly 1% of n8n's conversions. n8n's survey data put the figure at roughly 9%. Organic and paid search showed similar shares in both GA4 and the survey, which acts as a check on the method.
That's one company, the workflow automation platform n8n, and the published analysis doesn't give the sample size or the survey wording. Graphite also sells AEO services, so it has a stake in the answer. Read the gap as a direction, then measure your own. One company's ratio isn't a multiplier to apply to your GA4 numbers.
Which CRM Fields Should Hold AI Attribution?
Create dedicated fields on the contact record, lock the first value each one receives, and copy them to the deal or opportunity when it's created. Leave HubSpot's Original Source and Salesforce's Lead Source alone. Those stay your software's view of the path, and your new fields hold the buyer's view.
Why separate fields? If you overwrite Original Source with a buyer's answer, you lose the referrer record you need for comparison. And if a later form fill can replace the first answer, a returning buyer's "Google" can erase the "ChatGPT" they gave on the demo form. We'd set it up like this:
| Field | Type | Filled by | Rule |
|---|---|---|---|
| Self-reported source (raw) | Long text | The buyer, from the form or a call | Never edited |
| Self-reported source (normalized) | Picklist, same options as the signup question | Ops or a workflow | Map free text weekly |
| AI engine named | Picklist: ChatGPT, Perplexity, Google AI Overview or AI Mode, Gemini, Claude, Copilot, none | Ops | Only when an AI option applies |
| First-touch UTM source and referrer | Text, hidden fields | The form | First value locked |
| Deal AI source | Picklist | Workflow copy at deal creation | Not overwritten later |
The raw field is the record of what the buyer said, so nobody edits it. Ops maps each free-text answer to the normalized picklist once a week, which keeps reports clean without losing the original wording. Copying the value to the deal matters because pipeline reports run on deals, and contact fields can change after a deal opens.
Give one person in revenue operations ownership of the weekly mapping. If nobody owns it, the normalized field fills with blanks and "Other," and the report stops meaning much.
When the Buyer and the Referrer Disagree
Sooner or later, the buyer's answer and the referrer will disagree. These are the rules we'd apply:
- Buyer names an AI engine, referrer says Direct or branded search. Count it as AI-sourced. The buyer's answer sets the source, and the referrer shows the path they took afterward.
- Referrer shows an AI hostname, buyer names something else. Keep both values and count it as AI-assisted. Flag it for review.
- Both agree. Count it as AI-sourced, with high confidence.
These are working rules, not findings. Adjust them once you've seen a quarter of your own data, and write down whatever you change so the numbers stay comparable over time.
How Do Sales Calls Catch the AI Influence Forms Miss?
Sales calls catch AI influence when the rep asks about it on the first call. Many buyers arrive with a shortlist already built, and some skip the form question or answer it loosely. The rep asks where the shortlist came from, then logs the answer to the same raw field the form uses.
The form asks how someone heard about you. On a call, ask how they built their list instead. We'd use two questions: "Which tools or assistants did you use to build your shortlist?" and "Which other vendors are you looking at?" The first invites a buyer to mention ChatGPT or Perplexity without feeling tested. The second tells you which rivals share the shortlist with you.
Why not repeat the form question? By the first call, the buyer has already answered "How did you hear about us?" once, and asking again can feel like admin. Asking about the shortlist moves the question from marketing credit to the buyer's own research, which is closer to what they actually did. A buyer who answered "Google" on the form may name ChatGPT here.
Make logging part of the call notes, not a separate task. The rep adds the buyer's words to the raw source field, next to any form answer and without changing it. Ops then normalizes both together. If you record calls, tagging them for engine names is an optional extra check on the notes.
6sense's 2025 B2B Buyer Experience Report surveyed nearly 4,000 buyers across North America, APAC, and EMEA. It found the point of first contact shifted from 69% to 61% of the journey. That's still well past halfway. Buyers also fill an average of 3.4 spots on their shortlist on Day One. If an AI answer helped build that list, it did its work before you spoke.
Build a Monthly AI Pipeline Report
Build one monthly report with a view for each conversion path. Show AI-sourced pipeline created and AI-sourced closed-won revenue for each path, split by engine. Put AI referral sessions and AI share of voice per engine beside them, so you can see visibility, clicks, and pipeline on one page.
Each input answers a different question. Share of voice tells you how often engines name you. The method is in our guide on how to measure AI share of voice for a SaaS product. Referral sessions come from the GA4 channel. Pipeline comes from the deal fields above. When all three move together, you have a story you can defend.
Read the trends over a quarter, not a month. Monthly counts are often small, so a few deals can swing the figure a long way. Watch the share of answers marked "Other" or left blank too. If it climbs, check the form first, then the weekly mapping.
Share the report with sales as well as marketing. Reps log part of the data on calls, and seeing it used gives them a reason to keep logging.
Don't benchmark your figures against another company's published ratio. Your buyers, paths, and form wording are different. Our review of the pipeline evidence for GEO in B2B SaaS covers what published results show so far.
Frequently Asked Questions
What is self-reported attribution?
Self-reported attribution means asking buyers how they found you and recording their answer. You ask in a form field or on a sales call. It captures influence that analytics tools can't see, such as a ChatGPT answer that named you without linking to your site.
Why does ChatGPT traffic show up as Direct?
Visits from AI apps often carry no referrer, so GA4 files them as Direct. Many buyers also never click at all. They read an answer, then search your name or type your URL later. According to PPC Land's write-up of a Similarweb panel study of consumer brands, the AI channel itself accounted for only 8.8% of AI-influenced visits.
What should a "How did you hear about us?" field include for AI search?
List each AI engine as its own option: ChatGPT, Perplexity, Google AI Overview or AI Mode, Gemini, Claude, and Microsoft Copilot. Add Google search, review sites, a colleague or friend, social media, podcasts or events, and Other. On demo forms, lead with a required free-text field, because typed answers can name the engine and even the prompt.
How do you track leads from ChatGPT in HubSpot or Salesforce?
Create dedicated fields for the buyer's raw answer, a normalized source, and the AI engine named. Lock the first value and copy it to the deal or opportunity at creation. Leave HubSpot's Original Source and Salesforce's Lead Source untouched, because they hold the software's view of the path.
How accurate is self-reported attribution?
It's imperfect, because buyers misremember and may answer "Google" after an AI engine suggested you. It still sees what referrer data can't. According to Graphite's analysis of n8n, GA4 credited AI search with roughly 1% of conversions, while survey data put it at roughly 9%. That's one company, so pair both views rather than trusting either alone.
Conclusion
No single report will attribute SaaS pipeline to ChatGPT and AI search for you. Ask buyers on both paths, with a different question on each. Store the answers in dedicated fields and copy them to the deal. Ask again on the first sales call. Then report AI-sourced pipeline beside referrer data and share of voice.
Agencies such as Kubnal Bridge specialize in connecting AI visibility work to the pipeline fields your sales team already uses. If you'd like a second opinion on your setup, a short review is a sensible place to start.
About the Author
Kubnal Bridge Editorial Team. The Kubnal Bridge editorial team researches and writes about generative engine optimization and AI search visibility across ChatGPT, Perplexity, Google AI Overviews, Gemini, and Claude.
Sources
- Similarweb, "Closing the AI Search Attribution Gap," 2026-06-18, https://www.similarweb.com/corp/reports/the-downstream-impact-of-ai-visibilty/
- PPC Land (Luis Rijo), "Your analytics are lying: Similarweb traces AI recommendations to real traffic," 2026-06-21, https://ppc.land/your-analytics-are-lying-similarweb-traces-ai-recommendations-to-real-traffic/
- Graphite (Ethan Smith), "Last-Touch Attribution Only Captures 10% of n8n's AEO Conversions," 2026-06-01, https://graphite.io/five-percent/n8n-attribution-gap
- 6sense, "2025 B2B Buyer Experience Report," 2025-11, https://6sense.com/report/buyer-experience/