How Do You Attribute a SaaS Deal That Started in an AI Assistant?

How do you attribute a SaaS deal that started in an AI assistant?

Combine three sources: the referrer and tracking data your analytics captured, a required "how did you hear about us" field on your demo form, and the buyer's own words from the first sales call. No single source is reliable for AI assistant traffic, but together they give you an answer you can defend in a pipeline review.

Many B2B buyers now start research by asking ChatGPT, Claude, Gemini, or Perplexity which tools fit their problem. They get a short list, read a few pages, and show up on your demo form days later. In your CRM, that deal often looks like direct traffic or organic search. The assistant that put you on the short list gets no credit.

That matters for SaaS marketing teams, because budget follows attribution. If AI assistant influence stays invisible, the content that earns those mentions looks like it does nothing, and it gets cut. This is the method I use to make that influence visible without pretending the data is cleaner than it is.

Why is AI assistant attribution so hard?

It is hard because the path is broken into pieces. A buyer asks an assistant, reads an answer, maybe clicks a link, maybe searches your brand name later, and maybe tells a colleague who books the demo. Most analytics tools see only the final visit, and that visit often carries no sign of the assistant at all.

Some visits from AI assistants arrive with a referrer that names the assistant, and some arrive with none, depending on the app, the device, and how the link was opened. A visit with no referrer usually lands in your direct channel. A buyer who reads an answer and then types your brand into Google shows up as branded organic search.

There is also the zero-click problem. Many buyers never click at all. They read the assistant's summary of your product, decide you are worth a call, and go straight to your site later. That influence is real, and no tracking parameter will ever capture it.

What is the difference between AI-sourced and AI-influenced deals?

An AI-sourced deal is one where the first recorded touch came directly from an AI assistant, through a tracked click or a clear referrer. An AI-influenced deal is one where the buyer says an assistant shaped their shortlist, even if the tracked first touch was something else. You need both labels, kept separate.

Keeping them apart protects your credibility. Sourced numbers come from systems, so finance can audit them. Influenced numbers come from buyers, so they are softer but often larger. Mixing them into one number invites someone to challenge the whole thing the first time a single deal looks wrong.

In practice I report sourced pipeline as the hard number and influenced pipeline as context beside it. The gap between them is itself useful. A big gap tells you assistants matter more than your tracking can show, which is a reason to invest in being cited, not a reason to doubt the data.

How do you capture AI referral traffic in analytics?

Create a dedicated channel group or segment for AI assistant referrers in your analytics tool, so those visits stop hiding inside generic referral traffic. Match on the referring domains of the assistants your buyers use, and review the list every quarter, because new assistants and new domains appear regularly.

The exact setup depends on your stack. In Google Analytics 4, teams often use a custom channel group for this, so check Google's own docs for the current steps. In other tools, it is a saved segment or a filter. Some assistants also add a tracking parameter to the links they show, so check each vendor's publisher documentation for current behavior instead of assuming.

The second half is getting that first-touch data into your CRM. Capture the original source and landing page on the first visit, store it in hidden fields on your forms, and write it to the contact record when they convert. I covered the CRM side of this in tagging every deal with its first campaign in HubSpot, and the same pattern works for AI referrers.

Should you add a self-reported attribution field?

Yes. A required open text field asking "How did you hear about us?" is the single best tool for AI assistant attribution. Buyers will often write "ChatGPT recommended you" or "Claude listed you as an option," which no tracking system would ever show. Make it free text, not a dropdown, so you capture their real words.

A dropdown forces buyers into your categories. If the options are Google, LinkedIn, referral, and other, the AI answer gets buried in other. Free text lets them name the assistant, the question they asked, and sometimes the competitor they compared you with. That detail is worth more than a clean picklist.

The tradeoff is cleanup. Someone has to read the answers and tag them, at least monthly. For a small SaaS team with a few dozen demo requests a month, that is an hour of work, and a small classification step in your automation can do a first pass for you.

How should sales capture AI influence on the first call?

Add one question to the discovery call script: "What did you look at before reaching out to us?" Then ask a follow up if they mention an assistant: "What did it say about us?" Log the answer in a dedicated CRM field. That second question is the one that tells marketing what to fix.

Buyers are often happy to explain how they built their shortlist. They might say an assistant described you as expensive, or listed you as best for small teams, or confused you with a competitor. Each of those is a direct message about how AI engines understand your positioning.

Make the field short and structured enough to report on, such as a yes or no for AI influence plus a notes field. If it lives only in call recordings, nobody will ever count it. The way these leads should be handled once they arrive is a separate question, which I covered in qualifying a lead that came from an AI assistant.

Which attribution model works best for AI-influenced deals?

Use a simple first touch and last touch view for system data, plus the self-reported field as a separate dimension. Complex multi-touch models assume you can see every touch, and with AI assistants you cannot. A model that pretends to be precise with missing data is worse than a simple one that admits its gaps.

I like reporting three numbers side by side. Pipeline where the first tracked touch was an AI referrer. Pipeline where the buyer self-reported AI influence. And pipeline where both are true. The overlap shows how much your tracking catches, and the self-reported number shows how much it misses.

Review those numbers by quarter, not by week. AI-influenced pipeline moves slowly and in lumps, because one good citation can drive several deals over months. Weekly reporting makes the line look random and tempts people to cut the work too early.

How do you connect attribution back to content?

Ask which pages the AI assistant likely drew from. When buyers quote what an assistant said about you, compare it with your site. If the quote matches your pricing page, comparison page, or a specific article, that page is earning influence. If the quote is wrong, a page is missing or unclear.

This is where attribution turns into a content plan. Pages that keep showing up in buyer quotes deserve updates and internal links. Topics where assistants describe you wrongly deserve a clear page that states the fact plainly, near the top, in words a model can quote.

For the analytics side of this on a Webflow build, the setup I use is described in AI search referral traffic attribution on Webflow. The principle is the same on any platform: track what you can, ask about the rest, and connect both back to specific pages.

What mistakes make AI attribution misleading?

The biggest mistakes are counting every direct visit as AI influence, treating one enthusiastic quote as a trend, and changing definitions mid-quarter. Each makes the number look bigger or more certain than it is. Once leadership catches one inflated number, they stop trusting the rest of the report.

Direct traffic has many causes, including bookmarks, email clients, and typed URLs. Attributing a share of it to AI assistants is a guess, so label it as one or leave it out. Self-reported answers can also be vague. "Saw you somewhere online" is not AI influence, and tagging it that way pads the count.

Write your definitions down before the quarter starts and keep them fixed. If you need to change them, start a new series and keep the old one for comparison. Honest, boring attribution beats exciting attribution that falls apart under one question.

What should you do next?

This week, add a required free text "How did you hear about us?" field to your demo form and one AI question to your discovery script. Set up an AI referrer channel in your analytics. After one quarter, report AI-sourced and AI-influenced pipeline side by side, with your definitions written above the numbers.

My view is that AI assistants are already a real part of the B2B SaaS buying journey, and the teams that measure it honestly will be the ones that keep investing in the content that earns citations. The teams that wait for perfect tracking will wait a long time.

If you want help wiring up AI attribution across your forms, analytics, and CRM, reach out through pravinkumar.co. Let's chat about what your buyers are actually hearing about you.