How to Capture AI-Sourced Deals With a Self-Reported Attribution Survey

Your best deal this quarter probably started inside an AI assistant, and your analytics will never tell you. The buyer asked ChatGPT for tools in your category, read a paragraph that named you, opened a tab, and converted a week later as "Direct" traffic. Every dashboard you own credits that pipeline to the wrong source. A self-reported attribution survey, one question asked at the right moment, is the cheapest reliable way to see the demand that AI discovery is already generating for you.

This is not a theoretical gap. It is the single largest blind spot in measuring answer engine optimization, and it is getting wider every quarter as more buyers start their research in a chat window instead of a search bar.

Why AI-sourced pipeline is invisible to your analytics

Traditional attribution reads two things: the referrer header the browser sends, and the UTM parameters on the link. AI assistants break both.

When a buyer clicks a citation inside ChatGPT's mobile app or Perplexity's in-app browser, the outbound link frequently omits the referrer entirely. Industry measurement in 2026 puts the share of AI referral sessions arriving with no referrer at roughly 35 to 70 percent, depending on the engine and device. Those sessions land in Direct, indistinguishable from someone typing your URL by hand.

The larger problem is zero-click. When an engine names your brand in an answer and the buyer reads it without clicking, nothing registers anywhere. Conductor's 2026 benchmarks estimate that around 93 percent of AI search sessions end without a single website click. The recommendation still moves the deal. It just leaves no trace in your funnel.

Google added a native AI Assistant channel to GA4 on May 13, 2026, which recognizes ChatGPT, Gemini, and Claude automatically. It helps, but it is a partial fix, as several independent analyses of the channel have pointed out. Here is what it does and does not catch.

The pattern is clear. Instrumentation catches the sessions that behave like clicks. It cannot catch the conversation that happened before the click, and the conversation is where AI discovery does its work. That is the same reason a clean referrer so rarely survives the journey, a problem worth understanding in depth if you want to prove AEO pipeline when the buyer leaves no referrer.

The one question that catches what tracking cannot

Self-reported attribution asks the buyer to tell you what software cannot see. The mechanic is old, direct-response marketers have used "how did you hear about us" for decades, but it is newly essential because the dark funnel now runs through AI.

The rule that makes it work: use one open-text field, never a drop-down.

Drop-downs anchor answers to the channels you already know about. A buyer who found you through a Claude recommendation will pick "Search" or "Word of mouth" from a list, because those are the closest options, and the AI signal vanishes into a bucket you cannot act on. An open box lets them write "ChatGPT suggested you when I asked for alternatives to [competitor]," which is the exact sentence you need. The unexpected answers are the entire point.

Where to place the survey, and where not to

Placement decides data quality more than wording does. Ask at a low-commitment moment and you get noise. Ask at the point of high intent and you get the truth while it is fresh.

Put the field on:

• Demo request forms
• Free trial or account sign-up
• Pricing or sales-contact forms
• The first question a sales rep asks on a discovery call, logged as a CRM field

Keep it off newsletter signups, gated PDF downloads, and confirmation pages. Those capture people too early or after their attention is gone, and the recall is unreliable. Make the field optional but visible. A required field tanks your form conversion; a buried one gets skipped.

If you run product-led signup with almost no form, add the question as an optional step in onboarding, or ask it in the first in-app welcome message. The point is to catch the answer while the buyer still remembers the specific moment of discovery.

How to word the question so buyers name the AI assistant

Generic wording gets generic answers. "How did you hear about us" often produces "online" or "a friend." You want the buyer to name the surface and, ideally, the moment. Small changes in phrasing pull far more specific responses.

Compare these:

• Weak: "How did you hear about us?" (invites one-word answers)
• Better: "What first led you to check us out?" (invites a story)
• Best for AEO: "Where did you first come across us? If an AI assistant like ChatGPT, Claude, or Perplexity mentioned us, we would love to know which one." (names the surface, gives permission to be specific)

That last version does two things. It signals that AI answers are a valid, expected path, which makes buyers comfortable naming them. And it plants the specific engine names, which improves recall. You are not leading the witness, you are lowering the barrier to a truthful, detailed answer. Keep a plain follow-up too, "Anything else that put us on your radar?", because AI discovery usually sits alongside a Reddit thread, a YouTube review, or a peer recommendation, and you want the full assist chain.

Turning open-text answers into a number your board trusts

Open text is powerful and messy. The work is converting free responses into a defensible percentage. Run this as a monthly loop.

1. Export the month's responses alongside deal value and stage from your CRM.
2. Build your channel tags from the responses themselves, not from a preset list. Let categories like "AI assistant (ChatGPT)," "AI assistant (Perplexity)," "AI assistant (unspecified)," "Reddit," and "peer referral" emerge from what people actually wrote.
3. Tag each response. Allow multiple tags per answer, because "I saw you on a Reddit thread, then ChatGPT confirmed you were the top pick" is two real touches.
4. Weight by pipeline, not just count. Ten low-intent trials that name ChatGPT matter less than three enterprise demos that do. Report AI-influenced pipeline dollars, not only lead counts.
5. Compare the self-reported column against your software-attributed column. The delta between them is your dark funnel gap, and for AI it is usually large.

After three to six months you have a trend line, not a snapshot, and a trend is what survives board scrutiny. Feed the tags into your CRM so this becomes a standing report rather than a spreadsheet you rebuild each month. If you want the exact field structure and deal-stage records to make that automatic, we walk through how to instrument your CRM to report AI-sourced pipeline elsewhere on the blog.

Self-reported versus software attribution for AI deals

Neither method is complete alone. They cover each other's blind spots, and the honest reporting move is to show both.

Read them together. Software gives you precise counts of the traffic it can see. The survey gives you the shape of the demand it cannot. When the survey says 22 percent of new pipeline names an AI assistant and GA4 credits AI with 6 percent of sessions, that 16-point gap is the value of your AEO work that your analytics was silently handing to Direct. This is the same reconciliation logic behind a full model for attributing pipeline and ROI from AI-driven discovery, and the survey is the input that model needs most.

How to keep self-reported data honest

Self-reported attribution has one real weakness: memory. Buyers misremember, compress a five-touch journey into one, and credit the last thing they recall. Three habits keep it trustworthy.

Triangulate. When the survey says AI is driving pipeline, look for corroborating signals: rising Direct traffic with strong engagement, more branded search, growing AI referral sessions in GA4. If self-reported AI mentions climb while those move too, the story holds.

Ask early. Recall decays fast. A buyer at demo-request remembers the discovery moment far better than the same buyer at renewal.

Watch the trend, not the number. Any single month's percentage is noisy. The direction over a quarter is the signal. If you have also improved how AI engines see you, and you can confirm they are ingesting your content through a structured feed, self-reported AI mentions should rise in step. Making your content that machine-readable is exactly what the AI Feed Engine is built to do, and pairing feed changes with survey trends turns correlation into something closer to proof. A quick first step you can take today is to publish a clean llms.txt file so crawlers can find your best pages, which you can build in a minute with our free llms.txt generator.

Common mistakes that ruin the data

• Using a drop-down. It hides every channel you did not predict, which is exactly where AI discovery lives.
• Asking on low-intent forms. Newsletter signups produce guesses, not memories.
• Making the field required. It costs you conversions and breeds junk answers from people trying to get past it.
• Reporting counts instead of pipeline. One enterprise deal that names ChatGPT outweighs fifty low-fit trials.
• Quitting after a month. Patterns need a quarter to stabilize before you reallocate budget on them.
• Treating it as the whole answer. It is one instrument. Run it beside GA4's AI channel and CRM tracking, never instead of them.

The teams that win here treat self-reported attribution as a permanent instrument, not a one-off survey. It is nearly free, it takes one form field and an hour of tagging a month, and it is the only method that reliably sees the zero-click, no-referrer demand that AI assistants now generate. When you can show a board a clean trend of AI-influenced pipeline, the case for investing in answer engine optimization stops being a leap of faith and becomes a line on a chart. That is the same story we helped surface in the FastTrackr AI case study, where AI-sourced discovery showed up in the pipeline well before it showed up in any analytics tool. Once the survey proves the demand is real, the next question is how much to invest to grow it, and our pricing page lays out where a lean team can start.

Frequently asked questions