“How did you hear about us?”: the most valuable question you’re not asking

Analytics is very good at counting clicks. But more and more customers find a business without clicking anything: they ask ChatGPT or Gemini, get a name, and then search for it or pick up the phone. No tracking script can see that. The customer can tell you in two seconds, if you ask.

Why it matters more now

For years, self-reported source (“how did you hear about us?”) was seen as the soft, unreliable cousin of analytics. People forget, pick the first option, or say “Google” for everything. All true. But AI assistants changed the balance. A growing share of discovery happens inside a chat window that passes nothing to your website. Self-reported source is now the only signal for a big slice of AI-driven customers. Paired with tracking, it’s far more reliable than either alone.

What the data shows so far

Two public studies from 2026 put numbers on the gap:

• n8n (software, research by Graphite, June 2026): GA4 last-touch credited AI search with roughly 1% of conversions. A one-question “how did you hear about us?” survey put it at roughly 9%. Organic and paid search came out similar in both, which is why the AI gap is hard to wave away. Read the study.
• An independent hotel (Nicolas Sitter, September 2026): 35 website visitors said an AI assistant sent them. 24 of those 35 carried no trace of any assistant in the tracking data. Read the study.

Different industries, same pattern: most customers who credit an AI assistant arrive without a click analytics can recognise. Neither study measured phone enquiries, where service businesses get most of these answers.

How to ask it on your form

• Label it exactly: “How did you first hear about us?” The word “first” matters. Otherwise people answer with the last thing they did, which is usually Google.
• Use a dropdown for the answer you count: Google, ChatGPT, Gemini, Other AI, Facebook/Instagram, Referral, Other. Naming the AI tools explicitly makes people recognise them. “Online search” lumps ChatGPT in with Google.
• Add one optional free-text box for the detail: “What did you search or ask?” Some practitioners prefer free text alone, because a list can put words in people’s mouths. A dropdown plus an optional box gives you both: a number you can report and the actual prompt, which tells you what AI assistants recommend you for.
• Keep it optional but visible. Put it right above the submit button, not buried among the address fields.
• Don’t add more than one source question. The optional detail box counts as part of it. One good question beats three that people skip.

“How did you hear about us?” options: a list you can copy

Use these as the dropdown answers. Keep the list short (7 to 9 options), put the most common sources first, and always name the AI assistants on their own line so people recognise them.

• Any business: Google search · Google Maps · ChatGPT · Gemini · Other AI assistant (Perplexity, Copilot, Claude) · Facebook or Instagram · Friend or family · Other
• Home services (movers, plumbers, HVAC, roofers): Google search · Google Maps or Business Profile · ChatGPT or another AI assistant · Yelp or Angi · Truck, yard sign or flyer · Repeat customer · Friend or neighbour · Other
• B2B and software: Google search · ChatGPT, Gemini or Perplexity · LinkedIn · Podcast, newsletter or article · Review site (G2, Capterra) · Colleague or referral · Event · Other
• Online store: Google search · ChatGPT or another AI assistant · Instagram · TikTok · YouTube · Friend or family · Ad · Other

Two rules make the answers usable. Keep “Google search” and “ChatGPT” as separate options, because one combined “online search” option hides exactly the customers you want to count. And when someone picks “Other”, show the optional “What did you search or ask?” box, so you get their words instead of a blank.

How to ask it on the phone

For service businesses, this is the part that decides whether the data is any good.

• Ask on every new enquiry, right after the name: “And how did you first hear about us?”
• If they say “Google”, follow up once: “Did anything point you to us before you searched, like a friend or ChatGPT?” It’s surprising how often the answer is yes.
• Log the answer somewhere structured, in the CRM field or a click in your attribution tool, not a notes box.
• Log that you asked, even when the answer is “don’t remember”.

The number that tells you if it’s working: asked rate

If you only record answers, “nobody said ChatGPT” and “nobody was asked” look identical. Track the share of leads where the question was asked. Aim for 70%+ before you trust the AI numbers. Under that, the result says more about your intake process than about AI.

Turning answers into attribution

Answers on their own are a tally. They become attribution when each one is attached to a lead, and each lead to a sale:

• A customer who clicked from ChatGPT and says ChatGPT is as solid as it gets.
• A customer with no tracked AI visit who says ChatGPT is a Declared AI customer: real, in their own words, and worth counting separately.
• A customer who came through Google but told you on the call that ChatGPT shortlisted you is Assisted, if someone writes that down.

That’s exactly how Which AI Sent labels leads. See Confirmed vs Declared. It adds the question to your forms for you, lets reps log phone answers in a click, and shows the asked rate on every report.

Related: AI lead tracking for home service businesses · AI lead attribution.