AI Assistants Are Our Top Self-Reported Discovery Channel — Ahead of Google (n=51)

Key finding

31% of inbound hiring leads self-reported discovering MDliaison via an AI assistant (ChatGPT, Claude, Copilot, or "AI search") — ahead of explicit Google (27%) and generic web search (22%), in a sample of 51 consecutive intake-form submissions as of July 2026.

How to cite this data

MDliaison. AI Discovery Channel Study: Self-Reported Lead Attribution in Healthcare B2B Hiring. July 2026. n=51. https://mdliaison.com/ai-discovery-channel-healthcare-2026

What we measured — and what we didn't expect

MDliaison is a marketplace where healthcare companies hire pre-vetted independent medical sales reps. Every company that submits our hiring intake form answers one free-text question:

"How did you hear about us?"

We recently classified every answer from our 51 most recent inbound hiring inquiries. We expected Google to dominate. It didn't.

Three caveats before anyone quotes this:

1. Small sample (n=51). Useful as a directional signal, not a population estimate.
2. Self-reported attribution. Some "web search" answers may actually be AI Overviews or AI-assisted search; some "ChatGPT" answers may be imprecise recall.
3. One vendor, one niche. This reflects healthcare companies hiring contract medical sales talent in the US — not all B2B categories.

We're publishing anyway because almost nobody publishes bottom-of-funnel channel data. The AI-search conversation runs almost entirely on traffic and impression statistics — not on where signed leads actually say they came from.

Verbatim answers (anonymized)

A few examples of how buyers described AI-assisted discovery:

• "ChatGPT"
• "AI Platform"
• "I asked Copilot"
• "AI search"

We kept this field free-text rather than a dropdown menu. Dropdowns would have collapsed these answers into a generic "online" bucket — and we would have missed the shift.

Why this aligns with broader industry data

Our 31% is a leading indicator, not an outlier, when placed next to what the wider market is measuring:

• Gartner projected that traditional search engine volume will drop roughly 25% by 2026 as chatbots and virtual agents absorb queries.
• Semrush's study of AI search's impact on SEO traffic found AI-referred visitors are more likely to convert than traditional organic visitors — they arrive pre-qualified because the assistant already did comparison work.
• Industry trackers put ChatGPT at the large majority of AI referral traffic, with Gemini and Claude growing fastest.

The mechanism matters more than any single stat: when a founder asks ChatGPT "how do I hire a medical device sales rep without paying a recruiter fee," the assistant answers with a short list of named vendors. There is no page two. Either you're in the answer or you don't exist for that query.

What we changed after seeing this data

This wasn't a traffic report — it was a measurement failure. Our Google Search Console dashboards showed 0.10% company-side CTR (6 clicks on 6,046 impressions in a trailing 28-day window), while roughly 80% of inbound leads were search-driven. The AI channel was invisible in every mart we had. Here's what we actually did — not a generic SEO checklist.

1. We made the intake form our AI share-of-voice metric

The finding only existed because our hiring intake form asks "How did you hear about us?" as a free-text field, not a dropdown. We manually classified 51 consecutive submissions into four buckets. A dropdown would have collapsed "ChatGPT," "AI Platform," and "I asked Copilot" into "online" — and we would never have seen the 31% figure.

This form is now our primary AI share-of-voice measurement. We don't rely on AI-referral traffic in Google Analytics (which undercounts assistant-driven discovery) or manual spot-checks of ChatGPT answers. We ask every buyer, in their own words, and classify the responses. That's how we know AI assistants are ahead of explicit Google — and how we'll track whether that share grows or shrinks over time.

Takeaway for other marketers: your intake form may be the only instrument that captures AI-assisted discovery accurately. Fix attribution before you fix content.

2. We stopped treating all search as Google search

Our GSC data had no way to separate demand-side queries (hiring companies evaluating CSOs) from contractor-side queries (reps researching salaries and jobs). We built a user_type classifier — a Gemini Flash prompt running against every keyword in our GSC mart — that labels each query company-side, contractor-side, both, or none. Promoted to production June 2026 after an A/B test on our top 100 keywords.

That changed prioritization: we now segment SEO reporting by who is searching, not just what they're searching for. Company-side content gets weighted toward intake-form conversion; contractor-side impressions no longer inflate how "well" we're doing with buyers.

3. We started publishing the numbers we already had — in citable formats

Generic "we're a marketplace" copy doesn't get quoted by assistants. Specific, sourced numbers do. So we began a research brief series designed for citation:

Each brief follows the same template: headline stat block, data table, explicit methodology, "how to cite" line, and ungated PDF. The goal is to become the source assistants name when someone asks a question we can answer with real data.

4. We consolidated 40+ pages into seven AI-citable pillars

We audited 40+ overlapping buyer pages that were cannibalizing each other — five URLs competing for "medical sales outsourcing," three for "CSO comparison," seven for physician liaison hiring. We merged them into seven canonical pillars, each structured for AI extraction: question-style H1, direct answer in the first 100 words, at least one comparison table with specific fees and timelines. All 301 redirects are live.

Example: our CSO comparison page publishes what most CSOs won't — a per-rep cost framework ($133K loaded cost → $185K–$240K client price), named competitors with contract minimums, and a rep-replacement-rate benchmark (15–22% industry average). That's the template every surviving pillar now follows.

5. We added structured data to commercial pages

Every canonical pillar now carries FAQ and Article schema — so machines can parse comparison tables, fee ranges, and definitional content that humans skim. Research pages (including this one) add Dataset schema on top, so the stat blocks are explicitly machine-readable.

6. What's still in progress

• Lead-source in the data warehouse. Attribution still lives in Typeform/HubSpot, not our BigQuery marts. Wiring original source into the DWH is the next step to track AI vs. Google over time without manual classification each quarter.

We'd rather publish a small honest dataset and name the one remaining gap than claim we've "solved AI SEO."

What this means for B2B healthcare marketers

The playbook that wins AI citations is mostly the playbook that was always good marketing: be the source of specific, verifiable, generously explained answers in your niche.

The difference is the payoff curve. In classic SEO, position #4 still earned traffic. In an AI answer, being the cited source is closer to winner-take-all.

Practical next step: If your intake form doesn't ask how buyers found you — in their own words — start there. You may be blind to a channel that already drives a double-digit share of leads.

Methodology

Questions about the data or methodology: contact us.