Skip to Main Content
Kraftwerk Design Inc.

GEO numbers don't add up

Transmissions Johnny Hanshew

Wineries are being sold the same pitch: get "AI-visible" before your competitors do, or disappear from the next generation of search. GEO for wineries — Generative Engine Optimization, the practice of getting a business recommended inside AI answers instead of just ranked in search results — is being pitched like it's urgent.

But there's an obvious question to ask first: is anyone actually using AI to find your wine?

That's a measurement problem. It deserves an answer before an ongoing retainer.

What the data actually says

Growth in AI use is real. But the numbers used to describe it don't necessarily measure the audience or the behaviour a winery is paying to reach.

Conductor's 2026 AEO/GEO Benchmarks Report puts measured AI referral traffic at 1.08% of website traffic across ten industries. That benchmark uses U.S. monthly traffic data from May–September 2025, across 1,215 enterprise customer domains. It's about one percent in that sample, not thirty. Not fifteen.

Measured AI referral traffic

1.08%

Share of website traffic in Conductor’s enterprise benchmark

Conductor, 2026 AEO/GEO Benchmarks Report. U.S. monthly data, May–September 2025; 1,215 enterprise customer domains across ten industries. Not a winery benchmark.

That is an enterprise benchmark, not a winery forecast. It doesn't tell us whether a five-thousand-case operation should expect more or less. And it doesn't separately capture Google's AI Overviews or AI Mode, or visits influenced by an AI answer that arrive through another channel.

The Reuters Institute's Digital News Report 2026 measures something different: averaged across 48 surveyed markets, 10% of respondents said they used a standalone AI chatbot for news in the previous week, compared with 7% in 2025. News use rose overall, while several Western markets reported no year-on-year increase. This is an online, self-reported survey of news use — not general AI adoption, and not wine discovery.

GEO isn't fake. But neither dataset tells us how often people use AI to choose wine or book a tasting. Treating either as the reason for a winery's budget skips the question that matters: what is happening with this winery's customers?

There's also a mechanical reason the referral figure can't be the whole picture. Someone can hear a brand recommended in an AI conversation, later search that brand's name on Google, and click an unpaid result. That visit may arrive as organic search rather than a chatbot referral.

That's an attribution limit, not a measured effect anyone has quantified for this winery. It doesn't turn one percent into the curve being sold. It means direct referrals and total influence are different things, and neither should be confused with the other.

One recommendation. Two recorded sources.

AI answerDirect clickAI referral
AI answerLater unpaid Google search clickOrganic search

The part worth checking before you buy

Before a winery spends real money finding out whether ChatGPT can recommend it, somebody should check whether ChatGPT — or Perplexity, or Google's AI Overviews — can retrieve the product and booking information in the first place. Content quality isn't the only issue. The way the site is assembled matters too.

Three failure modes are worth checking when a website, a booking tool, and a commerce platform get wired together as separate pieces rather than built as one system:

  • Script-dependent club information. A club tier only appears after a script runs, so a system that doesn't render that script may miss it.
  • Disconnected booking information. A tasting reservation lives inside a third-party booking tool, without the same details available on the winery's own page.
  • Prices trapped in images. A price sits inside a photograph instead of typed-out page text, so retrieval that doesn't interpret the image may miss it.

Capabilities differ. Google can render JavaScript; not all bots can. Important content that depends on user interaction, image interpretation, or a separate tool can still be harder to retrieve reliably. Make essential details available as accessible text, then test the actual pages and systems rather than assuming they're unreadable.

Visible to a visitor ≠ available to a machine

Club tier

Visitor view

Revealed by a script

Possible retrieval gap

May be absent without rendering

Tasting reservation

Visitor view

Available in a booking widget

Possible retrieval gap

May sit outside the retrieved page

Tasting price

Visitor view

Visible in a photograph

Possible retrieval gap

Not available as page text

An AI system trying to answer "which winery near me has a Saturday tasting under fifty dollars" needs access to those details to make a useful recommendation. It might find them elsewhere, but missing accessible information on the winery's own site makes an accurate, current answer harder to support. Better writing alone doesn't fix that.

None of that is necessarily negligence. It's a risk of how winery sites get assembled — a website platform, plus a reservation tool, plus a separate club and commerce platform, each owning a piece of the customer experience. A seam that's easy for a human visitor to cross may be harder for a machine to follow.

That's the check that belongs before the GEO retainer, not after it. Verify the foundation before paying for more visibility on top of it.

What to actually ask before signing anything

If someone's pitching a GEO engagement, two questions come first.

1. What adoption data is this budget sized against?

A dated, named source with a sample size — not a slide from a conference, not "everyone's moving this direction." Ask what the source actually measures: website visits, chatbot use, citations, or purchases. Ask which markets and customers it covers. A global headline isn't a substitute for evidence about the audience a winery sells to.

2. Can an AI system read the site's product and booking information?

If nobody has checked, that's the work that needs to happen first. Ask for a scoped assessment and a price before committing to an ongoing GEO retainer.

What actually shows up when you check

Three checks you can start without hiring anyone.

Referral traffic

Analytics can show identifiable arrivals from AI platforms. Inspect the actual sources and source/medium data, not just one default channel: analytics may group chatbot arrivals separately, while Google's AI Overviews and AI Mode are included in organic search. Check whether those visits lead to bookings, purchases, or enquiries, as well as how many arrive.

The enterprise average in a sales deck isn't a forecast for a five-thousand-case DTC winery. Start with your own traffic and outcomes, and be clear about what your tracking can't see.

Branded search

Google's Search Console — the free tool that reports search impressions and clicks — gives another view across a rolling twelve months. Look at queries containing the winery's name, with the understanding that some queries are omitted for privacy and other reporting limits.

An AI recommendation followed by an unpaid Google search click may appear as organic search, not a chatbot arrival. But Search Console can't tell you why someone searched. A rising or flat branded-query trend alone neither proves nor rules out AI influence. Treat it as context alongside campaigns, publicity, seasonality, and what customers tell you — not as attribution on its own.

The platforms themselves

Ask them the way a customer would: the same question, phrased several different ways, over several days. Record the questions, dates, location, and account or session conditions, then check whether the recommendations and booking details are accurate.

These systems don't always return a fixed answer. One clean-sounding result in a vendor demo isn't representative evidence, and inconsistency alone doesn't prove the work has no value. Repeated checks give you a better basis for deciding what needs fixing.

Before you buy GEO

  1. Check AI referral traffic.
  2. Review branded-search trends.
  3. Repeat customer-like questions across platforms.
  4. Confirm product and booking information is accessible.

If those checks don't turn up meaningful exposure or outcomes, what's being pitched may be a bet on where the market is headed rather than an investment supported by your current measurements. There can be a case for that bet. Know which conversation is actually happening before signing.

Sources