This is a field note from my own journey of building a next-gen AI marketing lab. Whether you’re an AI enthusiast or a business owner trying to figure out what actually matters, this is written for you.
User experience for buyers has quietly changed from “search, scan, click” to “ask, read, decide.” More and more people use LLMs and AI-powered search — in ChatGPT, on Google, in Perplexity — to get answers instead of links. If you run a business, this is not a marketing curiosity. It’s a change in how your next customer finds you, and it deserves to be treated as its own channel.
Here’s the thesis up front: AI search sends you fewer visitors than Google ever did — and the visitors it sends are better. Better as in further along, more decided, more likely to buy. That trade-off changes how you should think about being found online, and it’s especially good news if your business is niche.
I’ve spent the past year and a half watching this shift from the inside — experimenting with AI search as a lead generation channel, tracking who gets cited and why, and watching leads arrive in the pipeline having already made up their minds before anyone spoke to them. This post isn’t a write-up of those experiments. It’s the thinking behind them: how we got here, what changed in buyer behavior, and what I believe you can and should do about it.
When did this start?
The honest answer is: earlier than most people think, and in a different order than most retellings suggest.
| When | What happened | Why it mattered |
|---|---|---|
| Dec 2022 | Perplexity launches its public answer engine | First to make “answers with citations” feel normal |
| May 2023 | Google announces Search Generative Experience (SGE) | AI summaries on top of results, Labs waitlist only |
| May 2024 | SGE becomes AI Overviews, live for all US users | No opt-in; 100+ countries by Oct 2024 |
| Oct 31, 2024 | ChatGPT Search launches | Real retrieval, real citations — the inflection point |
| Dec 2024 – Feb 2025 | Rollout to all logged-in users, then everyone | AI search stops being a paid-user feature |
| Mar – Aug 2025 | Google AI Mode: Labs → all US → 180+ countries | Google’s own search becomes conversational |
Two things in this table deserve a closer look.
First, Google moved before OpenAI on the search side — AI Overviews beat ChatGPT Search to market by five months. The popular retelling has it backwards.
Second, why October 31, 2024 is the date that matters: before ChatGPT Search, ChatGPT’s “citations” were often hallucinated. It would confidently invent URLs and attribute papers to real researchers who never wrote them, because it was pattern-matching from training data, not retrieving anything. That launch is when ChatGPT’s recommendations became checkable — and when SEO professionals started treating it as a channel rather than a toy.
So the sequence is: Perplexity proved the format, Google put AI summaries in front of billions, ChatGPT made citations real and conversation the default, and Google responded by making its own search conversational. By 2026, this isn’t an emerging behavior. McKinsey’s research found roughly half of consumers — across every demographic, including boomers — now use AI search for purchase decisions.
Why should you, as a business owner, care about the buyer journey?
Because the journey determines where your marketing money actually works.
For twenty years, the model was stable: a buyer typed keywords into Google, got ten blue links, clicked several, compared, and formed a shortlist themselves. Your job was to be on page one so you could get into that comparison set. The buyer arrived at your site as a skeptical evaluator — early, undecided, and doing the filtering work themselves.
AI search collapses that filtering step. The buyer describes their problem in a sentence or a paragraph, and the AI does the comparing — then hands back a synthesized answer with three to five names in it. The buyer who then clicks through to your site isn’t evaluating a list of ten anymore. They’re confirming a recommendation they’ve already half-accepted.
This is not a subtle difference, and the 2026 data makes it concrete:
| What buyers are doing now | The number | Source |
|---|---|---|
| B2B buyers starting research in an AI chatbot more often than Google | 51% | G2, Apr 2026 |
| B2B buyers who chose a different vendor than planned, based on AI guidance | 69% | G2, Apr 2026 |
| B2B buyers who bought from a vendor they’d never heard of before AI surfaced it | ~33% | G2, Apr 2026 |
| Consumers starting searches with AI instead of Google | 37% | Eight Oh Two, late 2025 |
| Consumers who’ve used AI to help make a purchase decision | 47% | Eight Oh Two, late 2025 |
| AI-referred shoppers’ conversion lift vs. non-AI traffic (e-commerce) | +42% | Adobe Analytics, Apr 2026 |
The row to read twice if you’re a smaller company: a third of B2B buyers purchased from a name that wasn’t on their radar until an AI put it there. Under the old model, an unknown brand had to out-rank incumbents for years to get seen. Under this one, the AI’s shortlist is the discovery mechanism — and it doesn’t care how big your ad budget is.
And the pattern isn’t B2B-specific. Whether you sell software or skincare, AI does the pre-work — comparing, filtering, shortlisting — and the human arrives closer to a decision. Adobe’s data on AI-referred shoppers shows them spending more time on product pages and generating more revenue per visit, which is exactly what you’d expect from someone confirming a recommendation rather than starting research from scratch.
What this looks like in your analytics — and why “fewer clicks” is the wrong thing to panic about
Here’s where most businesses get the story wrong, because the top-line numbers look scary in isolation.
AI referral traffic is still small — roughly 1% of total website traffic per Conductor’s cross-industry benchmark — and AI Overviews are eating clicks on the Google side: Pew Research found users click a source cited inside an AI summary in only ~1% of visits, and click any traditional result about half as often as when no summary appears.
So: less traffic from Google, and only a trickle from AI tools. Sounds bad. It isn’t — because the trickle behaves nothing like normal traffic:
| Signal | AI-referred visitors | Organic search visitors | Source |
|---|---|---|---|
| Time per session | ~943 sec | ~849 sec | Microsoft Clarity, 30B sessions, Oct 2025–Mar 2026 |
| Behavior pattern | Fewer pages, deeper focus | More pages, broad browsing | Microsoft Clarity |
| Sign-up conversion (publisher study) | 1.66% | 0.15% | Microsoft Clarity via Pixis |
| Conversion premium across studies | ~1.3x to 10x+ | baseline | Pixis roundup |
Clarity’s framing is the one I’d steal: AI acts as a pre-qualification layer. It filters out the information-seekers before they ever reach you, so the people who do click are disproportionately in-market.
That’s the “fewer clicks, better customers” trade in one picture. And here’s the kicker: most businesses can’t see it happening. Only around 14% of marketers track AI search as a separate channel — the rest let these visits get misfiled as “direct” or generic “referral” traffic in their analytics, which means the channel is invisible in their reporting even while it quietly sends them their best-converting visitors.
The customer you never saw coming (and the one you never saw leave)
Somewhere in the last month, there’s a good chance this happened to you.
Someone with a real problem and a real budget opened ChatGPT and described what they needed — in a full paragraph, the way they’d explain it to a colleague, not in three keywords. They got back a short list of companies. They spent twenty minutes checking those names — a website here, a few reviews there, maybe one Google search to make sure nothing was off. Then they reached out to one of them.
If you were on that list, congratulations: a well-qualified buyer arrived at your door, and in your analytics they probably showed up as “direct traffic” — a visitor from nowhere. You won the deal and never learned why.
If you weren’t on that list, nothing happened at all. No lost-deal notification. No bounced visit. Your name simply never entered the conversation — and that silence is what makes this shift so easy to underestimate.
Now, before this story convinces you to move your whole budget by Friday — three things that same buyer would want you to know.
They’re a trickle, not a flood. AI referrals are still around 1% of web traffic; if your business runs on volume, this channel can’t carry it yet. They’re skeptical, too: trust in AI answers actually fell from 82% to 54% in one year even as usage climbed — which is why that buyer spent twenty minutes verifying, checking 2.4 platforms on average, instead of blindly trusting the shortlist. Your website, reviews, and Google presence still closed the deal. And the shortlist itself isn’t a ranking anyone controls — ask twice, get different names; new model, new list. Anyone selling you “guaranteed position one in ChatGPT” is running 2010’s guaranteed-Google-rankings pitch in a new jacket.
So the honest picture is this: an and channel, not an or channel. Small, invisible in your reporting, immune to guarantees — and quietly deciding which businesses make it into conversations you’ll never witness.
What you can and should do
The mental model first: when an AI names your business, that’s not a ranking — it’s a recommendation, delivered in prose, with reasons attached. The buyer treats it like advice from a knowledgeable friend. Everything below follows from that.
1. Measure it like a channel, because it is one. Right now, AI-referred visitors are likely hiding in your “direct traffic.” Separate them out — GA4 can segment referrals from chatgpt.com, perplexity.ai and friends, and Microsoft Clarity now does it out of the box. You can’t manage what you’ve misfiled — and since barely 14% of marketers track this, simply seeing the channel is a head start.
2. Fix what you already own before buying anything labeled “GEO.” 86% of AI citations come from sources brands control or influence — your website first, then listings and directories, then reviews. So: make your site machine-readable (direct answers up top, real headings, structured data), and keep your listings and review profiles current. These were “nice to haves” in the SEO era. They’re load-bearing now.
3. If you’re niche, lean into it — hard. The original GEO research out of Princeton and Georgia Tech found that smaller, lower-visibility sites gain the most ground in AI answers, because these systems reward specific, expert content over raw domain size — and auditors keep finding big brands that dominate Google yet are invisible in AI answers, while specialists get cited. If you know one thing extraordinarily well, write like it: specific, factual, attributed to real humans. That depth is exactly what content farms can’t fake.
One warning before you go: don’t rob Google to pay ChatGPT. AI builds the shortlist; traditional search still closes the deal. Show up coherently in both — same facts, same proof, everywhere a model or a human might look.
The click economy trained us to want traffic. This channel gives you something better: customers who did their homework elsewhere and arrived ready. Fewer clicks, better customers — for most businesses, that’s the deal you always wanted.