Is AI Search Only a Top-of-Funnel Channel?

Most marketing leaders have already made peace with AI search. They concede the top of the funnel and keep the bottom.

The line I hear in almost every audit kickoff sounds like this: "Sure, people ask ChatGPT for ideas. But when it is time to buy, they go back to Google and our site."

That hedge held up for about a year. It does not hold up against the numbers anymore.

Similarweb's 2026 Generative AI Brand Visibility Index measured AI versus traditional search at every stage of the purchase funnel. At discovery, AI beats search by a wide margin. That part everyone expected. The part nobody priced in: AI now edges traditional search even at the final purchase decision.

AI owns the top of the funnel. The new fact is that it now competes at the close.

If you have been treating AI visibility as an awareness play, your budget is sized for a 2024 reality. This post walks through what the data shows, why the top-of-funnel myth survived as long as it did, and how to fix a program that stops investing right where the buyer is deciding.

For the conversion side of this story, we already covered why AI search traffic converts 4.4x better than SEO and why AI referrals behave like decision-stage visits.

What the Similarweb funnel data actually shows

Similarweb tracked 113 brands across six sectors and ran more than 11,000 prompts in Finance alone, across ChatGPT, Gemini, Copilot, and Perplexity. The finding is blunt: AI tools out-draw traditional search at all three funnel stages, and the gap only narrows at the very bottom. AI has taken the top of the funnel and is now contesting the close.

At discovery, AI outdraws Google 2.6 to 1

At product discovery, 35% of US consumers now use AI tools versus 13.6% who use traditional search. That is a 2.6x advantage at the moment a buyer first forms a shortlist.

This is the number most teams already believe. It maps to the behavior everyone has watched friends and colleagues adopt: you open ChatGPT to learn a category before you ever type a query into Google.

At evaluation, AI still runs 2x ahead of search

At the evaluation and comparison stage, AI holds 32.9% against search's 15%. The advantage barely shrinks from discovery.

That matters because evaluation is where most B2B and considered-purchase decisions are actually won or lost. If your brand is absent from the model's comparison, you are not in the room when the buyer narrows the field.

At the final purchase, AI edges search for the first time

Here is the data point that breaks the hedge. At the final purchase decision, AI sits at 24.3% against traditional search at 22.1%.

The gap is small. The direction is the story. The stage everyone assumed belonged to Google and the brand's own site is now a stage where AI is marginally ahead. Similarweb's own framing: AI took the top of the funnel, and it is now closing the sale too.

The funnel did not get shorter. It moved inside the model.

The visitor who does arrive is worth more

The funnel-share data pairs with an engagement gap. AI-referred visitors spend roughly 15 minutes on site versus 8 for Google referrals, view 12 pages per visit versus 9, and convert at 7% on transactional sites versus 5%.

Fewer visitors, higher intent. The buyer arrives having already done the comparison work inside the model, so the click is closer to a decision than a Google click usually is.

Why the "AI is only top of funnel" myth survived this long

The myth was not stupid. It was a reasonable read of bad measurement. Five things kept teams convinced AI was an awareness toy, and each one is a measurement artifact, not a fact about buyer behavior.

Reason #1: Referral volume stayed flat, so dashboards missed the shift

Similarweb notes AI referral volume held roughly flat even as AI platform traffic grew 28.6% year over year. A flat traffic line on a dashboard reads as "nothing is happening here." So teams stopped looking.

The flat line is real. It is also the wrong line to watch. Adoption at the decision stage moved while the click count did not.

Reason #2: A 1% click-through rate hid the real influence

Pew and a 68,879-search academic analysis both land near the same number: only about 1% of users who see an AI summary click a cited source. Two independent sources converging on roughly 1% is a strong signal.

Teams read that 1% as "AI does not drive purchases." Wrong inference. The buyer reads the answer, absorbs the brand impression, and acts later without a trackable click.

A 1% click-through rate is not a 1% influence rate.

Reason #3: Most analytics still bucket AI traffic as "direct"

If your GA4 setup files ChatGPT, Gemini, Claude, and Perplexity referrals under "Direct" or "Other Referral," the AI funnel is invisible by construction. You cannot see a channel you have not defined.

We walk through the channel-grouping fix in the AI search measurement guide. Until that split exists, every conclusion about AI's funnel role is drawn from a blind spot.

Reason #4: B2B teams assumed consumer data did not apply

Similarweb's index is consumer data, so B2B marketers waved it off. That instinct is backwards. The B2B buyer is a consumer who learned these habits on their own time and brought them to work.

The B2B numbers, covered below, point the same direction. The behavior transfers.

Reason #5: The funnel was measured by clicks, not decisions

Classic funnel reporting counts sessions and assigns a stage to each click. AI breaks that model because the decision work happens in a conversation that produces no session at all.

When you measure a click-free decision process with a click-based instrument, the instrument reports zero. That zero got read as "AI is not in the funnel here."

The B2B version of the same shift

Similarweb measured consumers. The B2B data tells a matching story. Half of B2B software buyers now start their search with an AI chatbot instead of Google, and 74% of those buyers use ChatGPT as the starting surface, per G2's reporting on the AEO category's 2,000% growth. The buyer is already inside the model before sales ever hears from them.

The CMO side moved too. A Wynter survey of about 110 CMOs at companies above $50M ARR found 84% now use LLMs for vendor discovery, up from 24% a year earlier, as documented in Profound's AEO playbook. That is a 60-point jump in twelve months.

McKinsey frames the scale of the consumer shift behind all of this. About half of US consumers already use AI-powered search, and McKinsey projects $750 billion in US spending will flow through it by 2028, per its analysis of AI search as the new front door. McKinsey does note that more than 70% of AI-search users ask top-of-funnel questions, but the same research finds the technology used across the entire decision journey, not parked at discovery.

Put the consumer and B2B reads side by side and the gap between dashboard and reality gets obvious.

What your analytics dashboard shows:

• •AI sends roughly 1% of sessions
• •That share looks flat month over month
• •Most of it lands in "Direct" or "Other Referral"

What the buyer is actually doing:

• •Starting 35% of product discovery in AI, against 13.6% in search
• •Comparing options inside the model before any click happens
• •Arriving on your site already shortlisted, then converting at 7% versus 5%

The first column is why the myth survived. The second column is why it is wrong. For the deeper B2B procurement read, see why B2B buyers now start their search in ChatGPT and G2.

How to optimize for the full-funnel AI buyer

The fix is not "spend more on AI." It is to stop building only awareness content and start covering the comparison and decision stages where the buyer is now active. Four moves, in order.

Define an "AI Search" channel before you argue about budget

You cannot reallocate toward a channel you cannot see. Build a custom channel grouping in GA4 that pulls ChatGPT, Gemini, Claude, Perplexity, and AI Overview referrals into one "AI Search" row, then mirror that split in your CRM.

Once the channel exists, the 7% versus 5% conversion gap shows up in your own numbers rather than a vendor's slide. Now the budget conversation has a denominator. The GEO measurement guide covers the setup.

Audit citation share at the comparison stage, where the gap is widest

Run a prompt set that mirrors the real buyer journey: a few discovery prompts, several comparison prompts, and a handful of decision prompts ("best X for Y," "X vs Z pricing," "is X worth it for a team our size"). Check which stages cite you and which do not.

Most brands we audit are present at discovery and absent at comparison. That is the exact stage where the Similarweb gap is widest in your favor if you show up, and most costly if you do not.

Build comparison and pricing pages that a model can quote

If AI is contesting the close, the pages that win the close have to be extractable. Comparison pages, pricing pages, and decision-stage FAQs need clean 40-to-60-word answer blocks that a model can lift verbatim.

Awareness blog posts will not get cited at the purchase stage because they do not answer purchase-stage questions. We documented the page-type playbook in comparison pages and AI citations and pricing pages AI can quote.

Judge AI visibility by influence, not click volume

Drop "how many sessions did AI send" as your headline metric. With a 1% click-through rate, that number will always look small and will always undersell the channel.

Track citation share by funnel stage, conversion rate of the AI Search cohort, and assisted pipeline where your CRM can attribute it. Those metrics measure influence. Session count measures the one behavior AI deliberately removed.

FAQ

Is AI search really used at the bottom of the funnel, or just discovery?

Both. Similarweb's 2026 index shows AI tools at 35% of product discovery versus 13.6% for search, and still ahead at the final purchase, 24.3% versus 22.1%. AI is strongest at discovery, but it now edges traditional search even at the buying decision. The bottom-of-funnel hedge no longer matches the data.

Does the Similarweb consumer data apply to B2B SaaS?

The behavior transfers. G2 reports half of B2B software buyers start their search in an AI chatbot and 74% use ChatGPT, and a Wynter survey found 84% of CMOs use LLMs for vendor discovery, up from 24% a year earlier. The B2B buyer learned these habits as a consumer and brought them to work.

If AI sends so little traffic, why does full-funnel presence matter?

Because the click is not the influence. Roughly 1% of users who see an AI summary click a cited source, per Pew and a 68,879-search analysis, yet the brand impression in the answer still shapes the decision. The visitor who does click converts at 7% versus 5% for Google. Low volume, high intent.

How do I measure AI's influence across the funnel?

Define an "AI Search" channel in GA4 and your CRM, run a staged prompt audit that separates discovery, comparison, and decision prompts, and track citation share by stage plus the AI cohort's conversion rate. Session count alone hides the channel because AI is built to answer without a click.

Will the purchase-stage gap hold as AI search adoption grows?

The 24.3% versus 22.1% margin is the read as of early 2026 and it is narrow. As AI search broadens to more casual top-of-funnel use, the close-stage gap could compress or widen depending on how buying flows evolve. Treat it as a live trend to monitor, not a fixed law.

The bottom line

The comfortable story was that AI handles curiosity and Google handles commerce. Similarweb just retired it.

AI leads at discovery by 2.6x, leads at evaluation by more than 2x, and now edges traditional search at the final purchase. The buyer is using AI across the whole journey, and the visitor who clicks through converts harder than the Google one.

Teams that keep funding AI visibility as an awareness line item will keep showing up for the first question and vanishing by the third. The brands that win the next two years will be cited at the comparison and decision stages, where the money actually moves, and they will measure influence instead of counting the clicks AI was designed not to send.