The 9x conversion stat is how vendors are selling you AI search. Look at the denominator.

TL;DR: The stat used to sell AI search programs is real and misleading. At honest sample sizes AI traffic converts about 1.3x better than organic, not 9x, on under 2 percent of traffic. AI search did not add a channel to capture. It removed the capture click. The move is to fund demand creation, not stand up a GEO program.

Every pitch I have seen for an AI search program in the last year leans on the same number. AI traffic converts several times better than organic. One widely shared case study put ChatGPT referrals at a 15.9 percent conversion rate against 1.76 percent for Google organic, roughly nine times better (Seer Interactive). The slide writes itself. New channel, enormous multiple, move now before your competitors do.

The multiple is real. The conclusion drawn from it is wrong, and the reason is sitting in the same study, one line down.

That nine times figure came from a single client where AI traffic was 0.07 percent of organic. Just under 11,000 sessions against almost 14 million, over seven months. So the honest way to state the finding is this: on a rounding error of traffic, the visitors who did arrive converted very well. That is a much less exciting sentence, and it is the true one.

The denominator is the whole story

I have spent twelve years being handed conversion rates and asked to make decisions from them. The first thing an operator does with a rate is ask what it is a rate of. A 15.9 percent conversion rate on 11,000 sessions is about 1,750 conversions. A 1.76 percent rate on 14 million sessions is roughly 246,000. The channel everyone wants me to chase produced under one percent of the outcomes of the channel they are telling me is dying.

A nine times conversion rate on 0.07 percent of traffic is not a strategy. It is a rounding error with good manners.

It gets worse for the pitch when you widen the sample. The nine times multiple comes from one client. Look across 94 ecommerce brands and ChatGPT traffic converts at 1.81 percent against 1.39 percent for non-branded organic (Search Engine Land). That is a 31 percent edge, not a 900 percent one. The multiple did not just shrink. It collapsed by an order of magnitude the moment the sample stopped being a single hand-picked account.

Both things are true at once, and holding them together is the actual insight. AI referral traffic does convert better, because the model pre-qualifies the visitor before they ever click. By the time someone arrives from ChatGPT, they have already had the comparison conversation, narrowed the field, and decided you are worth a look. You are not earning the conversion. You are receiving a visitor the model already sold. That is selection bias, and selection bias does not scale just because you build a program around it.

What actually happened is worse than a new channel

The story the vendors tell is additive. A valuable new channel appeared, go capture it. The truth is subtractive, and it is the part nobody is selling because there is no service to attach to it.

AI search did not add a channel. It ate the bottom of the funnel where capture used to live.

The numbers on this are not subtle. When Google shows an AI summary, people click a traditional search result in about 8 percent of visits, against 15 percent when there is no summary, and they click a link inside the summary itself just 1 percent of the time (Pew Research Center). Ahrefs, looking across 300,000 keywords, found that the presence of an AI Overview is associated with a 58 percent lower click-through rate for the top-ranking page (Ahrefs). For every hundred clicks that page used to earn, Google now keeps 58.

For fifteen years, performance marketing has been a capture discipline. Someone develops intent, they search, you intercept them at the moment of intent with a well-ranked page or a bid, and you measure the click that follows. The entire apparatus, last-click attribution, branded search defense, the conversion-rate obsession, assumes a click exists to be captured at the moment of intent. The AI answer removes that click. It resolves the intent before the user ever reaches you.

So the question is not how do I get cited in the answer. The question is what makes a model name me in the first place, and the answer to that is uncomfortable for a performance marketer, because it is not a performance-marketing activity.

The thing AEO cannot manufacture

A model recommends the brands it has learned the category associates with. It builds that association from what it has read across the open web: analyst coverage, press, forum threads, comparison content other people wrote about you, the body of earned material that says you are a serious answer to this category. You cannot schema-markup your way into that. You cannot bid on it. It is the accumulated residue of demand creation, the brand and category work that performance marketing has spent fifteen years defunding because it does not produce a clean last-click number.

This is the part I want to be precise about, because it is the actionable core. The most durable way to win AI search is to be the brand a buyer already wants to ask the model about by name. When someone types “is Exotel any good for CPaaS in India” instead of “best CPaaS India,” the model’s answer barely matters, because the buyer has already chosen and is looking for permission. That branded, named query is the output of demand creation, not capture. It is the one input AEO genuinely rewards and the one no GEO vendor can sell you, because it is slow, it is unattributable in a last-click model, and it looks like the budget line everyone has been trained to cut first.

What I would actually do

Four moves, in the order I would make them.

Report the absolute number, not the rate. Put the denominator next to every AI dashboard. A conversion rate with no volume beside it is a vanity metric, and the nine times figure is the most seductive vanity metric in marketing right now. My rule of thumb: a channel does not earn dedicated headcount or budget until it clears a few percent of real pipeline. Until then you instrument it, you watch the trendline, and you do not staff it.

Reallocate from capture, not from brand. The instinct will be to fund a GEO program out of the brand budget. That is exactly backwards. With capture click-through down 58 percent where AI Overviews appear, the marginal dollar in capture-side SEO is depreciating in real time. The money should move toward the demand creation that makes a human name you to the model, not away from it.

Fix measurement before you touch budget. Last-click is now structurally broken, because the AI answer is a real and influential touch that leaves no trackable referrer. Most AI-influenced buyers show up later as direct or branded-search traffic, not as an “AI” source in analytics. So I lean harder on self-reported attribution, the “what made you reach out today” question, and I watch branded search volume and direct traffic as leading indicators. If you wait for an “AI referral” line in GA to grow, you will both overstate it when it is small and miss the influence that shows up everywhere else.

Do not hire a Head of GEO. Spinning up a dedicated function ratifies the framing that this is a new specialist channel, which is the framing the vendors need you to accept. Even Google’s own position is that there is no special AI optimisation, just the same fundamentals of crawlable, helpful content (Google Search Central). The technical hygiene that helps a model read you belongs to SEO as it always has. The work that actually moves citations, owning a category point of view, earning analyst and press mentions, being the named brand, belongs to brand and product marketing. De-silo it.

The uncomfortable version

Here is the part I would put in the board deck and watch the room go quiet. The correct response to AI search is to move money back into the demand creation that performance marketing spent fifteen years cutting, and to do it on the strength of a channel that is currently under two percent of web traffic and converts, at honest sample sizes, about 1.3 times better than what you already have.

That is a hard sell, and it is the right one. The operators who win the next few years will not be the ones who bought a GEO audit. They will be the ones who understood early that AI did not hand them a new place to capture demand. It quietly removed the place they were capturing it, and rewarded the brands that had done the patient, unfashionable work of being worth recommending.

Answers are the new page. But you do not win the answer by optimising for it. You win it by being the brand the buyer was already going to name.