How will customers find brands when AI answers their questions?
The short answer
Customers will increasingly find brands through what an AI assistant chooses to cite in its answer, rather than through a page of blue links they scan themselves. OpenAI expanded ChatGPT Ads to 31 European markets on August 24, 2026, about six months after the US launch, which shows how fast assistants are becoming a commerce surface, not just a research tool. Nobody in the industry has reliable, standardized attribution for how much of that influence actually drives a sale yet, so treat any specific conversion-lift number for assistant referrals as directional, not settled.
Customers are starting to find brands the way they find a restaurant recommendation from a friend: someone, or something, tells them the answer directly, and they act on it without comparing ten options themselves. That someone is increasingly an AI assistant, and the mechanism that decides whether your brand shows up in that answer is different from the one that decided whether you ranked on a search results page.
What’s actually changing
For two decades, discovery ran through one box: type a query, get ten blue links, click through and compare. That box is splitting into many. A shopper might ask ChatGPT, Claude, or Gemini a question directly and act on whatever the assistant cites, never touching a traditional results page at all. OpenAI expanded ChatGPT Ads to 31 European markets starting August 24, 2026, roughly six months after the US launch, with ads currently running on Free and Go plans and a self-serve Ads Manager announced as coming (per OpenAI’s announcement and Search Engine Land’s coverage). That’s a signal about direction, not proof of scale: it shows a major assistant is building commerce infrastructure into the answer itself, not that it has already replaced search for most categories.
Why early citations might compound the way early links did
The pattern worth naming, carefully, is a structural one. In the early years of search, sites that earned links and citations first tended to keep compounding that advantage, because search engines used existing authority as a signal for future ranking. Something similar could be forming with assistants: a brand or a page that gets cited today becomes part of the training and retrieval signal an assistant draws on tomorrow, which makes it more likely to get cited again. If that compounding effect turns out to be real and durable, being an early, well-structured answer to a category question is worth more than it looks today.
That’s a position, not a settled fact, and it comes with a kill condition: if assistant answers stay dominated by a small number of large, generic sources regardless of how well a smaller brand’s page is structured, the compounding read is wrong and citation share won’t behave like the old link economy at all. Too early to know which way it goes.
The honest counterweight
Nobody has reliable attribution for how much an assistant answer actually influences a purchase. Some industry coverage, including reporting from Yotpo, points to assistant-driven referral traffic converting at a higher rate than traffic from organic search, but the underlying methodology varies by source and the category isn’t standardized the way search attribution eventually became. Any specific lift number you see quoted right now should be read as directional. The honest position is that discovery is visibly fragmenting away from one search box, and measurement hasn’t caught up to explain exactly how much that shift is worth yet.
Old search box vs. assistant answers
| Traditional search | Assistant answer | |
|---|---|---|
| What the customer sees | A ranked page of links to compare | One synthesized answer, often with a citation |
| What earns a spot | Backlinks, on-page SEO, domain authority | Structured, answer-shaped content an assistant can cite cleanly |
| Attribution today | Mature: rankings, click-through, conversion tracking | Immature: no standard way to measure assistant-driven influence yet |
| Compounding effect | Well documented over two decades | A live hypothesis, not yet proven |
What we actually do about it
We publish our own site’s content in answer-shaped pages, meaning each page opens with a direct, quotable answer to a real buyer question rather than a slow lead-in, on the theory that an assistant can only cite what it can parse cleanly. We maintain an llms.txt file and structured schema markup so assistants have a clean map of the site to draw from. And we track AI-assistant referrals as their own segment in GA4 on our own site, specifically so we’re not guessing at this shift, we’re watching our own numbers move, or not move, in real time. For the mechanics of building answer-shaped pages, see our companion piece on answer engine optimization.
What we don’t know yet
We don’t know how much of our own traffic mix will come from assistant citations a year from now, and we’re not going to pretend the early trend line is a forecast. What we do know is that watching it requires instrumenting for it now, before there’s a mature playbook, which is the same bet our systems work is built around more broadly: build the measurement before you need the answer, not after.