Buyers are getting answers from AI before they ever click a result. AI Overviews, ChatGPT and Perplexity now sit between your content and your customer, and for a growing share of searches the journey ends right there, in the answer.
If you run a website that depends on organic traffic, you have probably already felt it: impressions holding steady or climbing while clicks flatten. The information your page holds is still being consumed. It is just being consumed through a summary that may or may not mention you.
TL;DR: Search is moving from ranking links to synthesising answers, and the unit of competition is now the citation, not the position. Clicks fall hardest on informational queries; commercial and comparison queries are where citations decide revenue. The brands that win are the ones AI can read, trust and quote accurately, and in most categories that bar is still low.
What has actually changed
Classic search was a referral system. Google ranked documents, users picked one, and the click was the currency everyone optimised for.
AI-mediated search is a synthesis system. Google's own framing when it launched AI Overviews was "let Google do the searching for you", and that is exactly what happens: the engine reads a set of sources, composes an answer, and cites a handful of them. Ahrefs measured the top-ranking result losing roughly a third of its clicks when an overview appears above it. Three consequences follow:
- Fewer clicks overall. For informational queries especially, the answer satisfies a large share of users on the results page itself.
- Winner-takes-most citations. An AI answer typically leans on a small number of sources. Being the third-best page on a topic used to earn a steady trickle of traffic. Now it can earn nothing at all.
- The brand impression happens without you. If ChatGPT describes your product inaccurately, or recommends a competitor when buyers ask "what is the best X", that impression is formed before your website ever enters the picture.
None of this means SEO is dead. It means the unit of competition has changed from the ranking position to the citation.
Not every query loses equally
The panic headlines treat "search traffic" as one thing. It is not, and the impact splits sharply by what the searcher is trying to do:
| Query type | What AI answers do to it | What you should do |
|---|---|---|
| Informational ("what is X", "how does X work") | Heaviest click losses. The answer is the product. | Compete for the citation, not the click. Structure content so it can be quoted. |
| Commercial and comparison ("best X for Y", "X vs Y") | Fewer clicks, but the citations shape the shortlist. | This is where revenue is decided. Own the comparison content in your category. |
| Branded and navigational (your name, your product) | Largely intact, and often growing as answers create curiosity. | Make sure what AI says about your brand is accurate. Watch branded search volume as a signal. |
| Local and transactional ("near me", "buy X") | Least affected so far. Intent still needs a website or a shop. | Business as usual, plus clean structured data so you surface in AI-assisted local results. |
If most of your organic traffic is informational, you feel this shift first and hardest. If it is commercial, the clicks were never the point: the shortlist was, and the shortlist is now written by a model.
This is an opportunity, not a threat
Here is the part that gets missed in the panic: most of your competitors are doing nothing about this. The brands that win are the ones AI can read, trust and quote accurately, and right now that bar is surprisingly low in most categories.
We have seen this play out with our own clients. When we rebuilt Loco Skates' content around the questions buyers actually ask and the authority signals AI engines trust, LLM-driven revenue grew 774% year on year, and the brand took a 30% share of voice in its category's AI answers. The traffic that arrives this way also converts well, because the answer has already done the qualifying: by the time someone clicks through from a citation, they have read a comparison, seen your name recommended, and arrived with a decision half-made.
How AI decides who to cite
Reverse-engineering every engine is a mug's game, but the patterns across AI Overviews, ChatGPT and Perplexity are consistent enough to act on:
- Extractable structure. Clear headings that match real questions, direct answers in the first sentence or two, lists and tables where they genuinely fit. AI quotes what it can cleanly lift. A page that buries its answer in paragraph six of brand storytelling gives the model nothing to work with.
- Verifiable claims. Specific numbers, named sources, dates. "We are a leading provider" is invisible to an engine looking for something citable. "Founded in 2018, working with 40 ecommerce brands across the UK" is a fact a model can repeat with confidence.
- Entity clarity. A consistent name, description and category for your brand across your site, your profiles and the wider web. Engines cite brands they can confidently identify. If your homepage says one thing, your LinkedIn another and your directory listings a third, the model hedges, and hedging models do not recommend.
- Genuine authority. Mentions and links from sources the engines already trust. This is the old authority game with a new referee, and the referee has a strong preference for named expertise: authored content, credentials it can verify, a brand that shows up where its industry talks.
What extraction-ready actually looks like
The single highest-leverage change is answer-first structure, and it is easier to show than describe.
Before: a heading that says "Our approach to sizing" followed by three paragraphs of context before any usable information appears.
After: a heading that says "How should inline skates fit?" followed by one direct sentence: "Inline skates should fit snugly with toes just brushing the front, roughly half a size below most trainers." Then the nuance underneath, for the humans who keep reading.
The second version answers the question a buyer actually asks, in a form a model can lift verbatim with your name attached. Nothing about it is worse for human readers. That is the general pattern: the changes that win citations are mostly just clarity, applied ruthlessly.
What to do this quarter
- Audit your visibility. Ask the engines your buyers' top twenty questions and record who gets cited, what is said about you, and what is wrong. This baseline takes an afternoon and changes the conversation.
- Restructure your best content for extraction. Not new content, your existing winners. Question-led headings, answer-first paragraphs, clean markup.
- Fix your entity. Consistent organisation schema, aligned profiles, an about page that says plainly what you do and for whom.
- Monitor monthly. AI answers move constantly. Share of voice in your category's answers is now a metric worth tracking alongside rankings.
This is exactly the work our AI Search & GEO service productises: visibility audits, extraction-ready restructuring, entity work and monthly share-of-voice tracking, run alongside the SEO you already do rather than instead of it.
The measurement question
Traffic from AI answers is measurable, but it hides by default: some arrives with referral data, some looks direct, and some shows up as branded search from people who read an answer and then went looking for you.
Practically, that means three jobs. First, capture the referrers you can see: chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com all pass referral data on their clickable citations, so build a custom channel group for them in GA4 rather than letting them dissolve into "referral". Second, watch branded search demand as the lagging indicator: people who read an answer and later search your name are answer-driven traffic wearing an organic disguise. Third, accept that some of it stays dark and measure the trend, not the absolute.
For the deeper version of this argument, two of our long reads take it further: The Overview Effect follows a publisher rebuilding its visits after AI Overviews decoupled visibility from traffic, and The Phantom Query shows why the sub-queries behind AI answers never show up in your keyword tools at all.
The goal is simple: be the answer, not a link nobody clicks.
Search is not shrinking, it is concentrating. Fewer clicks will be distributed among fewer sources, chosen for structure, clarity and authority. Every quarter you wait, the engines settle further on the sources they already trust. Being early to this is still possible. It will not be for long.
