Every second LinkedIn post in 2026 says SEO is dead. Every second SEO consultancy is selling a new "AI search optimisation" offering. Both are overstating the case. The data across nine client sites, measured every month since March 2025, tells a less dramatic but more useful story.
Organic search did not die. It split. The long-tail and product-intent traffic from traditional search is roughly stable. Informational top-of-funnel traffic is falling to AI answer boxes, which are sending a new, smaller, better-qualified stream of their own. Between the two, the net for our clients is flat to up, not down.
The winners are the teams that understood the split and adjusted the content strategy. The losers are the teams that kept publishing informational pillar pages in the post-2023 style, watched AI Overviews summarise the answer above the fold, and wondered where the traffic went.
The numbers we have
Nine clients, each with twelve months of GA4, Google Search Console, and Bing Webmaster data. Sectors: FinTech, legal, B2B SaaS, AU e-commerce.
Traditional Google organic. Median year-on-year traffic change: minus 12 per cent on informational queries, flat on commercial-intent queries, plus 6 per cent on product queries. The loss is concentrated at the top of the funnel.
AI answer traffic. Measured three ways. Search Console's AI-attribution query tag, Bing Webmaster's AI traffic breakout, and referrer strings matching ChatGPT, Perplexity, Claude, and You.com. This channel grew from roughly zero to between 4 and 11 per cent of organic traffic, depending on the sector. Conversion rate on this traffic is 1.8 to 2.4 times the conversion rate on Google organic.
Conversion quality is the part the headline numbers miss. Someone who clicks a citation inside an AI answer has been pre-qualified by the AI's framing. They already know what the company does, what the angle is, and why the source is cited. They arrive closer to the brief than a cold Google click.
What the AI answer channels reward
This is where the playbook changed, and where most teams have not updated.
Clear, self-contained sections. An AI summariser extracts the paragraph that answers the question. If your answer is spread across three sections, you do not get cited. If paragraph two of section three is the full answer, you do. Rewrite for extractability.
Named entities and structured data. AI models build their citations around entities they recognise. Schema.org markup on Organization, Product, Article, and Author (with a verified sameAs URL) has had disproportionate weight in what we have measured. The Person schema on authors in particular, with a verified LinkedIn or publication URL, has pulled clients into citation rotations they were not in six months earlier.
Authority signals that survive abstraction. AI answers strip visual brand. A hero image means nothing in an AI citation. A bio line that says "Chief Data Officer at Company X since 2019" means a lot, because it becomes the citation context. Rewrite your author bios accordingly.
First-party data. Surveys, benchmarks, datasets, original research. AI models cite these disproportionately because they need a source for the claim. We have had two clients go from zero AI mentions to weekly citations within two months of publishing a single piece of primary research. Nothing else we have tested comes close.
Fresh date stamps. AI models favour recent sources for current-state questions. A page dated 2022 on a fast-moving topic is invisible in AI answers now. Date stamps are being read again, and "Last updated" dates get honoured. We touch every evergreen page on a quarterly review cycle now.
What traditional Google still rewards, unchanged
The fundamentals did not reset.
Topical authority at the cluster level. Still the strongest signal for commercial-intent rankings. Ten well-linked pages on one tight topic beat fifty scattered pages.
Real links from real sites. The link graph still works. Digital PR that earns links from recognised publications still works. Link-building from directories and low-authority guest posts still does not, and Google is getting better at ignoring them.
Green Core Web Vitals. Still a ranking factor, still the bar the majority of sites we audit fail on. A 2 MB hero image is a 2 MB hero image regardless of who is running the algorithm.
E-E-A-T. Expertise, Experience, Authoritativeness, Trust. The Helpful Content system enforces it, and the AI answer boxes reward it too, because citation ranking is shaped by the same signals.
Internal linking. Tight clusters, varied anchor text, no stuffing. Unchanged for a decade.
The workflow we run now
One content calendar. Two distribution targets.
Every commercial-intent piece we publish is structured for both channels. For Google: the full deep article with internal links, proper schema, E-E-A-T signals, and mid-funnel keyword targeting. For AI answers: a clean TL;DR paragraph in the first 200 words that answers the implicit question, a structured-data block that makes the claims machine-readable, an author byline with Person schema.
We no longer publish short informational pieces unless they have a specific commercial tie-in. Pure informational content used to be fodder for links and brand; now it gets abstracted and replaced by the AI answer. The commercial pages that used to sit at the bottom of the funnel have to earn their own authority, and that is where the investment has shifted.
The measurement calendar changed too. Monthly dashboards with a single "organic traffic" number are not useful anymore. We split the dashboard into four panes: Google informational, Google commercial, AI answer traffic, direct brand. Each pane is treated separately, because the signals that move each are different, and bundling them hides the real story.
Where this is heading
Our working model for the next eighteen months.
Traditional Google organic will keep compressing at the top of the funnel. The compression will plateau because Google cannot abstract away every query without losing its advertising revenue model. Commercial and transactional queries will stay roughly where they are; Google needs those users to click through for Ads to work.
AI answer traffic will keep growing until it settles at somewhere between 20 and 30 per cent of the old organic share. It will not replace organic, because people still want nuance, alternatives, and shopping surfaces that a synthesised answer cannot serve. But it will be a permanent second channel.
The teams that win both channels will be the ones that stopped writing for SEO specifically and started writing for understanding, in structured prose, with proper authority markup on the pages. The SEO craft has not died. It has become slightly more honest.