How AI Is Changing SEO: The Fundamentals That Still Win

The rules of ranking are being rewritten in public. The job underneath them is not. If you are betting a content platform on how AI is changing SEO, the useful separation is between the parts that churn month to month and the parts that have held for twenty years. Build your defaults on the second set. Serve the person asking. Earn the kind of trust that credible sources will vouch for. Keep learning as the machinery shifts under you.

Duane Forrester makes this case with more standing than almost anyone. He helped launch Schema.org in 2011 and built Bing Webmaster Tools during years inside Microsoft’s search operation (Search Engine Land, 2026). His read on the current moment is blunt. AI is rewriting SEO, but the mission is the same.

How AI is changing SEO: what actually changed

Start with the interface, because that is where the disruption is real. AI Overviews now reach more than 2 billion monthly users, a number Sundar Pichai gave on Google’s July 23, 2026 earnings call (Digiday, 2026). Search answers in place more often than it sends a click. Zero-click searches hit 68% in the first four months of 2026, up from 60% two years earlier (Search Engine Land, 2026). When an Overview shows, the click rate collapses further. Pew Research clocked it at 8%, against 15% for results with no Overview above them (Digiday, 2026).

So rank and visibility have come apart. A page can hold the #1 organic position and still open below an AI-generated answer, which we traced in AI search visibility beyond organic rank. The retrieval layer is new. The presentation is new. The list of who gets cited is rewritten weekly, and that churn is what Forrester calls the wild west: unsettled and easy to over-fit a strategy to.

What earns a citation now

The mechanics beneath the churn are more learnable than the leaderboard suggests. Start with structured data. Marking up an article, its author, a product, or an FAQ with Schema.org vocabulary tells a machine what each element on the page really is. That legibility is what a retrieval system leans on when it assembles an answer. Forrester co-founded that vocabulary in 2011 for search crawlers. Fifteen years on, it feeds the models too.

Then there is the evidence of who is speaking. Google’s raters were told to weigh Experience in December 2022, when an extra E joined E-A-T in a guideline set that runs past 170 pages (Google Search Central, December 2022). In practice that shows up as a named author with a real bio and first-hand testing anyone can check. Cite primary sources, not a rewrite of the consensus. A page that says “we ran this, here is the number” carries weight that borrowed summary never will. Original data travels far. Aggregated opinion stalls.

Being quotable is the last piece. An Overview is stitched from passages, so the pages it pulls tend to answer the question in the first two sentences, under a heading that matches how people ask it. Bury the answer four screens down and the model quotes whoever led with it. That is the new snippet game. It rewards the page that puts the payoff first, aimed at a reader who may never click through. When only 8% of Overview searches end in a click, the citation is often the whole prize (Digiday, 2026).

The constants worth building on

Here is the part that has not moved. Underneath every interface change, search still does one thing. It matches a person’s intent to the most trustworthy answer it can find. Forrester’s verdict on what matters now is direct: trust has become the algorithm, and models weigh how often credible sources mention you more than any single on-page trick (Search Engine Land, 2026).

Three principles survive the transition intact, and they are the ones worth wiring into a content engine’s defaults:

• Serve the user first. Understand what someone wants at the moment they want it, and be the best answer. That mandate predates Google and will outlast AI Overviews.
• Build trust that others confirm. Accurate claims, cited sources, real expertise, and consistent mentions from places search already trusts.
• Learn continuously. The tactics turn over fast, so the system has to keep re-reading the terrain rather than freezing a playbook.

None of these is new. The rater guidelines were built on this exact basis years before the models arrived, and the models inherited the same test.

A builder’s read: defaults versus configuration

The design question is where to draw the line. Constants become defaults you never expose. Tactics become configuration you expect to change.

Serving the user is a default. So an engine should generate against a real query and a real intent, never a bare keyword slot. Trust is a default too, which puts quality checks in the pipeline rather than the review queue. We built that discipline into an E-E-A-T checker driven by an AI agent, where the model reasons over Google’s rater framework instead of hitting a score endpoint.

Continuous learning is the harder default to encode. It means the platform reads the current surface before it publishes, then routes each piece to where it belongs. That routing logic is the tactical layer, and it should change often. We mapped one version of it in a search journey content strategy, pairing each buyer doubt with the surface that resolves it.

Build for the mission, not the month

The temptation in a wild-west market is to chase the newest tactic and rebuild around it. Resist it. The teams that stayed durable across the last three search eras did the unglamorous thing, holding to users and trust while the interface thrashed. So ask a content platform one question. Does it encode the constants as defaults and leave the tactics as knobs you can turn? That is the design that survives the next rewrite, and the one after it.