Google Rewired Search. Here's What I Shipped in Five Days.

Google flipped the default. AI Overviews now sit above the blue links for the overwhelming majority of English-language queries, and "AI Mode" — a conversational answer surface with no list of results at all — is rolling out the same way Gmail's tabs rolled out: quietly, then everywhere.

For an indie founder, the translation is short. The traffic model the entire indie SaaS playbook was built on for the last ten years just got rewired. Ranking on page one used to mean clicks. Now it means the model might cite you. Might.

I spent the last week shipping five changes to vibeflow.marketing in response. Here's what they are, why each one matters in the new game, and the playbook I'd run if you're staring at the same problem on your own site.

The Citation Game vs. the Ranking Game

Classic SEO was a popularity contest. Write the keyword-targeted page, earn the backlinks, climb the rankings, harvest clicks from page one. The page that ranked won the click.

AI search is a citation contest. A model reads a handful of pages, picks one or two, and quotes them by name. The cited page wins the entire query. The other results become invisible, because there is no result list.

The signals that win have shifted with it.

Server-rendered content beats client-rendered shells. RAG cannot read what JavaScript would have rendered. If your content lives in a hydrated bundle, the model sees an empty page.

Schema types that match how AI structures answers beat generic markup. FAQPage, ItemList, HowTo, AboutPage, Person — these tell the model exactly how your content fits into an answer. Generic Organization blocks don't.

Non-commodity content beats templates. A "10 best tips for X" listicle gets filtered. An opinionated, first-hand, specific recommendation gets quoted. The model is trained on ten thousand generic guides. It remembers the one that has a unique point of view.

Named authorship beats anonymous pages. Per Google's own AI optimization guidance, pages with verifiable authorship and clear entity grounding appear more often in AI Overview citations. The model treats authorship as a citation-safety signal — if it can verify who wrote it, it can stand behind the answer.

This is the new scoreboard. The old one still exists — ranking still matters for queries that don't get AI Overviews. But it's not the only game anymore. You need to optimize for both.

Five Changes I Shipped This Week

None of these are guesses. Each one targets a specific signal in the new game.

Schema enrichments — give the model a knowable entity. I expanded the JSON-LD across the site from a single SoftwareApplication block to a real entity graph: Organization, SoftwareApplication, WebSite (with the SearchAction), Person (me, with sameAs pointing to my LinkedIn and X), and BlogPosting with author linked to the Person @id and a real dateModified. A complete entity graph is the difference between "this is a source I can stand behind" and "this is text I scraped."

/about — a reference page built for entity-knowledge queries. Entity-knowledge questions ("what is VibeFlow", "who runs VibeFlow") are exactly the kind AI Overviews answer by pulling from the cleanest, most schema-rich source they can find. An /about page with AboutPage markup, linked via mentions to the SoftwareApplication and Person @ids in the root layout, is the canonical answer surface. Without it, the model has to guess from a homepage hero — and guesses get filtered.

/stack — opinionated tool roundup, not a listicle. I shipped a public /stack page with opinionated picks across thirteen categories of tools VibeFlow doesn't replace — image, video, email, landing pages, analytics, forms, the works. Each category names a first pick with a one-line reason, then three to five runners-up. ItemList schema with numberOfItems, BreadcrumbList, and mainEntity pointing back to the SoftwareApplication. Commodity roundups get ignored — the model has read ten thousand of them. "Loops if we were starting an ESP today. Resend if we were going to code the integration ourselves. Kit if the business is the newsletter." That specificity is citable.

/learn/seo — one differentiated primer as a test. We had eight /learn/[agent] primers running on essentially the same template. Generic-marketing-primer template is the textbook definition of commodity content. So I rewrote the SEO primer first — first-hand framing, specific examples, a clear point of view on what to do first — and left the other seven untouched as a control group. The hypothesis: the rewritten primer shows up in AI Overviews for "what is SEO marketing" and the long-tail variants. The other seven stay invisible. If the hypothesis holds, I'll roll the same treatment to the other primers one at a time, with real data per agent instead of guessing.

AI Search Audit — a new SEO agent feature. The existing SEO agent has audit modes that check what classic SEO cares about: title tags, meta descriptions, canonical, schema presence, OG tags. Useful for page one, but those are ranking signals. So I added a new audit mode (internally called evaluate_ai_search) that checks something different: whether your page is citable by AI Overviews, ChatGPT, and Perplexity. Eight evidence-based checks against the live page snapshot — server-rendering visibility (can RAG actually read your content?), non-commodity content scoring on four axes, schema type mapping against the types AI Overviews actually prefer, heading hierarchy for query fan-out, direct-answer lead detection, authorship and freshness signals, citation-ready format checklist, and a ranked priority-fix list. A page can pass classic SEO audits with flying colors and still be invisible to AI Overviews because it's a JS-rendered shell, or because the content is commodity, or because the model has no author to anchor the citation to. It went live this week.

The Playbook If You're Staring at the Same Problem

The honest version is that nobody has six months of clean citation data yet. The Google announcement is recent enough that most SEO commentary is still vibes. But the directions are clear and they don't conflict with anything that was already good practice.

Stop optimizing only for ranking. Start optimizing for citation. Two different scoreboards. Run both. A page can dominate one and fail the other.

Server-render or you're invisible. RAG fetches HTML and parses it. If your content is rendered client-side after a JavaScript bundle loads, the model sees an empty shell. Check by viewing the raw source, not the inspector.

Pick schema types that match how AI structures answers. FAQPage for question-shaped pages. ItemList for curated lists. HowTo for step-by-step. AboutPage for entity queries. Person for any byline. Organization and SoftwareApplication for the entity itself. Linked together with @id references, not floating as separate islands.

Lead every page with the direct answer in the first paragraph. Two sentences, then expand. AI Overviews almost always lift the first quotable answer-shaped sentence on the page. If yours is a hero tagline followed by positioning copy, you've given the model nothing to quote.

Ship non-commodity content. Opinions. Recommendations. First-hand experience. Specific numbers from your own work. The kind of writing a competitor cannot copy because they didn't live it. This is the hardest one and it's also the one that matters most.

Add named authorship with verifiable linkage. Real bylines. Real Person schema with sameAs pointing to real social profiles. dateModified that's actually maintained.

Build for query fan-out. A model receiving "best email tool for indie founders" will reformulate the query four or five ways under the hood. The page that wins is the one that answers all four reformulations on a single URL. Use H2s shaped like real user questions, not feature-list nouns.

Track citation, not just ranking. Set up alerts for when your domain appears in AI Overview snippets. Monitor which pages get cited and which don't. The data you collect now becomes your competitive edge later.

What I'm Watching Next

The thirty-day window from now is the one that tells us whether the work compounded. The signals I'm watching: which pages start showing up in AI Overviews for our target queries, whether the differentiated /learn/seo primer outperforms the seven control primers, and whether /stack starts getting cited for "best X for indie founders" categories.

I'll post the data when it's real. Not before.

The old SEO playbook isn't dead. The new one runs on top of it, and the founders who ship to it first are the ones AI Overviews will keep citing in six months. The work is small. The window for being early is not.

Found this useful? Share it.