GEO is still SEO: what Google actually said about AI search, llms.txt, and small business websites

On May 15, 2026, Google published a generative AI Search guide that says the quiet part plainly: for Google Search, GEO and AEO are still SEO. If you run a small WordPress site, the priority is not llms.txt. It is crawlable, useful, specific pages that Google can index and show with a snippet.

TL;DR

• Google says AI Overviews and AI Mode rely on core Search ranking and quality systems, so "GEO" for Google Search is not a replacement for SEO.
• A page must be indexed and eligible to appear in Search with a snippet to be eligible as a supporting link in AI Overviews or AI Mode.
• llms.txt can be a sensible documentation artifact, especially for developer-heavy sites, but Google says it is not required or specially supported for generative AI Search visibility.
• Google's June 3, 2026 Search Console changes are limited rollouts, not a global switch every site owner can already use.
• Small service businesses should fix crawlability, service pages, internal links, visible business details, page experience and structured data that matches visible content.

Table of contents

• What Google actually said
• What eligibility still depends on
• The GEO claims I would ignore first
• Where llms.txt does make sense
• What changed in Search Console on June 3 2026
• What a small service business should fix first
• When it is not worth caring about GEO yet
• Key takeaways

What Google actually said

The strongest source is Google's own guide, Optimizing your website for generative AI features on Google Search, updated on May 15, 2026. It explains that Google's generative AI Search features use retrieval-augmented generation and query fan-out, but the retrieved material still comes from Google's Search index and core quality systems.

That matters because it cuts through a lot of agency language. Google explicitly addresses "AEO" and "GEO" and says that, from Google Search's perspective, optimizing for generative AI Search is optimizing for the Search experience. In short: new interface, same foundation.

The interface is not small. AI Overviews and AI Mode can use query fan-out, which means Google may run several related searches across subtopics before building an answer. That does not mean you need a doorway page for every synthetic long-tail query. It means your real service scope, constraints and process need to be findable on pages that deserve to exist.

John Mueller made the same point in a May 21, 2025 Search Central post: focus on unique non-commodity content, page experience, technical access, preview controls and structured data that matches what visitors can see. No new discipline was announced there. The practical advice was normal SEO, applied to a Search results page that now has AI features.

What eligibility still depends on

Google's AI features documentation gives the simplest eligibility rule: to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to appear in Google Search with a snippet. Google's technical requirements for Search then reduce the technical side to access for Googlebot, HTTP 200 and indexable content.

For WordPress sites, that translates into very ordinary checks:

• important pages are not accidentally set to noindex;
• Googlebot is not blocked by robots.txt, a security plugin, a CDN rule or a host-level firewall;
• service pages return HTTP 200, not soft 404s or redirect loops;
• the core service content is present as indexable text, not only as JavaScript-rendered decoration;
• internal links connect the homepage, services, cases, articles and contact pages;
• snippets are allowed if you want eligibility in AI Overviews and AI Mode.

This is also where some modern architecture choices can quietly hurt you. A headless or heavily JavaScript-driven WordPress build can be rational, but only if the public HTML and routing still make sense for Search. I wrote about that trade-off in my headless WordPress analysis. If the main service copy only appears after client-side code runs, fix that before buying any GEO package.

Preview controls still matter. Google points site owners to nosnippet, data-nosnippet, max-snippet and noindex for Search AI features. Google-Extended is different: Google describes it as a control for training and grounding in some other Google AI systems, not as an AI Overviews or AI Mode opt-out.

The GEO claims I would ignore first

Google's mythbusting section is unusually direct. If a vendor leads with any of these, ask what evidence they have beyond a dashboard and a new acronym.

The common pattern: these tactics are not always harmful by themselves, but they become harmful when they replace the boring work. A clean service page with clear pricing constraints, a visible contact path and honest examples will usually do more than a bundle of AI-specific markup nobody asked for.

Where llms.txt does make sense

The original llms.txt proposal by Jeremy Howard, published on September 3, 2024, is not nonsense. It proposes a root /llms.txt Markdown file that gives language models a short overview of a site and links to LLM-readable Markdown resources. The proposal is especially coherent for documentation, APIs, software libraries and other sites where an agent may need concise reference material at inference time.

That is a real use case. It is just not the same as Google Search visibility.

For a developer platform, llms.txt might be worth generating automatically from the docs. For a small service business, it is usually a maintenance liability unless it costs almost nothing and mirrors the real site accurately. A stale llms.txt file that says you serve Rotterdam while the actual site says Amsterdam does not make the business easier to understand. It adds another place for inconsistency to hide.

My rule: if llms.txt falls out of your documentation pipeline for free, fine. If someone sells it as the reason Google AI Search will cite your WordPress site, keep your wallet closed.

What changed in Search Console on June 3 2026

There is one recent change worth watching carefully. On June 3, 2026, Google announced Search Generative AI performance reports in Search Console, with dedicated views for Search and Discover. The Search report shows impressions, pages, countries, devices and dates for generative AI features such as AI Overviews and AI Mode. Google says the reports are rolling out to a subset of websites while it tests and collects feedback.

Google also documented a Search generative AI control. The help page says the control is rolling out to a subset of website owners, appears under Settings > Search generative AI, and lets owners include or exclude links and content from AI Overviews, AI Mode and generative AI features in Discover. Excluding a site means no traffic or impressions from those features, and Google says the control is not used as a ranking or inclusion signal for other parts of Search.

Two cautions.

First, this is not a global, mature control at publication time. The June 3 Keyword Blog announcement framed the rollout in a UK subset context, and the Search Console help page says the control is still a subset rollout. Treat it as a limited test unless you see it in your own verified property.

Second, the new reports do not remove the need to measure business outcomes. The Generative AI performance report help page describes impressions and page-level visibility. It does not turn Search Console into a lead-quality report. For a service business, calls, forms, booking requests and qualified email clicks matter more than whether an AI surface showed a URL 400 times.

What a small service business should fix first

For a Dutch or EU service business, I would prioritize this order.

1. Make the money pages indexable. Homepage, service pages, contact page, about page, service-area pages and useful articles should be reachable through normal links, return HTTP 200, contain visible text and allow snippets. Search Console verification is not optional here.

2. Replace generic content with specific service pages. "7 tips for choosing an accountant" can be written by anyone. A useful page says which clients you serve, which region you cover, what the intake looks like, what documents are needed, what affects the price and where you draw the line.

3. Keep business details visible and current. Google specifically mentions Business Profile and business details. Show the business name, service area, contact options, opening hours, legal details where relevant, qualifications and real staff or practitioner information.

4. Use internal links like a map. Link service overview pages to individual services, service pages to cases or testimonials, and articles back to the relevant service page. Query fan-out does not remove the need for a coherent site structure. It makes weak structure more obvious.

5. Fix page experience without pretending it is magic. Slow pages waste attention, especially on mobile. If your WordPress site has high TTFB, heavy page-builder JavaScript or unstable layout shifts, fix those because visitors feel them. Core Web Vitals are useful diagnostics, not a standalone AI Search strategy.

6. Use structured data only where it describes visible content. LocalBusiness, Organization, BreadcrumbList and real FAQ markup can be useful when they match the page. Fake FAQ blocks and invisible schema fields are just another way to make the site less trustworthy.

7. Keep bilingual content clean. Dutch and English pages should have separate URLs, crawlable language switches and localized terms. Do not publish thin machine translations. The question a Dutch client asks about "WordPress onderhoud" is not always the same as an English query for "WordPress maintenance."

8. Separate Search visibility from AI features you add yourself. If you add an AI chatbot to the site, that is a different project with disclosure, privacy and operations questions. My EU AI Act chatbot disclosure article and WordPress AI API key policy article are better starting points for that side of the work.

This is not glamorous. It is also the part that keeps working when the acronyms change.

When it is not worth caring about GEO yet

Do not start with GEO if your site is not verified in Search Console, your main service page is a thin paragraph, or your contact form breaks silently. That is not caution. That is sequencing.

I would also ignore GEO for now if you cannot measure leads from organic traffic. Without form submissions, call clicks, booking requests or qualified email clicks, you will end up optimizing for impressions because they are easy to screenshot. AI Search reporting makes that temptation stronger, not weaker.

There is another case: very local businesses with no real content gap. If you run a two-person practice with a simple appointment flow and your current bottleneck is availability, not visibility, do not spend a month on AI Search theory. Keep the Google Business Profile current, keep the site technically healthy and move on.

Finally, do not chase llms.txt or special AI schema while your real pages contradict each other. AI systems are not magic. If your homepage says you serve all of the Netherlands, your contact page says Amsterdam only and your Google Business Profile has old opening hours, the problem is not GEO. It is messy information.

Key takeaways

• For Google Search, GEO and AEO are still SEO: Google's May 15, 2026 guide says generative AI Search is rooted in core Search ranking and quality systems.
• Eligibility starts with Search basics: indexed pages, snippets, Googlebot access, HTTP 200, indexable text and internal links.
• llms.txt is a reasonable optional documentation artifact for some sites, not a Google AI Search ranking lever.
• Ignore AI-only rewriting, artificial chunking, fake mentions and special AI schema claims unless the vendor can tie them to official guidance.
• The June 3, 2026 Search Console controls and reports are limited rollouts. Useful to watch, too early to treat as a universal planning assumption.
• Small service businesses should fix specific service content, current business details, structured data that matches visible content, page experience and lead measurement before buying anything called GEO.