What Stays the Same

Why So Much Carries Over

There is a structural reason, and it is worth stating before the list, because it explains why this is not wishful thinking from people invested in the old discipline.

When an assistant needs current information it searches the web. It does not have a private index of its own built from scratch. Google AI Overviews sit directly on top of Google's results. Copilot is grounded in Bing. Assistants that browse use conventional search infrastructure to find candidates.

That means the entrance to the AI answer pipeline is the same entrance as before. If you are not indexed, not crawlable or not competitive in ordinary search, you are eliminated at step two of the process described in How AI Decides What to Recommend, before anything specific to GEO becomes relevant.

The Fundamentals That Transfer Unchanged

Crawlability and indexing

Unchanged, and now with an additional set of user agents to account for. If a page is blocked, orphaned, buried behind a login, or only exists once javascript has executed, it is out of the running.

The AI crawlers are broadly less capable at javascript execution than Googlebot, not more, which makes server-rendered HTML more important than it was rather than less.

Site architecture and internal linking

Clear hierarchy, sensible URLs, meaningful internal links between related pages. This helps machines understand what your site covers and how its parts relate, which is exactly what entity resolution needs. Internal linking is one of the few tactics that is genuinely more valuable now, because relationships between concepts are what these systems are built to work with.

Page speed and technical health

Slow pages get abandoned by crawlers with limited budgets, and by agentic systems working through a task on a timer. Broken canonicals, redirect chains and duplicate content confuse machines exactly as they always have.

Genuine expertise

Google's E-E-A-T framing was always a description of what good sources look like rather than a set of tags to add. That description transfers completely. Named authors with real credentials, first-hand experience, sources for claims, transparent corrections: these were good practice before and they are the specific properties that make a passage safe for a system to repeat.

Earned links and mentions

Links still matter to the search layer that feeds retrieval, and independent mentions still matter to what a model learns about you. The activity that earns both, being genuinely worth writing about, was always the durable version of link building. It is now the durable version of two things at once.

Understanding intent

Working out what someone actually wants when they ask a question, rather than what words they used, was always the core skill. It matters more now, because the system rewrites the question before searching, and what it rewrites toward is intent rather than phrasing.

What Changed Is Emphasis, Not Validity

Several practices remain correct but shift in weight. This is the part most "SEO is dead" arguments get wrong: they mistake reweighting for invalidation.

Practice	Still valid	Change in emphasis
Keyword research	Yes	Shifts toward the questions and phrasing behind the keyword, since the system searches its own rewritten terms
Long-form content	Yes, when the length is substance	Padding now costs you at the selection step, where it previously cost nothing
Title and meta tags	Yes	Still control conventional search appearance; carry less weight inside a generated answer
Structured data	Yes	Clarifies entities and relationships, which is more useful now, but is not confirmed as a direct input to answer selection
Backlinks	Yes	The mention alongside the link now carries independent value, whether or not the link is followed

The Uncomfortable Diagnosis

In practice, sites that perform badly in AI answers are usually failing at something that was already broken and was previously survivable, rather than at anything GEO-specific.

That is a pattern from twelve years of technical SEO audits rather than a measured proportion, and the audit at the end of this lesson is how you find out whether it applies to you.

Search was forgiving. A slow site with messy architecture and thin author information could still rank on the strength of a strong domain and a lot of pages. Retrieval is less forgiving, because the selection step throws away material before it is judged on merit, and a poorly structured page is exactly what gets thrown away.

So the honest first move may not be to learn GEO. It is to fix the SEO they deferred.

Audit The Foundation First

Do this before spending anything on GEO tooling or content. Half a day, and it regularly finds the actual problem:

1. Check robots.txt for rules affecting AI user agents. Confirm you are allowing or blocking deliberately rather than by accident. Managing AI Crawler Access covers the trade-offs.
2. View source on your five most important pages, with javascript disabled. If the substance is not in the HTML, retrieval systems may not see it at all.
3. Verify your site in Bing Webmaster Tools and compare indexed pages against Google Search Console. A large gap means you are effectively absent from Copilot.
4. Check that every important page names an author with a real biography and carries a visible date. Absence of both is the most common trust failure and the cheapest to fix.
5. Take your ten most valuable pages and check whether each answers one clear question in one clear place, or spreads the answer across the whole document.
6. Only after all five: start on anything specific to GEO.

Key Takeaways

• The retrieval layer behind AI answers runs on conventional search infrastructure, so the entrance is the same entrance as before.
• Crawlability, architecture, speed, expertise, earned links and intent understanding transfer unchanged.
• Several practices shift in emphasis rather than validity. Padding is the clearest loser, internal linking the clearest gainer.
• AI crawlers handle javascript less reliably than Googlebot, which makes server-rendered HTML more important, not less.
• Most poor AI visibility is deferred technical SEO rather than a GEO-specific failure. Audit the foundation before buying anything.