Google AI Overviews vs classic SEO: what changes

AI Overviews do not replace classic SEO, they add a second contest on top of it. Ranking still decides whether your page is in the pool Google considers. Being quotable decides whether anything from your page ends up in the generated answer above the blue links. You can win the first and lose the second, and most sites currently do, because the two are optimised by different things.

What carries over from classic SEO?

More than the "SEO is dead" posts suggest. Everything that determines whether a page can be found and trusted still applies, unchanged:

• Crawlability and indexation. A page nothing can fetch cannot be retrieved, summarised or ranked.
• Page speed and rendering. Content assembled by JavaScript after load is content a parser may never see.
• Canonical hygiene. Conflicting canonicals and duplicate URLs split whatever authority a page has.
• Topical authority and links. When two pages say the same thing, the signals that used to break the tie still break it.
• Titles, headings, structured data. Machine-readable context is worth more now, not less.

If your technical SEO is broken, GEO is not a separate project you can start. It is the same project, and it is blocked.

What actually changes?

Three things, and they are the reason a well-ranked page can be absent from the answer sitting directly above it.

The unit of success changes. In classic search the unit is a position on a results page. In a generated answer the unit is a citation: a link the engine attaches to a claim it made. There is no position two through ten to fall back on. You are in the source set or you are not.

The reader changes. A person skims a page and forgives a slow start. A model extracts. It looks for a passage that answers the question on its own, and a page whose first 200 words are a warm-up has offered nothing extractable. Content that reads well and content that extracts well are not the same thing, and only one of them has been optimised for over the last twenty years.

The measurement changes. Rank is a number you can watch daily. Citation is a yes or no per question, and it varies between runs because the engine re-searches each time. That makes single checks useless and repeated checks necessary.

Does ranking still get you into an AI Overview?

It helps and it is not sufficient. Ranking is what puts your page in front of the retrieval step; extractability is what decides whether anything is taken from it. The failure mode is specific and common: a page ranks in the top three, the AI Overview above it cites three other sites, and nothing in the analytics explains why.

The usual reason is that the ranking page answers the question eventually and the cited pages answer it immediately. Extraction reads the top of the page. So does a reader in a hurry, which is why this change costs nothing to accommodate.

What does a GEO check measure that a rank tracker does not?

myGEOscore scores a page on 51 factors in seven categories. Six of them are causes, things about your page that you control, and they carry the score. How much each one counts depends on the kind of page: a homepage is not weighted like an article. On an article, the weighting is:

• Content structure and extractability, 25%
• Authority and trust, 25%
• Fact density and citability, 20%
• Structured data and schema, 15%
• Technical AI readiness, 10%
• Brand entity and mentions, 5%

The methodology page prints the same table for all five page types. The seventh category, AI platform visibility, is the outcome, and it is reported on its own, beside the score rather than inside it. No page-type rubric gives it any weight, so it moves the grade in neither direction: the score measures whether your page is citable, the checks measure whether it was cited. The AI Overviews check asks Gemini the same five questions a buyer would ask, with grounding enabled, and looks for your domain in the grounding sources behind the answer. Appearing in one of the five scores 4 out of 10, two scores 6, three scores 8, and four or more scores 10. Two sibling checks do the same against ChatGPT and Perplexity.

The score becomes a grade on a scale calibrated against real pages rather than against a wish: 85 and above is A+, 75 is A, 62 is B, 50 is C, 40 is D, and anything lower is F. Most sites land lower than they expect the first time.

Where should you start?

In weight order, because the weights are the answer to "what matters most".

1. Rewrite your openings. For the pages you care about, make the first two sentences answer the question the page is about, with no preamble. This is the single highest-leverage change available and it takes an afternoon.
2. Make your headings questions. They become the anchors an engine matches a query against.
3. Add extractable facts. Replace "significantly faster" with the number. Date your claims. Name your sources and link them.
4. Fix reachability. Confirm GPTBot, ClaudeBot, PerplexityBot and Google-Extended are allowed in robots.txt and not blocked at the CDN. This is binary and it silently cancels everything above it.
5. Ship the schema. Article, FAQPage, Organization and breadcrumbs give an engine machine-readable context it does not have to infer.
6. Measure again next month. One reading is a number, three readings are a direction.

Each of those maps to scored factors with their own guides, including what we check and how we score it, in the factor library.

What is worth measuring now?

Two things, monthly: whether the pages you care about are citable, and whether they are actually cited. The free check covers the first. It scores 36 factors, needs no signup, and returns a grade with the specific failures behind it in about a minute.

The second needs the live checks, which is what the paid report at $19 adds, together with the full 51 factor breakdown, code-level fixes and 3 monthly re-scans that re-run everything and email you what moved. The pricing page lays out the difference, the FAQ covers the common questions, and the about page explains who builds this and why.