How Google Uses AI in SEO: From RankBrain to AI Overviews

AI in SEO means two different things, and this page covers the one most people are actually asking about: how Google uses AI in search, and what that changed about ranking. If you’re looking for how to use AI tools in your own SEO workflow, that’s a separate question covered on our AI SEO hub; if you want the picture specific to this market, start with AI SEO in Ireland.

The short version: Google has been using machine learning in ranking since 2015, and almost none of it required anything new from site owners. What changed in 2024 and 2025 was not the ranking systems but the interface answers generated on the results page instead of a list of links. Google’s own documentation states there are no additional requirements to appear in AI Overviews or AI Mode, and no special optimisation is necessary. That single sentence is worth more than most AI SEO advice published since.

AI in SEO: the systems, in order

Hummingbird (2013). A rewrite of the core algorithm to interpret the meaning of a whole query rather than matching individual words. Pre-dates the machine learning era but set the direction.

RankBrain (2015). Google’s first machine learning ranking component, used to interpret queries it had never seen before. Google described it at the time as one of its most important ranking signals.

Neural matching (2018) and BERT (2019). Both improved Google’s understanding of language BERT specifically for how prepositions and word order change meaning. Google said at launch that BERT affected around one in ten English-language queries.

MUM (2021). A multimodal, multilingual model announced at Google I/O, used in specific features rather than broad ranking.

The helpful content system (2022–2024). Launched August 2022 as a separate site-wide classifier, then formally incorporated into Google’s core ranking systems in March 2024. Google deprecated it as a standalone system; helpfulness is now assessed across multiple core systems rather than by a single classifier.

SGE to AI Overviews (2023–2024). Introduced as the Search Generative Experience at Google I/O in May 2023, rebranded to AI Overviews and launched in the US in May 2024, reaching the UK in August 2024 and over 100 countries that October.

AI Mode (2025). First an opt-in Labs experiment, described by Google as rooted in its core quality and ranking systems. It arrived in the UK on 28 July 2025 as a tab on the results page.

What query fan-out actually changed

AI Mode introduced the one genuinely new retrieval behaviour in this list. Google’s documentation confirms that both AI Overviews and AI Mode may use a query fan-out technique issuing multiple related searches across subtopics and data sources to build a response.

The practical consequence is that your page may be retrieved for a sub-question Google generated internally, not for the query the user typed. You will not see that sub-question in Search Console. It doesn’t appear in any keyword tool, because it was never a keyword it was written by a model at the moment of the search.

This is the strongest argument for covering a topic properly rather than building one page per keyword variant. A page that answers the main question and the four obvious follow-ups has more surfaces available to it than a page targeting a single phrase.

What Google has confirmed, and what the industry has inferred

Most AI SEO content blurs this line. It’s worth drawing clearly, because the confirmed list is short and the inferred list is long.

Confirmed by Google:

• Generative AI features are grounded in core Search ranking and quality systems, and retrieve from the Search index (Google’s generative AI optimisation guide).
• Eligibility requires the same things as a normal search result: indexable, snippet-eligible, crawlable, useful.
• Query fan-out may be used by both AI surfaces.
• No third-party tool has access to Google’s internal ranking or AI systems Google explicitly warns against tools claiming otherwise.

Inferred by the industry, not confirmed:

• Any specific content structure that “gets you cited”. Answer-first writing is good practice for readers; it is not a documented retrieval requirement.
• Author credentials as a direct ranking input. E-E-A-T is a concept from the quality rater guidelines, not a measurable signal Google has confirmed.
• Word counts, paragraph lengths, or passage limits. Claims of this kind circulate widely and none are sourced to Google.

If you take one thing from this page: treat the second list as hypotheses worth testing on your own site, not as rules.

The regulatory layer, which matters more here than elsewhere

This is where UK and Irish businesses are genuinely in a different position from the US-written guidance most people are reading.

On 10 October 2025 the Competition and Markets Authority used its powers under the Digital Markets, Competition and Consumers Act 2024 to designate Google’s general search and search advertising services with Strategic Market Status. That designation let the CMA impose binding conduct requirements without having to prove illegal conduct first.

Two of those requirements landed in June 2026. The publisher conduct requirement, imposed on 3 June, gives publishers tools to prevent their content being used to power AI features such as AI Overviews, and strengthens their hand in negotiating content deals with Google. Google must also attribute publisher content properly with clear links in AI-generated results, and allow publishers to opt out of their content being used for fine-tuning AI models. The fair ranking and data portability requirements followed on 17 June 2026.

The visible result for site owners arrived at the same time. Google launched Search Generative AI performance reports in Search Console, giving dedicated views of impressions within AI Overviews and AI Mode, and rolled them out to all websites worldwide as of 31 August 2026. The rollout began in the United Kingdom, with an accompanying control to block content from appearing in AI features.

Read that sequence again. UK site owners got first-party AI visibility data before anyone else, because a UK regulator required it. That is a real, checkable advantage for businesses operating here, and almost no US-authored guide mentions it.

Two limits are worth knowing before you open the report: it covers impressions only, with no click data, and there is no historical backfill. It also doesn’t separate AI clicks, CTR, queries or conversions, and it groups AI Overviews and AI Mode together rather than reporting them individually.

Northern Ireland and Ireland are on different clocks

Worth knowing if you operate across the border, because the two markets got these features months apart.

Northern Ireland, as part of the UK, had AI Overviews from August 2024 and AI Mode from July 2025. The Republic of Ireland got AI Overviews in March 2025, in the EU expansion alongside Germany, Belgium, Italy, Austria, Poland, Portugal, Spain and Switzerland. AI Mode followed in October 2025, when Google extended it to Ireland, Germany, Spain, Portugal, Italy and Belgium France excepted.

So a Belfast business and a Dublin business searching the same query have had different results pages available to them for well over a year. If you’re benchmarking AI visibility across both markets, your baselines are not comparable and shouldn’t be charted as one line.

Core updates in this period

For context when you’re diagnosing a traffic change, the confirmed Search updates since the AI surfaces landed: the December 2025 core update, the March 2026 core update (27 March to 8 April), and the May 2026 core update (21 May to 2 June). February 2026 saw a Discover core update, which ran from 5 to 27 February a distinction worth keeping straight, since it’s often reported as a Search core update.

Google’s own guidance is to wait at least a week after a rollout completes before drawing conclusions from Search Console.

Checklist: is your site set up for how Google’s AI systems actually work?

Ten checks, all of which trace back to something Google has documented rather than something the industry has guessed.

1. Every page you care about is indexed and eligible for a normal search snippet.
2. Your key content is in crawlable HTML, not rendered only by client-side JavaScript.
3. Internal links connect related pages so retrieval can reach your deeper content.
4. Structured data matches what a visitor actually sees on the page.
5. Each page answers its main question and the obvious follow-up questions, so it can surface for generated sub-queries.
6. Your Google Business Profile details match your site, if you trade locally.
7. You have opened the Generative AI performance report in Search Console and recorded a baseline.
8. You know whether your AI-features opt-out toggle is on or off, and that it was a deliberate choice.
9. You’re tracking UK and Republic of Ireland visibility separately, not as one figure.
10. Your content contains something a competitor’s summary of the same topic could not contain.

FAQs

Ready to see where you stand?

ProfileTree, a Belfast-based web design and digital marketing agency, works with SMEs across Northern Ireland, Ireland and Great Britain on search visibility in both traditional results and AI answers. If you want a baseline before you change anything, request a free AI visibility check against our 12-point checklist.