First published: 2 May 2026 · Last updated: 2 May 2026
SGE to AI Overviews: A Quick History (and Why It Matters)
Google announced Search Generative Experience at I/O 2023 as an opt-in Search Labs feature. For about 12 months it lived behind the Labs toggle, accessible only to users who explicitly enabled it. SEOs studied it. Most CMOs ignored it. In May 2024, Google rebranded SGE to AI Overviews and launched it to all US users by default. The opt-in was gone. Within weeks, SEMrush and Ahrefs reported AI Overviews appearing on roughly 6% of US searches. By late 2025, that number had crossed 30% on informational queries. By Q1 2026, the surface had expanded to most major markets including Singapore. Why this history matters: every "SGE optimization" article published before mid-2024 was written against an experimental interface that had different ranking logic than what shipped. Many of the schemas, query types, and content patterns those guides recommended were tuned for SGE's beta behaviour. After the rebrand, Google rewired the ranking signals around the production system, and several pre-2024 tactics now produce zero lift. If your team is still working off a 2023 SGE playbook, you are optimising for a thing that no longer exists.How AI Overviews Actually Work (Technical Architecture)
At a high level, AI Overviews runs a three-stage pipeline whenever Google decides a query qualifies for an AI-generated answer.Query qualification
Google's classifier decides if the query is "AI Overviews eligible" based on intent, ambiguity, and whether a synthesized answer adds value over the standard SERP. Most informational and "how to" queries qualify. Most pure transactional queries do not.
Source retrieval
Google retrieves a candidate set of pages from its index. The retrieval signals overlap with classical ranking (relevance, authority, freshness) but skew heavier toward content that contains directly liftable answer passages, structured data, and recognized entity associations.
LLM synthesis + citation
An LLM (Gemini, in 2026) synthesizes a 2 to 6 sentence answer from the retrieved passages, attributes citations to source URLs, and renders the result above the standard organic results. The cited URLs receive both a visible link and a "from [domain]" attribution.
The 6 Ranking Signals Google Uses for AI Overviews
Synthesizing public Google guidance, the Quality Rater Guidelines, the academic GEO research, and field testing across hundreds of queries, the signals that drive AI Overviews citation cluster into 6 categories.Entity recognition
Google must associate your brand and pages with the entities the query references. Wikidata presence, consistent NAP, Organization + LocalBusiness schema, and Knowledge Graph entries all feed this layer.
Answer-shaped content
The LLM lifts passages, not pages. Open every section with a 1-3 sentence direct answer before context, examples, or caveats. Buried answers do not get cited.
E-E-A-T evidence
Author bios with credentials, original first-hand experience, sourced statistics, and explicit dates of publication and last update. The 2024 Quality Rater Guidelines weight E-E-A-T heavier than the 2022 version.
Valid structured data
FAQPage, HowTo, Article, and Speakable schema where applicable. Schema gives the LLM machine-readable confirmation of what the page is about. Invalid or partial schema is sometimes worse than none.
Topical authority
One excellent article in a thin topic area underperforms a cluster of 8-12 related articles. Google treats topical breadth as evidence the site can be trusted on the subject.
Information Gain
The 2024 patent and Quality Rater Guidelines describe a concept of "Information Gain", the score for whether your content adds new information beyond what is already in Google's index. Regurgitation scores zero. Original data, contrarian analysis, and proprietary case studies score high.
Schema Markup That Actually Helps AI Overviews
The schema you deploy matters more than the volume. A page with one valid, complete FAQPage block usually outperforms a page with six different schema types where two are partial or invalid.<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [{
"@type": "Question",
"name": "What is Search Generative Experience?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Search Generative Experience (SGE) was Google's experimental AI search interface launched in May 2023. It was rebranded to AI Overviews in May 2024."
}
}]
}
</script>Validate every block in Rich Results Test before deploying. Partial or invalid schema can suppress AI Overviews eligibility.
The schema types that consistently move AI Overviews citation rates in 2026:
- FAQPage for question-answer blocks. Strongest signal for "how to" and "what is" queries.
- HowTo for step-by-step instructional content. Pairs with How-to AI Overviews specifically.
- Article with `author`, `datePublished`, `dateModified` populated. Reinforces E-E-A-T signals.
- BreadcrumbList for navigational context. Helps AI Overviews understand topical hierarchy.
- Speakable for content optimised for voice answers. Lower priority but adds incremental signal.
Schema types that do less than people think:
- Organization schema on every page (deploy once on homepage, link via `@id` elsewhere).
- WebPage schema (mostly redundant given the URL itself).
- Person schema if not tied to verified author entities.
For a deeper dive into schema implementation, see our guide to schema markup types.
Information Gain: The Scoring Concept Most SEOs Miss
In 2023 Google was granted a patent (US 11,755,569) describing a system for scoring "Information Gain" — the degree to which a web document provides information beyond what already exists in the index. The Quality Rater Guidelines update in 2024 introduced explicit guidance for raters to identify content that "adds value over and above" what other sources offer. Both signals feed into AI Overviews source selection.
- Original survey data ("we audited 50 Singapore sites and found...")
- Proprietary case study with named clients and real numbers
- Contrarian analysis of accepted SEO advice with supporting evidence
- First-hand experience from a verified expert (author bio matters)
- Original frameworks named after their creator
- Definitions paraphrased from other top-ranking pages
- "What is X" articles that summarise existing content without adding insight
- Generic "best practices" lists with no source-specific data
- Spun or AI-generated content without editorial enhancement
- Listicle aggregations of existing tools without comparative testing
Practical implication: a 1,200 word article with one original chart of proprietary data will outrank a 4,000 word article that synthesises information already in Google's index. This is why the "longer is better" SEO heuristic is dead for AI Overviews specifically. Density and originality beat length.
The 8-Step AI Overviews Audit (Technical Walkthrough)
Run this audit on your top 20 pages targeting informational queries. Total time: 2 hours.
- Identify AI Overviews eligibility per page. Search the target query in Google (logged out, Singapore IP). Note whether AI Overviews appears for that query. If it does, capture which sources are cited.
- Check entity association. Search "[your brand]" in Google. Look for Knowledge Panel. If absent, audit Organization + LocalBusiness schema, Wikidata entry, and Wikipedia eligibility.
- Audit opening 100 words of each page. Does it open with a direct, citable answer? If the answer is buried after intro paragraphs, rewrite.
- Validate schema in Rich Results Test. Run every page. Note which schema types are deployed and which are reported as valid. Fix invalid blocks first, then add missing FAQPage and Article where applicable.
- Score Information Gain per page. Does this page contain original data, frameworks, or analysis not present in the top 10 competing pages? If no, add at least one original element (proprietary stat, named framework, comparison test, expert commentary).
- Audit author signals. Every page targeting an AI Overviews-eligible query needs a visible author byline with credentials and link to author bio page with verifiable expertise.
- Check topical cluster depth. Does this page link to and from at least 5 other related pages on your site covering adjacent subtopics? If no, build out the cluster.
- Track AI Overviews position monthly. Use a tool (Profound, Athena HQ, Otterly) or manual prompt testing to record AI Overviews appearance and citation rate per query, monthly. Build the baseline now, measure delta in 90 days.
For a deeper technical walkthrough on the schema and entity layers, our technical SEO service covers the full audit and remediation programme.
Tools to Track AI Overviews Visibility
The market for AI search visibility tools matured rapidly in 2025. As of Q2 2026, the practical options are:
- Profound — best for enterprise multi-engine tracking (Google AI Overviews, ChatGPT, Perplexity, Claude). Strong dashboards, expensive.
- Athena HQ — focused on AI Overviews specifically. Good for SEO teams who want depth on Google's AI surface.
- Otterly — lighter weight, good for agencies tracking multiple clients.
- Manual prompt testing — for teams under SGD 5,000/month budget, a simple Google Sheet with 20 queries tested monthly across 4 engines beats over-investing in tooling.
None of these tools are free. None are perfect. The honest assessment is that manual testing on a stable query set still produces the highest signal-to-noise for most teams. Pair it with one paid tool for trend tracking once the baseline is established.
For a broader SEO foundation that supports AI Overviews work, our content strategy service builds the topical authority layer that makes individual page optimisations multiply in value.
Frequently Asked Questions
Is SGE the same as AI Overviews?
Yes, AI Overviews is the production rebrand of Search Generative Experience. SGE was the experimental Labs version (May 2023 to May 2024). AI Overviews is the version that ships to general Google search today. The underlying technology (LLM-synthesized answers above search results, with cited sources) is the same. The ranking signals were retuned during the rebrand.
Does AI Overviews appear on every search?
No. AI Overviews triggers on roughly 30% of informational queries in major markets as of Q1 2026 and rarely on transactional queries. Google decides per query whether a synthesized answer adds value over the standard SERP. Pure navigational queries (e.g. "facebook login") and most commercial queries (e.g. "buy iPhone 17") do not trigger AI Overviews.
How is AI Overviews different from featured snippets?
A featured snippet is one passage extracted verbatim from one source page. AI Overviews is a synthesized answer combining multiple sources. Featured snippets show one URL. AI Overviews shows 3-8 cited URLs. Optimising for one helps the other (both reward clear answer-first content) but they are different SERP features. See our guide to featured snippets for the older surface.
Will AI Overviews kill organic traffic?
It will reduce click-through rates on informational queries where AI Overviews appears, particularly when the AI answer fully resolves the user's question. Sites that rank inside the AI Overviews citation set retain meaningful traffic. Sites that rank in positions 1-5 of the standard SERP but are not cited by AI Overviews lose the most clicks. This is why getting cited inside the AI Overviews matters more than ranking position 1 in the classical SERP.
What is the single most important thing to optimise for AI Overviews?
Information Gain. Every other signal (schema, E-E-A-T, topical authority) is necessary but commodity. Original information that does not exist elsewhere in Google's index is what consistently lifts pages from "retrieved but not cited" to "cited in the AI Overviews answer". One original chart, one proprietary stat, one contrarian framework on each page changes the math.