Project Magi: Google's Next Big Change to Search
Project Magi was the internal Google initiative that transformed search. Here's what it was, what it became, and what it means for your content.
Quick verdict: Project Magi was Google’s internal codename for the AI overhaul of its search engine, the initiative that became Search Generative Experience (SGE), and ultimately Google AI Overviews. What started as a competitive response to ChatGPT-powered Bing is now a live product reaching over 2 billion users monthly. For content creators and SEO professionals, this isn’t just a feature update. It’s a structural change to how Google surfaces, summarises, and credits web content.
What was Project Magi?
In April 2023, The New York Times reported that Google had assigned approximately 160 engineers to an urgent project codenamed “Magi.” The mission: rebuild the core Google Search experience around AI before Microsoft’s ChatGPT-integrated Bing could take meaningful market share.
Project Magi wasn’t a single feature. It was an umbrella initiative covering several parallel experiments:
- A conversational AI layer that could answer questions without users needing to click through to a website
- Personalised results based on individual search history and preferences
- The ability to complete transactions (flights, purchases, bookings) directly within Search
- Visual AI search to identify objects and places from images
- Natural language understanding sophisticated enough to handle multi-step, complex queries
The urgency was real. Google had spent a decade treating search as a link-ranking problem. ChatGPT showed users that an AI could just answer the question. Magi was Google’s recognition that the rules had changed.
The timeline: Magi to SGE to AI Overviews
Understanding what Project Magi became requires tracing the full timeline.
May 2023: Search Generative Experience announced At Google I/O 2023, the company unveiled Search Generative Experience (SGE) as an experimental AI layer. The Magi codename was retired; SGE was the public face. Access was limited to US users who joined a Search Labs waitlist.
August–November 2023: International expansion SGE expanded to India, Japan, then 120+ countries, still within Search Labs. It was clearly experimental: answers were sometimes wrong, citations inconsistent, and the product visibly unfinished.
May 2024: AI Overviews launch At Google I/O 2024, SGE became AI Overviews and rolled out to all US users without a Labs requirement. This was the moment Project Magi’s vision became a default part of Google Search. Google projected it would reach 1 billion users by end of year.
October 2024: Global scale AI Overviews expanded to 100+ countries. The 1 billion monthly user milestone was reached ahead of schedule.
November 2024: First ads Sponsored placements appeared inside AI Overviews on mobile. The commercial infrastructure was live.
March 2025: Gemini 2.0 integration Gemini 2.0 became the underlying model, improving math, coding, and health query handling. AI Overviews began appearing on thousands of additional health-related topics.
May 2025: 200 countries, 40+ languages Arabic, Chinese, Malay, and Urdu were added. User base reached 1.5 billion monthly. Desktop ads launched in the US.
July 2025: 2 billion users AI Mode (the full conversational extension of AI Overviews) went from experimental to mainstream on mobile.
January 2026: Gemini 3 becomes default The current state: Gemini 3 powers AI Overviews globally. “Show more” buttons on mobile transition users directly into AI Mode for deeper conversational follow-up. Source links now average 15 per overview, up from under 7 in late 2024.
What AI Overviews look like in 2026
What Project Magi envisioned in 2023 is now the default experience for hundreds of millions of searches per day. AI Overviews currently appear on approximately 60% of all Google queries, up from 28% in May 2025. They’re no longer a novelty.
The current feature set includes:
- AI-generated summaries at the top of search results, synthesising multiple sources into a direct answer
- Inline citations with an average of 15 linked sources per overview (over double the November 2024 figure)
- AI Mode for multi-turn conversational follow-up, now integrated as a standard mobile feature
- Shopping integration: AI Overviews appear on 14% of shopping queries as of March 2026, a 5x increase from 2024
- Sponsored placements above and below overviews in the US and major international markets
- Gemini-powered reasoning for complex, multi-step questions
The niche distribution is striking. Finance queries now trigger AI Overviews 78% of the time (up from 11% in mid-2025). Business, careers, and relationships are all above 70%. Informational queries, the bread and butter of content publishers, are the most affected.
What this means for search traffic
The traffic impact is significant and well-documented.
| Query type | Average CTR reduction (AI Overview present) |
|---|---|
| Informational | 30–50% for top 3 positions |
| Commercial investigation | 10–18% |
| Transactional | 5–8% |
| Zero-click rate (informational) | 65–70% |
Position 1 click-through rates on queries with AI Overviews have dropped to 8–12%, compared to 28–34% on the same queries without them. Sites heavily dependent on informational content have seen 20–40% organic traffic declines.
This isn’t Google breaking the web. It’s the logical end state of what Project Magi set out to build: a search engine that answers questions rather than pointing to answers. The business model consequences for ad-supported content sites are real, but the user experience goal was always frictionless answers.
How to get cited in AI Overviews
The more useful question for content publishers isn’t how to fight AI Overviews. It’s how to be cited by them. Pages cited inside AI Overviews receive both a traffic signal and an authority signal that compounds over time.
Research suggests structured data implementation is the clearest lever available. Pages with proper FAQ, HowTo, Article, and Organization schema are cited at 2–3x the rate of equivalent pages without it. Schema markup is now table stakes, not an advanced technique.
Beyond schema, the content formats earning the most AI Overview citations are:
Step-by-step guides with numbered structure. AI Overviews frequently pull numbered steps directly from well-structured how-to content.
FAQ-rich pages where question headings are answered directly in the following paragraph. The question-answer pairing maps cleanly to how AI models extract factual claims.
Comparison content with structured criteria. Tables comparing tools, products, or approaches give the AI a clear signal about the informational purpose of the page.
Data-driven analysis with original statistics. Proprietary data and original research are cited more frequently because they’re unique: the AI can’t synthesise the same claim from five other sources.
Fresh content updated within the past 3–6 months receives significantly more citations than evergreen pages left unchanged. An updatedDate in structured data signals recency directly to Google’s crawlers.
The top sources cited in AI Overviews as of early 2026 include YouTube, Reddit, Wikipedia, and established health and finance authorities like NIH and Bankrate. The pattern is consistent: sources with clear topical authority, strong entity signals, and structured content perform best.
What Project Magi got right (and wrong)
Looking back from 2026, the Project Magi vision was largely accurate. Conversational search is mainstream. Transactions are completing inside Google. Visual search is part of everyday use. The competitive threat from Bing turned out to be overstated, but it pushed Google into a transformation it had been putting off for years.
What the original Magi roadmap underestimated was the publisher backlash and the reliability problem. Early SGE had a documented hallucination problem: confidently wrong answers with credible-looking citations. Google has iterated significantly on accuracy, but AI Overviews still misrepresent sources on edge cases. The “just answer the question” model works well for established facts and much less well for niche, specialised, or rapidly-changing information.
The monetisation path also took longer to prove than Google suggested. Ads inside AI Overviews are now live globally, but click behaviour on an AI-first result page is fundamentally different from a ten-blue-links page. Google is still figuring out how to maximise revenue from a product that suppresses the clicks that historically drove it.
Frequently Asked Questions
What was Google Project Magi?
Project Magi was an internal Google initiative launched in early 2023 to rebuild Google Search around AI. Approximately 160 engineers worked on the project, which aimed to make search conversational, personalised, and capable of completing transactions without users leaving Google. The project became the public Search Generative Experience (SGE) in May 2023 and evolved into AI Overviews in 2024.
Is Project Magi the same as Google AI Overviews?
Yes. Project Magi was the internal codename; Search Generative Experience (SGE) was the experimental product; AI Overviews is the current public name for the same underlying vision. The product has changed significantly since the Magi phase. It’s now powered by Gemini 3 and reaches over 2 billion monthly users.
Did Project Magi replace Google Search?
No. AI Overviews is an additional layer on top of traditional search results, not a replacement. Standard organic results still appear below the AI Overview. However, AI Overviews now appear on roughly 60% of queries, which significantly changes how users interact with results and how much traffic flows to individual pages.
How does Project Magi / AI Overviews affect SEO?
The primary impact is on informational queries. Sites targeting how-to, what-is, and explainer content have seen CTR drops of 30–50% when AI Overviews appear above their results. The offset is citation traffic: pages cited within AI Overviews receive a trust and authority signal. Structured data (FAQ, HowTo, Article schema), well-organised content, and clear topical authority are the most reliable ways to earn those citations.
When did Project Magi launch publicly?
The underlying technology launched publicly as Search Generative Experience at Google I/O in May 2023, initially available to US Search Labs participants. It became a default feature for all US users in May 2024 under the AI Overviews name, and reached 100+ countries by October 2024.
Is Google AI Overviews hurting publishers?
For publishers heavily dependent on informational organic search traffic, yes. Traffic declines of 20–40% are documented for the most affected content types. The effect is not uniform: commercial, transactional, and niche technical content is far less affected than broad informational queries. Publishers adapting to the new environment are focusing on original data, expert perspectives, structured content, and conversion-focused traffic rather than raw informational volume.
