AI Search Timeline (2022–present)

1. The shape of AI search since 2022

AI search did not emerge gradually. The public launch of ChatGPT on 2022-11-30 marked its sudden arrival. On 2023-11-16, Aggarwal et al. gave the field the name “Generative Engine Optimization.” AI search then passed through four distinct phases. Together with the pre-history period, they form the five eras below, each beginning with a specific dated event:

Pre-AI ──── Shock ──── Fragmentation ──── Mainstream ──── Stabilization
≤2022.10    2022.11–   2023.02–          2024.05–        2025–present
            2023.02    2024.04            2024.12
─────────   ─────────  ─────────────     ─────────────    ──────────────
Featured    ChatGPT    SGE, Claude,       AI Overviews,       Agent browsers,
Snippets,   Bing Chat, Perplexity grows,  ChatGPT Search,     multimodal answers,
voice,      Bard       Aggarwal names     "GEO" enters        AI Mode reaches
GPT-3       preview    "GEO" on arXiv     marketing language  1B MAU

Each era introduced exactly one new constraint for content that appears in AI answers: users no longer arrive at the source (Shock), answer surfaces multiply (Fragmentation), AI answers become the default (Mainstream), and agents decide what to read (Stabilization). The dated events below show when each constraint took effect.

2. Pre-history through October 2022

Before 2022, three building blocks were already in place. Extractive answer features such as Featured Snippets and voice assistants had taught users to expect a single direct response. Large language models could generate text but did not yet retrieve current information. GPT-3, released in June 2020, had no live web access. Combining generation with retrieval created the form of AI search that arrived on 2022-11-30.

Extractive direct answers and generative language models had developed separately. AI search combined them in a single consumer product.

3. Shock: November 2022 to February 2023

During this 92-day period, one consumer product brought conversational answers into the mainstream and forced the two largest search providers to respond. As AI answers entered major search products, content teams had to begin considering whether their pages would appear in those answers. That concern has continued ever since.

Within ten weeks of ChatGPT’s launch, the two largest search providers faced a choice: integrate AI answers or risk losing users to a new kind of product. Microsoft added them to the search box on 2023-02-07. Google had previewed Bard one day earlier but kept it outside Search until SGE arrived in May. The shock came not only from ChatGPT but from the speed of the response. AI answers had become a strategic part of search rather than a research demonstration.

4. Fragmentation: February 2023 to April 2024

Over the next fourteen months, every frontier lab and major search vendor introduced an AI product. Users encountered more than six distinct AI-answer experiences, each with its own retrieval system, citation behavior, and traffic model. Content teams now had to manage the practical problem of appearing in someone else’s synthesized answer before the field had settled on a common name for that work. Aggarwal et al. supplied one during this period.

By April 2024, a content team had to consider its visibility across at least five distinct answer products: ChatGPT, Bing Chat / Copilot, Google SGE, Perplexity, and Claude. The academic term arrived in November 2023, almost a year after the practical problem appeared. As synthesis became the dominant mechanism, “GEO” (generative) became more widely used than “AEO” (answer).

5. Mainstream adoption: May to December 2024

Over these eight months, AI answers moved from a new feature to the default on the world’s largest search products. The term “GEO” moved from arXiv into mainstream marketing. This period also produced the traffic measurements that now inform discussion of zero-click search.

By the end of 2024, AI answers had become the default experience on the world’s largest search engine. Vendors had begun selling dedicated GEO tools, and publications were covering the subject regularly. The acronym was no longer confined to a niche audience.

6. Stabilization: 2025 to May 2026

By 2025, AI search had several stable engines and its first agentic browsers. Measurement tools had matured, peer-reviewed research had begun to challenge GEO tactics, and large-scale studies were measuring traffic shifts. Product development shifted from launches to evolution, including agent flows, multimodal answers, and deeper integration with personal data.

The I/O announcement on 2026-05-19 marked the first time a single AI search product crossed one billion monthly users. Further updates are likely to add rows within the Stabilization era. A new era would require a paradigm shift, such as agentic search becoming the default, a new modality taking precedence, or regulation imposing a cap.

7. Vendor strategies at a glance

This table shows when each vendor entered AI search, its current products, and its audience.

8. The inclusion test for timeline events

Content teams only need to track AI events that change their tactics. The deciding question is: “Would a content team’s tactics have changed because of this event?” A yes indicates a meaningful change. Prominence in the AI news cycle is not enough.

Three concrete applications of the test:

• Model checkpoints (GPT-4.5, Gemini 1.5, Claude 4, etc.) matter only when they materially change the answers users see. GPT-4 did so on 2023-03-14 by improving the answers available in ChatGPT. Most later checkpoint releases did not require content teams to change their tactics.
• Vendor strategies and acquisitions (executive changes, funding rounds, and M&A) matter only when they change the answer products available across vendors. Perplexity’s Series A on 2023-03-28 mattered because the funding and iOS app marked the point when the retrieval-native chatbot began to scale. Most later funding rounds did not affect content tactics.
• Pre-2022 search history (the evolution of Featured Snippets, voice-assistant releases, and ranking-algorithm updates) provides context rather than a tactical change. The GEO problem came later, and content teams do not need the detailed chronology of these events to respond to AI search.

Featured Snippets, voice assistants, and GPT-3 all preceded AI search. They provide the context needed to understand the four post-launch phases, but their detailed release histories do not affect AI search tactics.

References

The event dates draw on primary vendor announcements. Measurement evidence comes from Pew 2025 and the initial and follow-up Ahrefs studies in 2025. Aggarwal’s KDD ‘24 publication marks the academic adoption of GEO, while Puerto’s NeurIPS ‘25 C-SEO Bench provides a counterpoint. The term was first published on arXiv on 2023-11-16 in GEO: Generative Engine Optimization (Aggarwal et al. 2024).

The shift from ranked links to synthesized answers is central to both Generative Engine Optimization and SEO vs GEO. Those answers are produced by a Generative Engine, while earlier extractive systems provide the historical context for AEO vs GEO.