AI Search Timeline (2022–present)

1. The shape of AI search since 2022

AI search did not appear gradually. It arrived sharply on 2022-11-30 with ChatGPT, was christened “Generative Engine Optimization” on 2023-11-16 by Aggarwal et al., and has changed shape four times since. Five eras, each triggered by 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 scale, ChatGPT Search,  multimodal,
voice,      Bard       Aggarwal coins    "GEO" enters     AI Mode at
GPT-3       preview    "GEO" on arXiv    marketing vocab  1B MAU

Each era introduced exactly one new constraint for content surviving in answers: the user no longer arrives (Shock), the surfaces multiply (Fragmentation), the default surface flips (Mainstream), the agent decides what to read (Stabilization). The dated rows below populate each era.

2. Pre-history (≤ Oct 2022) — what “before” looked like

Pre-2022 search had three building blocks that AI search later fused. Extractive answer engines (Featured Snippets and voice assistants) had taught users to expect a single direct response. Large language models existed and could generate, but did not yet retrieve — GPT-3 (June 2020) had no live web access. The merger of these two worlds — generation with retrieval — is what 2022-11-30 unlocked.

Extractive direct answers and generative language models were two separate worlds. AI search is the merger of the two — and the merger arrived as a single consumer product.

3. The Shock — Nov 2022 to Feb 2023

A 92-day window in which one consumer product made conversational answers mainstream, and the two largest search incumbents had to react. Nothing before this window matters operationally for GEO; nothing after has stopped.

Within ten weeks of ChatGPT’s launch, the search duopoly had to choose between integrating AI answers and losing the surface. Microsoft was first into the search box on 2023-02-07; Google previewed Bard one day earlier but kept it outside Search until SGE in May. The “shock” was less about ChatGPT itself than about how fast incumbents had to react — the AI-answer surface was now a strategic asset, not a research demo.

4. Fragmentation — Feb 2023 to Apr 2024

Every frontier lab and every major search vendor shipped something in fourteen months. The user-visible surface shattered into six-plus distinct AI-answer experiences, each with its own retrieval, citation behavior, and traffic model. This is the era in which the practical problem of “appearing in someone else’s synthesized answer” outran any single label — and in which Aggarwal et al. supplied one.

By April 2024, a single content team had to think about appearing in at least five distinct answer surfaces: ChatGPT, Bing Chat / Copilot, Google SGE, Perplexity, and Claude. The academic coinage in November 2023 trails the practical problem by almost a year — typical pattern; the field needed a label, and “GEO” (generative) won the naming race over “AEO” (answer) once synthesis became the dominant mechanism.

5. Mainstream — May 2024 to Dec 2024

In eight months, AI answers stopped being a new feature and became the default on the largest search surfaces in the world. The term “GEO” left arXiv and entered marketing decks. The downstream traffic measurement that defines today’s zero-click discourse also begins here.

Mainstream is the era in which two things stopped being deniable: (1) AI answers are the default surface on the world’s largest search engine, and (2) GEO is a discipline with stable enough patterns that vendors started selling tools and publications started running beats. “GEO” was no longer a niche acronym by year-end.

6. Stabilization — 2025 to present (May 2026)

AI search has stopped shocking and started behaving like a category. Multiple stable engines, agentic browsers, mature measurement tooling, the first peer-reviewed counter-evidence on GEO tactics, and the first scale traffic-shift studies. Growth has shifted from product launches to product evolution — agent flows, multimodal answers, deeper integration with personal data.

The 2026-05-19 I/O announcement is the first time a single AI-search surface crossed one billion monthly users. The chronicle from here forward will add rows, not eras — until a paradigm shift (agentic search-by-default? a new modality? a regulatory cap?) earns a new bucket.

7. Vendor strategy at a glance

A recognition table — for readers who want “who is who” without scrolling the full chronicle. Each row is one sentence; the per-vendor history lives in the linked entry.

8. The inclusion test — what makes a row

Not every AI event is a chronicle event. The single test that decides what gets a row: “would a content team’s tactics have changed because of this event?” If yes, it earns a dated row. If no, it does not — regardless of how prominent the event was in AI-news cycles.

Three concrete applications of the test:

• Model checkpoints (GPT-4.5, Gemini 1.5, Claude 4, etc.) earn a row only when they materially changed the user-visible answer surface. GPT-4 (2023-03-14) is in §4 because it upgraded the answers users were actually seeing inside ChatGPT. Most subsequent checkpoint releases are capability events a content team’s tactics need not respond to, and are filtered out.
• Vendor strategy and acquisitions (executive shuffles, funding rounds, M&A) earn a row only when they changed the cross-vendor answer surface. Perplexity Series A (2023-03-28) is in §4 because the iOS app + funding were the moment the retrieval-native chatbot started to scale; most subsequent funding rounds are filtered out.
• Pre-2022 search history (Featured Snippets evolution, voice-assistant feature releases, ranking-algorithm updates) is filtered out. Pre-AI search is well-covered elsewhere, and the GEO problem post-dates it. §2 holds one anchor paragraph for context, not a chronicle.

The inclusion test is also why this entry’s pre-history (§2) is one paragraph and one small table rather than a full era. Featured Snippets, voice assistants, and GPT-3 set the stage for AI search, but no content team’s AI search tactics depend on knowing their detailed dates.

References

The dated rows above link directly to each event’s primary source. Cross-checks and measurement studies — Pew 2025, Ahrefs 2025 (initial + re-measurement), Aggarwal at KDD ‘24, Puerto at NeurIPS ‘25 — are in the sources block at the top. The academic coinage event (2023-11-16 arXiv) has a dedicated GEO Wiki paper page at GEO: Generative Engine Optimization (Aggarwal et al. 2024).

For the conceptual entries this chronicle anchors against: Generative Engine Optimization (the front-door definition), SEO vs GEO (the “what changed for content” comparison), Generative Engine (the technical mechanism), and AEO vs GEO (the naming question this chronicle’s vendor events drove).