How AI Search Works
1
Attention Is All You Need: The Google Paper That Accidentally Ended Google's Search Monopoly.
On June 12, 2017, eight researchers at Google Brain uploaded a 15-page paper to arXiv. It accidentally made Google's own search business obsolete.
2
How Large Language Models Actually Work: Tokens, Context Windows, and Why Your Content Gets Ignored.
The answer to "how do I get cited by AI search?" is technical, not strategic — rooted in how large language models actually process information.
3
What Is RAG: the Only Thing That Matters for AI Search.
Once you see how Retrieval-Augmented Generation works, the rest of AI search visibility clicks into place. It's the mechanism that connects your content to every answer these systems produce.
4
5
How Google AI Overviews and AI Mode Actually Work: Why Your SEO Rankings Don't Guarantee AI Visibility.
AI Overviews aren't a ranking bonus layered on top of existing positions. In May 2024, Google restructured how most people receive information — the blue links became supplementary.
6
How to Be Surfaced in AI Search: What the Platforms' Own Documentation Actually Says.
Most GEO content produces a list of tactics and implies they work across all platforms. This article goes directly to what each platform has actually published about how its own system works.
7
Why the Bing AI Performance Report Is the Most Underused Tool in GEO.
Microsoft's Bing AI Performance report is the only first-party citation dashboard from any AI search platform. Most practitioners aren't using it.
8
The GEO Documentation Gap: Why Three Platforms Haven't Told You How to Rank in Their Systems.
Google published a guide. Microsoft published measurement tooling. Anthropic, OpenAI, and Perplexity have published nothing equivalent. Understanding why is more useful than waiting.
9
Constitutional AI vs RLHF: Why the Alignment Method Affects What Gets Cited.
ChatGPT is aligned with RLHF. Claude is aligned with Constitutional AI. These aren't interchangeable — and the difference affects what each model treats as trustworthy and citable.
Writing  ·  Series: How AI Search Works  ·  Part 4 of 6

The AI Search Race: ChatGPT, Gemini, Claude, Copilot, and Perplexity. A Technical Timeline.

The five major AI search players are architecturally different, trained differently, and retrieve information differently. Those differences determine where your brand shows up, and why you might be visible in one engine and invisible in another.

Published1 June 2026
ByThomas Cox
Read time14 minutes
Filed underSeries · AI Timeline · Part 4 of 6

I've pulled this timeline together as a reference for anyone trying to make sense of where the AI search market actually came from. The models, the milestones, the architectural decisions that shaped every AI search product currently in market. It all moved fast enough that even people inside the industry lost the thread.

This article covers all five major platforms: Google/Gemini, OpenAI/ChatGPT, Anthropic/Claude, Microsoft Copilot, and Perplexity. Return to it when you need to place a model or milestone in context. In late 2021, the dominant question in technology was whether large language models would ever be practically useful. By mid-2023, the question was how quickly they would restructure the information economy. That shift happened in roughly eighteen months.

2022
Nov 2022
ChatGPT launches
OpenAI releases ChatGPT on 30 November. GPT-3.5 with RLHF behind a simple chat interface. One million users in five days, one hundred million in two months — the fastest consumer product adoption in recorded history.
For most of the world, this was the first encounter with a capable LLM. The implications for search became obvious within weeks.
Dec 2022
Perplexity AI launches
Founded by Aravind Srinivas, Denis Yarats, Johnny Ho, and Andy Konwinski. Unlike ChatGPT, Perplexity was built from day one as a search product: every response grounded in retrieved sources, every answer with citations.
The thesis: the answer engine replaces the link engine. At launch almost no one noticed. Within two years, Google was paying attention.
2023
Jan 2023
Microsoft $10B investment in OpenAI
Microsoft extends its OpenAI partnership with a $10 billion commitment, gaining exclusive commercialisation rights and Azure as the compute layer. Intent: bring GPT-class capability to Bing and challenge Google.
Signalled to the entire industry that search was existential — and that Google's moat was at risk.
7 Feb 2023
New Bing with GPT-4
Microsoft launches the new Bing powered by GPT-4 — the first large-scale LLM integration in a major search product. Later renamed Microsoft Copilot. Early responses were erratic; Microsoft applied restrictions within weeks.
The first direct commercial threat to Google's 25-year search dominance.
8 Feb 2023
Google Bard — $100B error
Google announces Bard one day after Bing. A promotional video shows Bard claiming the James Webb telescope took the first images of exoplanets — factually wrong. Alphabet loses ~$100B in market cap by close of trading.
The hallucination risk, made brutally expensive in public. A stark illustration of how much was riding on getting this right.
Mar 2023
GPT-4
The first widely accessible large multimodal model — text and image inputs. Passes the bar exam at ~90th percentile. Becomes the foundational model for ChatGPT Plus and the API.
Sets the benchmark every subsequent model is measured against for the following year.
Mar 2023
Claude (Anthropic)
Anthropic releases Claude using Constitutional AI — trained against an explicit set of principles rather than RLHF. More cautious, less prone to harmful outputs. Positioned as the safe, reliable alternative for enterprise.
A third viable alignment path beyond RLHF. Proves safety and capability are not mutually exclusive.
May 2023
Perplexity Series A — $25.6M
Perplexity raises a $25.6M Series A (including Nvidia), launches Pro subscription. Monthly visitors hit 2 million — faster growth than almost any observer predicted at launch.
The answer engine thesis starts looking credible to investors.
Jul 2023
Claude 2 — 100K context window
Anthropic releases Claude 2 with a 100,000 token context window (vs GPT-4's 8K at launch). Public access opens via claude.ai. The window allows processing of entire codebases or research document sets in a single pass.
Context window becomes the central competitive dimension for the next 12 months.
Sep 2023
Bing Chat becomes Microsoft Copilot
Microsoft renames Bing Chat to Copilot, signalling a broader strategy: AI as the layer across the entire Microsoft product ecosystem — Windows, Office, Teams, Azure, Bing — not a feature within one app.
Microsoft's AI strategy shifts from search challenger to enterprise AI platform.
Dec 2023
Gemini 1.0
Google DeepMind releases Gemini 1.0 — natively multimodal from architecture up, first model to achieve human-expert performance on MMLU, state of the art on 30 of 32 benchmarks. Three tiers: Ultra, Pro, Nano.
Google's ground-up rebuild works. Technical parity with GPT-4 regained after the Bard embarrassment.
2024
Feb 2024
Bard becomes Gemini; Gemini 1.5 Pro
Google renames Bard to Gemini and announces Gemini 1.5 Pro with a 1 million token context window — an order-of-magnitude expansion using Mixture of Experts architecture.
Context window competition escalates. Entire film scripts, codebases, or long document sets now processable in one inference call.
Mar 2024
Claude 3 family
Anthropic releases Haiku (fast/cheap), Sonnet (balanced), and Opus (most capable). Claude 3 Opus outperforms GPT-4 on most benchmarks at launch. The three-tier naming structure becomes the industry template.
Anthropic establishes itself as a genuine frontier lab, not just the "safe" alternative.
May 2024
AI Overviews launch at Google I/O
AI Overviews (successor to Search Generative Experience) launches in US Search — AI-generated summaries above blue links for a significant share of queries. Gemini 1.5 Flash ships simultaneously, optimised for cost-efficiency at search scale.
AI search becomes a live product for hundreds of millions of users. SEO as previously practised is directly disrupted.
May 2024
GPT-4o
OpenAI releases GPT-4o ("omni") — natively multimodal across text, image, and audio, with real-time voice capability. Significantly faster than GPT-4. Becomes the default model in ChatGPT.
Speed and multimodality at parity with capability. The "slow but smart" trade-off collapses.
Jun 2024
Claude 3.5 Sonnet
Best-in-class for coding at launch — outperforms GPT-4o and Gemini 1.5 Pro on SWE-bench. Introduces Artifacts, generating code and documents in a side panel — becomes one of Claude's most-used features.
Coding becomes Anthropic's clearest competitive differentiator.
Jul 2024
Perplexity Publisher Programme
First revenue-sharing arrangement from any AI search platform to content creators. A direct response to criticism that AI search extracts value from publishers without compensation.
First acknowledgement from an AI search platform that the publisher relationship needs to be sustainable.
Sep 2024
OpenAI o1 — the reasoning model
A new paradigm: instead of a single forward pass, o1 spends compute at inference time "thinking" — an internal chain-of-thought before answering. Substantially better on complex reasoning, maths, and coding. Slower and more expensive.
The scale paradigm ("bigger = smarter") is replaced by the reasoning paradigm ("think longer = smarter").
Oct 2024
ChatGPT Search
OpenAI integrates live web search directly into ChatGPT. Retrieved results, synthesised responses, citations — the same architecture as Perplexity, applied to the world's most-used AI product. Framed explicitly as a Google challenger.
The AI search competitive dynamic becomes direct, not adjacent. Google's core product now has a named competitor with 200M+ users.
Dec 2024
Gemini 2.0 Flash; Deep Research; "Agentic era"
Google closes 2024 with Gemini 2.0 Flash (new default for AI Overviews), Deep Research (agentic multi-step web research), and a declared "agentic era" — AI that plans and executes tasks, not just answers questions. OpenAI ships its own Deep Research the same month.
Agents arrive in production simultaneously at Google and OpenAI. The next phase of the race begins.
2025
Jan 2025
Gemini 2.0 Flash goes GA
Becomes Google's default across all AI products. Includes native tool use — code execution, web search, image generation within the model's context. Fast, multimodal, and tool-capable as the search default.
Google's production AI search model can now act, not just answer.
Feb 2025
Claude 3.7 Sonnet — extended thinking
Anthropic's reasoning model — "extended thinking" mode generates an internal reasoning trace before answering. The first "hybrid" model: rapid standard responses and deliberative reasoning in the same product. Outperforms o1 on several coding benchmarks.
Every major lab now has a reasoning model. The paradigm shift is complete.
Mar 2025
Gemini 2.5 Pro
Google's first native thinking model with chain-of-thought reasoning. Tops major leaderboards at launch including coding, maths, and multimodal understanding. Google's most capable public model to date.
Google leads the benchmarks for the first time since GPT-4 launched.
Apr 2025
o3 and o4-mini — reasoning with tool access
OpenAI adds tool access during the reasoning process — models can invoke web search and code execution as part of their thinking trace, not just in output. Substantially more capable for research and analysis tasks.
The line between reasoning models and agentic systems blurs. Thinking and acting become one process.
May 2025
Claude 4; Claude Code GA; $1B ARR
Anthropic releases Claude 4 (Opus and Sonnet), launches Claude Code into GA — an agentic coding tool that reads, writes, executes, and tests autonomously. Anthropic discloses $1B in annualised revenue.
Anthropic is confirmed as a commercially viable enterprise AI provider at scale, not just a research lab.
May 2025
Google I/O 2025: AI Mode and query fan-out
AI Mode launches — multi-stage RAG with query fan-out. User queries decomposed into multiple sub-queries, retrieved independently. Key finding: only 14% of URLs cited by AI Mode ranked in the traditional top 10 for the primary query.
Organic search rankings are a poor predictor of AI search visibility. The SEO playbook needs a new chapter.
Jun 2025
Perplexity: 1B monthly queries; $14B valuation
Perplexity reaches 1 billion monthly queries, raises $500M, hits a $14B valuation. Growth from 2M monthly visitors in May 2023 to 1B queries in June 2025 — among the fastest sustained adoption curves in consumer software history.
The answer engine thesis is validated at scale.
Aug 2025
GPT-5
OpenAI releases a unified model merging GPT-4o's generative capability and o-series reasoning into one architecture. Faster than o1 and o3 while matching or exceeding their benchmarks. Replaces both GPT-4o and o3 as the ChatGPT and API default.
The reasoning/generative split collapses into one model. One frontier model to rule them all — for now.
Oct 2025
Perplexity Comet browser
Perplexity releases an AI-native browser integrating the answer engine into the browsing experience — agentic web tasks (booking, forms, research) from within the browser.
Perplexity's most direct challenge to Google's control of the browser and search interface layer.
Dec 2025
Gemini 3 Flash
Google's new default model across AI Overviews and Google AI features. Faster and more capable than Gemini 2.0 Flash across multimodal tasks. Google's fourth default model swap in under two years.
The pace of default model turnover signals that the speed of the race is not slowing.
2026
Feb 2026
Claude Opus 4.6 and Sonnet 4.6
Opus 4.6 adds 1M token context and Agent Teams — networks of Claude agents coordinating on complex tasks. Sonnet 4.6 hits 72.5% on OSWorld (computer use). Anthropic's most agentic release to date.
Agent-to-agent coordination enters the mainstream product. AI that delegates to itself.
Feb 2026
Gemini 3 Deep Think
A Gemini variant optimised for extended reasoning in science, mathematics, and complex multi-step tasks. Google's response to the deliberative reasoning paradigm. Most capable Gemini model for analytical tasks.
Google now has both speed (Flash) and depth (Deep Think). Two-speed model strategy confirmed.
Feb 2026
Bing AI Performance report
Microsoft releases the first first-party citation measurement tool from any major AI search platform — showing publishers and SEOs how their content performs in Copilot and Bing AI answers.
The first acknowledgement from a platform that AI search visibility is measurable and needs to be reported.
Feb 2026
Perplexity goes subscription-only
Free tier removed. 34M monthly active users, $22.6B valuation. The subscription pivot signals confidence in product-market fit and a shift toward sustainable unit economics.
Perplexity transitions from growth-at-all-costs to a sustainable business. The answer engine is a real product, not a demo.
Apr 2026
GPT-5.5 and Claude Opus 4.7
OpenAI releases GPT-5.5 (state of the art for coding). Anthropic releases Opus 4.7 with 3× vision resolution and self-verification — the model checks its own outputs against retrieved sources before responding.
Self-verification is a direct architectural response to hallucination in enterprise deployments. Accuracy becomes a product feature, not just a benchmark.
May 2026
Google publishes first official AI search optimisation guide
Google Search Central confirms RAG/grounding as the underlying mechanism for AI Overviews and AI Mode, confirms the standard Search index is the retrieval source, and outlines content quality signals for AI visibility. The first official guidance from any platform.
The framework this series presents is officially validated. Build for the architecture, not the brand name on the wrapper.
Today
The race is still open
Five platforms compete. No durable capability lead. The frontier shifts every few months. What's state-of-the-art at launch is competitive-but-not-leading within six months.
Don't build a strategy that bets everything on one platform. Build for the architecture — crawlability, factual precision, clear structure, non-commodity value — and it works across all of them.

The shape of the race.

Looking across this timeline, three patterns stand out clearly.

The reasoning paradigm replaced the scale paradigm. From 2020 to 2023, the dominant assumption was that bigger models were smarter models. From 2024 onwards, the question became about compute at inference time: not just how large the model is, but how much it thinks before answering. o1, Gemini 2.5 Pro, Claude 3.7, and GPT-5 all reflect this shift. The race is no longer purely about parameter counts.

The agentic era arrived faster than expected. In 2022, "agentic AI" was a research concept. By 2025, Claude Code, Gemini Deep Research, and ChatGPT's Deep Research were all in production. The shift from AI as an answering tool to AI as an acting tool happened within three years of ChatGPT's launch. That's a genuinely compressed timeline.

No platform has a durable capability lead. The frontier shifts every few months. What's state-of-the-art at launch is competitive-but-not-leading within six months. The race is genuinely close, and the competitive dynamics between Google, OpenAI, and Anthropic aren't producing a stable winner in the near term.

For your AI search visibility, the practical takeaway is this: don't build a strategy that bets everything on one platform. The retrieval mechanisms are similar enough across platforms that content built to the universal requirements (crawlability, factual precision, clear structure, non-commodity value) performs well across all of them. Build for the architecture, not the brand name on the wrapper.

/ Next in the series

Article 5 explains exactly how AI Overviews and AI Mode work, and why your organic rankings don't guarantee AI search visibility. Read Article 5 →

SOURCES ↓
  1. OpenAI, "Introducing ChatGPT," 2022 — openai.com/index/chatgpt
  2. OpenAI, "GPT-4 Technical Report," 2023 — openai.com/index/gpt-4-research
  3. OpenAI, "Learning to Reason with LLMs (o1)," 2024 — openai.com/index/learning-to-reason-with-llms
  4. OpenAI, "Introducing ChatGPT Search," 2024 — openai.com/index/introducing-chatgpt-search
  5. OpenAI, "Introducing GPT-5.5," 2026 — openai.com/index/introducing-gpt-5-5
  6. OpenAI, Model Release Notes — help.openai.com/en/articles/9624314-model-release-notes
  7. OpenAI, ChatGPT Release Notes — help.openai.com/en/articles/6825453-chatgpt-release-notes
  8. Microsoft, "Reinventing Search," 2023 — blogs.microsoft.com/...
  9. Microsoft, "Announcing Microsoft Copilot," 2023 — blogs.microsoft.com/...
  10. Microsoft, "15 Milestones," 2025 — news.microsoft.com/...
  11. Google DeepMind, "Gemini," 2023 — deepmind.google/research/publications/gemini-...
  12. Google, Gemini API Changelog — ai.google.dev/gemini-api/docs/changelog
  13. Anthropic, "Introducing the Claude 3 Model Family," 2024 — anthropic.com/news/claude-3-family
  14. Anthropic, "Claude 2," 2023 — anthropic.com/news/claude-2
  15. Perplexity, Blog Hub — perplexity.ai/hub
  16. Perplexity, "Meet New Sonar" — perplexity.ai/hub/blog/meet-new-sonar
  17. Perplexity, "Introducing the Perplexity Publishers' Program," 2024 — perplexity.ai/hub/blog/introducing-the-perplexity-publishers-program
  18. Wikipedia: ChatGPT — en.wikipedia.org/wiki/ChatGPT
  19. Wikipedia: GPT-4 — en.wikipedia.org/wiki/GPT-4
  20. Wikipedia: Google Gemini — en.wikipedia.org/wiki/Google_Gemini
  21. Wikipedia: Gemini (language model) — en.wikipedia.org/wiki/Gemini_(language_model)
  22. Wikipedia: Claude (language model) — en.wikipedia.org/wiki/Claude_(language_model)
  23. Wikipedia: Microsoft Copilot — en.wikipedia.org/wiki/Microsoft_Copilot
  24. Wikipedia: GitHub Copilot — en.wikipedia.org/wiki/GitHub_Copilot
  25. Wikipedia: Perplexity AI — en.wikipedia.org/wiki/Perplexity_AI
  26. Wikipedia: Anthropic — en.wikipedia.org/wiki/Anthropic
How AI Search Works

The full series.

1
Available now
Attention Is All You Need: The Google Paper That Accidentally Ended Google's Search Monopoly
2
Available now
How Large Language Models Actually Work: Tokens, Context Windows, and Why Your Content Gets Ignored
3
Available now
What Is RAG: the Only Thing That Matters for AI Search
4
You are here
The AI Search Race: ChatGPT, Gemini, Claude, Copilot, and Perplexity, A Technical Timeline
5
Available now
How Google AI Overviews and AI Mode Actually Work and Why Your SEO Rankings Don't Guarantee AI Visibility
6
Available now
How to Be Surfaced in AI Search: What the Providers' Own Documentation Actually Says
tc
/ Written by

Thomas Cox

Twelve years in B2B SEO, most recently at VP level. Now independent — helping companies stay discoverable as buyer search moves into ChatGPT, Perplexity, and Google AI Overviews. Remote · UK.

About Thomas → Work with me →

Want this in your inbox?

Roughly one essay a month. No drip sequences, no upsells.

Subscribe →