Public record · MIT licensed
The SERP event registry.
A dated, sourced, confidence-tagged public record of AI model releases, search feature changes, confirmed algorithm updates, and observed SERP volatility, from November 2022 onward.
Coverage runs across the AI answer surfaces, including ChatGPT, Perplexity, the Gemini app, AI Overviews and AI Mode, alongside the classic Google SERP and Discover. The repository is the source of truth; this page renders it. It exists so that movement in your own data can be checked against what changed on the surfaces, and so that pipelines can make that check automatically.
Use
What the record is for.
When something moves in your rankings, your traffic, or your citations, the registry answers what changed on the surfaces around that date.
Event windows for your own analysis
Join your data against dated windows rather than against trends. The record carries its own demonstration of why that is harder than it looks: the December 2025 Google core update and the GPT-5.2 release share a start date, and both rows say so, because volatility in that window cannot be assigned to either event without separating the surfaces first.
Deploy and launch timing
Rows are dated to the day and name the actor and the surface. Where a vendor published a completion date, the row records it: Google’s dashboard put the December 2025 core update at complete on 29 December. Scheduling large content deploys and programmatic launches around what is mid-rollout is a use the record supports.
Machine consumption
data/events.jsonl is newline-delimited JSON at a fixed raw URL, checked against its published schema by CI on every pull request and every push to main. Each row has an anchor on this page. Pipelines and agent workflows can read the file directly: a deploy gate that checks whether a core update is mid-rollout before shipping SEO-sensitive changes, dashboard annotations generated from the file, a step that consults the record before a large publishing run.
Event studies
Dated discontinuities support designs that trend correlation cannot, and the attribution tier on each row states how much weight its date can carry. The ladder below defines the four tiers.
RampStack’s own content and launch pipelines consult this record when scheduling programmatic work.
The instrument
The attribution ladder.
Every row carries one of four tiers, and the tier describes the event and its date together. A well known event with a poorly attested date does not get the top tier. Rows move down the ladder as readily as up, and corroboration from a volatility sensor is recorded on the row without moving the tier.
- 1
Vendor confirmed
vendor_confirmed
The vendor stated the event and the date on its own property, and the date can be read off that property today.
- 2
Vendor announced
vendor_announced
The vendor announced the event, but the rollout window is uncertain, or the date was taken from contemporaneous reporting rather than a vendor dateline.
- 3
Community inferred
community_inferred
No vendor statement exists, and multiple independent observers converge on the event and its approximate date.
- 4
Hypothesis
hypothesis
We suspect the event happened and the row says so. It stays at this tier until a vendor statement or independent convergence moves it.
The record
Every event in the registry, newest first.
All 45 rows are on this page. Open a row for the full description and the primary sources it rests on.
| Date | Type | Surfaces | Event | Attribution |
|---|---|---|---|---|
| Model release |
| OpenAI releases GPT-6.1 Sol in ChatGPT Work and Codex DetailHide detailOpenAI released GPT-6.1 Sol, an upgrade to GPT-6 Sol. The vendor post states it is available starting today to all Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex, that GPT-6.1 Sol is not yet available in Chat, and that developers can access it as gpt-6.1-sol, so standard ChatGPT chat answers did not change on this date. The dated release note gives a narrower rollout, starting with Pro users and expanding to Plus, Business, Enterprise, and Edu, so the date a given non Pro subscriber received it is not recorded. The date is read from the dated entry in the vendor release notes; the announcement page carries no publish dateline.
| Vendor confirmed | |
| Model release |
| Anthropic releases Claude Sonnet 5.5 DetailHide detailAnthropic released Claude Sonnet 5.5, which the vendor describes as the second model in the Claude 5.5 family and a faster, lower-cost complement to Claude Opus 5.5, and states is now available on all platforms. The post does not state whether Sonnet 5.5 replaced Sonnet 5 as the default model on the Free and Pro plans, where Sonnet 5 was the stated default, so the date the Claude.ai default changed is not recorded. It arrived six days after Opus 5.5, so the two releases sit close enough that they should not be read as separate shifts in Claude.ai answers. The vendor page carries the announcement date.
| Vendor confirmed | |
| Spam update |
| Google September 2026 spam update DetailHide detailGoogle released the September 2026 spam update, logged on the Search Status Dashboard as applying globally and to all languages, with a start of 24 September 2026 at 09:15 US/Pacific. Google states the rollout may take up to two weeks to complete, the longest stated window of the four confirmed Google spam updates of 2026, after March, June, and August. As of 1 October 2026 the dashboard records no completion, so this row asserts none; a completion timestamp goes in this description when Google posts one.
| Vendor confirmed | |
| Model release |
| OpenAI releases GPT-6 Sol and GPT-6 Luna in ChatGPT Work and Codex DetailHide detailOpenAI released GPT-6 Sol and GPT-6 Luna. The vendor post states that GPT-6 Sol and GPT-6 Luna are available in ChatGPT Work and Codex starting today for all Plus, Pro, Business, Enterprise, and Edu users, that Free and Go users can access GPT-6 Luna in the desktop app, and that these models are not yet available in Chat; the API ids are gpt-6-sol and gpt-6-luna. Standard ChatGPT chat answers therefore did not change on this date, and a study joining on this row should not treat it as a change to the Chat default. The post says the rollout in ChatGPT was gradual throughout the day. The date is read from the dated entry in the vendor release notes; the announcement page carries no publish dateline.
| Vendor confirmed | |
| Model release |
| Anthropic releases Claude Opus 5.5 DetailHide detailAnthropic released Claude Opus 5.5, which the vendor describes as the first model in our new Claude 5.5 family and states is now available on all platforms. The post raises five-hour usage limits on Pro, Max, Team, and seat-based Enterprise plans alongside the release, but does not state whether Opus 5.5 replaced Opus 5 as the default model on any Claude plan, so the date the default changed for a given subscriber is not recorded. It ships with safeguards similar to those on Claude Fable 5.1, so some cybersecurity and biology requests may be handled differently than under Opus 5. The vendor page carries the announcement date.
| Vendor confirmed | |
| Search feature change |
| OpenAI retires automatic switching to Thinking for ChatGPT Plus and Pro DetailHide detailOpenAI changed how ChatGPT answers for paid users without changing the model. The dated vendor release note says the vendor is retiring automatic switching from Instant to Thinking (reasoning) for ChatGPT Plus and Pro users globally, that you can still select an available option in the model picker to give ChatGPT more time to think or reason, and that it is also removing the Higher intelligence setting from ChatGPT on the web for these plans. The note adds that ChatGPT can still switch automatically for safety purposes, so not every Plus or Pro answer after this date comes from Instant. Free and Go users and managed workspaces are not named in the note, so this row asserts no change for them. The date is read from the dated entry in the vendor release notes.
| Vendor confirmed | |
| Model release |
| OpenAI releases GPT-6 Astra DetailHide detailOpenAI released GPT-6 Astra in a staged rollout. The vendor post says GPT-6 Astra is rolling out today to a limited set of organizations and over the coming days will become available to all ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API, Microsoft Azure, and AWS Bedrock, and the dated release note adds: Astra is not yet generally available. Broader availability is planned over the coming days. No dated vendor entry records when paid plans received it, so the date ChatGPT answers changed for a given subscriber is not recorded here. An undated vendor help page states that GPT-6 Pro, powered by GPT-6 Astra, is available in ChatGPT for Pro $100, Pro $200, Business and Enterprise plans, and that Plus plans include GPT-6 Astra in ChatGPT Work and Codex, so Plus subscribers did not receive it in Chat. The date is read from the dated entry in the vendor release notes; the announcement page carries no publish dateline.
| Vendor confirmed | |
| Model release |
| Google releases Gemini 3.8 Flash DetailHide detailGoogle released Gemini 3.8 Flash, describing it as our most intelligent workhorse model and its third Flash release in only six weeks. The vendor post states that 3.8 Flash is available to Google AI Pro and Ultra subscribers across the Gemini app, AI Mode in Google Search and Gemini in Google Sheets, so it reached two consumer answer surfaces on the announcement date for subscribers only, and the post does not say it replaced the default model for other users. The same post announced Gemini 3.8 Flash Cyber, available to trusted defenders through the Fairwind Program, which does not meet the general audience test and is recorded here as context only. The vendor page carries the announcement date.
| Vendor confirmed | |
| Model release |
| Anthropic releases Claude Fable 5.1 and Claude Mythos 5.1 DetailHide detailAnthropic released Claude Fable 5.1 and Claude Mythos 5.1, which the vendor post describes as the same model with different levels of safeguards: Fable 5.1 is generally available, while Mythos 5.1 is available only through our trusted access programs. The post states Claude Fable 5.1 is available today on all platforms and that it defaults to Medium effort on Claude.ai, so it reached a consumer answer surface on the release date. Mythos 5.1 is available to vetted cyberdefenders and life scientists and, per the post, currently only to a set of US organizations, so it does not meet the general audience test and is named here as context only. The vendor post carries a month but no day; the 1 September date is read from the dated entry in the vendor release notes.
| Vendor confirmed | |
| Spam update |
| Google August 2026 spam update DetailHide detailGoogle released the August 2026 spam update, logged on the Search Status Dashboard as applying globally and to all languages. The dashboard records a start of 18 August 2026 at 09:27 US/Pacific and completion on 21 August 2026 at 01:49 US/Pacific, a rollout of two days and sixteen hours. It is the third confirmed Google spam update of 2026, after the March and June updates, and the longest of the three.
| Vendor confirmed | |
| Model release |
| Google releases Gemini 3.7 Flash DetailHide detailGoogle released Gemini 3.7 Flash, describing it as its most intelligent workhorse model yet for coding and agents. The vendor post states consumer access as available via Spark, the personal agent in the Gemini app, for Google AI Pro and Ultra subscribers in supported countries, with developer access through Google AI Studio, Android Studio, Google Antigravity, Vertex AI, and the Gemini Enterprise Agent Platform. The supported countries are not enumerated, so coverage on the announcement date cannot be read from the source. It arrived three weeks after Gemini 3.6 Flash, which is short enough that a study joining on model version should not assume the two rows describe separate populations of users.
| Vendor confirmed | |
| Search feature change |
| OpenAI makes GPT-5.6 Luna the ChatGPT default for free users DetailHide detailOpenAI changed the default model behind the ChatGPT free tier to GPT-5.6 Luna. The dated vendor release note states that GPT-5.6 Luna will become the default model for Free and Go users this week, and that starting next week they'll also have unlimited text chats and access to a new Think button for harder questions, so a single date does not describe when free tier answers changed. The same note says Plus and Pro users can now use an updated GPT-5.6 Sol in ChatGPT, so two consumer tiers changed under one date, and that the release applies to the chat experience while Work and Codex are not changing. The date is read from the dated entry in the vendor release notes, which replaces the contemporaneous reporting this row first rested on when openai.com could not be read.
| Vendor confirmed | |
| Model release |
| Anthropic releases Claude Opus 5 DetailHide detailAnthropic released Claude Opus 5, stating that it is available today on all platforms, that it is the new default model on Claude Max, and that it is the strongest model on Claude Pro. The vendor post names Claude.ai, Claude Code, and Claude Cowork among the surfaces it reaches, so the model became the default behind a consumer answer surface on the announcement date. When this row was added it was the first on the claude surface; the June 2026 Fable 5 release, suspension, and restoration and the Sonnet 5 release were backfilled on 1 October 2026, and earlier Claude events remain uncovered. The vendor page carries the announcement date.
| Vendor confirmed | |
| Model release |
| Google releases Gemini 3.6 Flash DetailHide detailGoogle released Gemini 3.6 Flash, which the vendor post makes available for everyone via the Gemini app, alongside developer access through the Gemini API, Google AI Studio, Android Studio, Google Antigravity, and the Gemini Enterprise Agent Platform. It reached a consumer answer surface on the announcement date rather than after a preview period. The same post announced Gemini 3.5 Flash-Lite and Gemini 3.5 Flash Cyber, which are recorded here as context only: Flash Cyber is described as limited government and partner access and does not meet the general audience test. The vendor page carries the announcement date.
| Vendor confirmed | |
| Search feature change |
| Google adds connected apps to AI Mode in Search DetailHide detailGoogle announced connected apps in AI Mode in Search, letting a searcher link a third party service and complete a task inside the answer rather than leaving for the app. The vendor post names Instacart, Canva, and YouTube Music, and says Google is working with a range of partners and looking forward to launching with more apps soon. It gives the rollout as starting to roll out this week in the U.S., so the announcement date is firm while the date any given searcher saw the feature is not. No market outside the United States is stated on the vendor page.
| Vendor announced | |
| Model release |
| OpenAI opens general availability of GPT-5.6 DetailHide detailOpenAI opened general availability of GPT-5.6 across its Sol, Terra, and Luna variants. A restricted preview had gone to roughly twenty partner organizations on 26 June 2026 while a United States government safety review was outstanding, so the model was already serving production traffic about two weeks before general access. The date recorded here comes from contemporaneous reporting rather than a dateline on the vendor page.
| Vendor announced | |
| Search feature change |
| Anthropic restores access to Claude Fable 5 DetailHide detailAnthropic restored general access to Claude Fable 5. The vendor post dated 30 June states that the export controls on Fable 5 and Mythos 5 had been lifted that day and that Fable 5 would be available starting Wednesday, July 1, to users globally on the Claude Platform, Claude.ai, Claude Code, and Claude Cowork; the vendor release notes record the restoration on 1 July. The post says Fable 5 was included for up to 50% of weekly usage limits on Pro, Max, Team, and select Enterprise plans through 7 July and available via usage credits after that, so subscriber exposure narrowed a week after restoration, and requests blocked by a new safety classifier are sent to Claude Opus 4.8. Mythos 5 access was restored for a set of US organizations following a US government approval on 26 June; that step does not meet the general audience test and has no row of its own.
| Vendor confirmed | |
| Model release |
| Anthropic releases Claude Sonnet 5 DetailHide detailAnthropic released Claude Sonnet 5, describing it as the most agentic Sonnet model yet. The vendor post states that from today, Claude Sonnet 5 is available across all plans: it is the default model for Free and Pro plans, and is available to Max, Team, and Enterprise users, so the default model behind Claude.ai changed for most users on the release date. It launched while access to Claude Fable 5 was suspended, one day before Fable 5 was restored, so the two changes on the claude surface fall in the same week. The vendor page carries the announcement date.
| Vendor confirmed | |
| Spam update |
| Google June 2026 spam update DetailHide detailGoogle released the June 2026 spam update, logged on the Search Status Dashboard with no country or language carve out. The dashboard recorded completion on 26 June 2026 at 10:00 US/Pacific. This is the most recent confirmed Google ranking update in the registry as of the seed date.
| Vendor confirmed | |
| Search feature change |
| Anthropic suspends access to Claude Fable 5 and Claude Mythos 5 DetailHide detailAnthropic suspended access to Claude Fable 5 and Claude Mythos 5 for all users three days after launch. The vendor statement says the US government, citing national security authorities, issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, and that the net effect was to disable both models for all customers; it says access to all other Anthropic models was not affected. The vendor sources name the US government and an export control directive but not the issuing agency, so this row does not name one either. The row is typed search_feature_change because the event changed which model answers on the claude surface rather than releasing one, the same typing the registry uses for a default model swap.
| Vendor confirmed | |
| Model release |
| Anthropic releases Claude Fable 5 and Claude Mythos 5 DetailHide detailAnthropic launched Claude Fable 5, which the vendor post describes as a Mythos-class model that we've made safe for general use, and states that Claude Fable 5 is available everywhere today. It is the first Mythos-class model released to a general audience: the same post records that Claude Mythos Preview, the first Mythos-class model, went in April only to a limited group of cyber defenders and critical software infrastructure providers. Claude Mythos 5, the same underlying model with some safeguards lifted, was restricted to Glasswing partners and does not meet the general audience test, so it is named here as context only. The post states Fable 5 was included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost through 22 June, and that queries on some topics receive a response from Claude Opus 4.8 instead, so not every Claude.ai answer in this window came from Fable 5.
| Vendor confirmed | |
| Core update |
| Google May 2026 core update DetailHide detailGoogle released the May 2026 core update, logged on the Search Status Dashboard with no country or language carve out. Google estimated a rollout of up to two weeks, and the dashboard recorded completion on 2 June 2026 at 05:40 US/Pacific. It began two days after the I/O 2026 announcement that AI Mode becomes the global Search default, so ranking movement and surface composition were both in motion across the same window.
| Vendor confirmed | |
| Search feature change |
| Google makes AI Mode the global Search default at I/O 2026 DetailHide detailAt Google I/O 2026, Google announced that AI Mode becomes the default Search experience worldwide, with Gemini 3.5 Flash serving it, alongside a redesigned query box and background agents that keep researching between sessions. Google placed those agents with Google AI Pro and Ultra subscribers in the United States over the months following, so the announcement date and the date any given user met the new default are not the same. This row records the announcement date and asserts no completion date.
| Vendor announced | |
| Core update |
| Google March 2026 core update DetailHide detailGoogle released the March 2026 core update, logged on the Search Status Dashboard with no country or language carve out. Google estimated a rollout of up to two weeks, and the dashboard recorded completion on 8 April 2026 at 06:00 US/Pacific. The March 2026 spam update had finished only two days before this rollout began.
| Vendor confirmed | |
| Spam update |
| Google March 2026 spam update DetailHide detailGoogle released the March 2026 spam update, logged on the Search Status Dashboard with no country or language carve out. The dashboard recorded completion on 25 March 2026 at 07:30 US/Pacific, under a day from start to finish. It began three days before the March 2026 core update, so the two rollouts sit close enough together that they should not be read as one event.
| Vendor confirmed | |
| Model release |
| OpenAI releases GPT-5.4 with native computer use DetailHide detailOpenAI released GPT-5.4 across ChatGPT, the API, and Codex, with the model able to drive software directly and to hold roughly a million tokens of context. A smaller GPT-5.4 variant followed on 17 March 2026 and reached free tier ChatGPT users, so the model behind free tier answers changed on a later date than the flagship release did. The date recorded here comes from contemporaneous reporting rather than a dateline on the vendor page.
| Vendor announced | |
| Core update |
| Google February 2026 Discover core update DetailHide detailGoogle released what the Search Status Dashboard names the February 2026 Discover core update, scoped at release to English language users in the United States, with Google saying it would look at widening that later. The dashboard recorded a start of 5 February 2026 at 09:00 US/Pacific and completion on 27 February 2026 at 02:00 US/Pacific, so the rollout ran well past the two weeks Google estimated. It is typed core_update because that is the vendor's own name for it, but it touched Discover rather than web Search and only one country and language, so pooling it with the web core updates in this file would be an error.
| Vendor confirmed | |
| Model release |
| OpenAI releases GPT-5.2 DetailHide detailOpenAI released GPT-5.2, starting the ChatGPT rollout with paid plans while making it available in the API straight away. The release shares a start date with the December 2025 Google core update, so volatility seen in this window cannot be assigned to either event without separating the surfaces first. The date recorded here comes from contemporaneous reporting rather than a dateline on the vendor page.
| Vendor announced | |
| Core update |
| Google December 2025 core update DetailHide detailGoogle released the December 2025 core update, logged on the Search Status Dashboard with no country or language carve out. Google estimated a rollout of up to three weeks, and the dashboard recorded completion on 29 December 2025 at 11:00 US/Pacific. This rollout shares a start date with the GPT-5.2 release, which any event study over this window should treat as a confound rather than a coincidence to wave through.
| Vendor confirmed | |
| Model release |
| Google releases Gemini 3 and deploys it into Search DetailHide detailGoogle released Gemini 3 and shipped it into Search and the Gemini app on the announcement date rather than staging it behind a preview period. A frontier model landing in Search the same day it is announced is the case this registry exists to capture, because a model event and a search side answer change share one timestamp and cannot be told apart by date alone. The vendor page carries the announcement date.
| Vendor confirmed | |
| Search feature change |
| Perplexity makes Comet free worldwide DetailHide detailPerplexity dropped the subscription and waitlist gate on Comet and released it free in every market. What changed is the addressable population of the surface rather than how it answers, which is why this is a separate row from the limited launch. The date recorded here comes from contemporaneous reporting rather than a dateline on the vendor page.
| Vendor announced | |
| Spam update |
| Google August 2025 spam update DetailHide detailGoogle released the August 2025 spam update, logged on the Search Status Dashboard with no country or language carve out. The dashboard recorded completion on 22 September 2025 at 00:00 US/Pacific, putting the rollout at roughly four weeks. Google issued no separate guidance alongside it.
| Vendor confirmed | |
| Model release |
| OpenAI releases GPT-5 and makes it the ChatGPT default DetailHide detailOpenAI released GPT-5 and made it the ChatGPT default, retiring the manual model picker in favor of routing each request automatically between quicker and more deliberative variants. Which model answers a given prompt therefore changed without any user action, which matters to any measurement series that assumed a fixed model sat behind the surface. The date recorded here comes from contemporaneous reporting rather than a dateline on the vendor page.
| Vendor announced | |
| Search feature change |
| Perplexity launches Comet browser to Max subscribers and waitlist DetailHide detailPerplexity launched Comet, a browser that carries its assistant alongside whatever page is open rather than in a separate destination. Distribution was invitation gated at launch, reaching Perplexity Max subscribers and selected waitlist entrants, so the population able to reach the surface grew on a schedule Perplexity did not publish. The date recorded here comes from contemporaneous reporting rather than a dateline on the vendor page.
| Vendor announced | |
| Core update |
| Google June 2025 core update DetailHide detailGoogle released the June 2025 core update, logged on the Search Status Dashboard with no country or language carve out. Google estimated a rollout of up to three weeks, and the dashboard recorded completion on 17 July 2025 at 01:00 US/Pacific. Both the start and completion timestamps are vendor published.
| Vendor confirmed | |
| Search feature change |
| Google makes AI Mode available to all United States Search users DetailHide detailAt Google I/O 2025, Google announced that AI Mode would reach every Search user in the United States, taking it out of Search Labs, with a custom build of Gemini 2.5 serving it. Google said the tab would show up on US searches during that week, so per user availability trailed the announcement by an interval nobody published. This row records the announcement date.
| Vendor announced | |
| Model release |
| Google releases Gemini 2.5 Pro Experimental DetailHide detailGoogle released Gemini 2.5 Pro Experimental, the first model in the Gemini 2.5 family, which Google characterizes as one that works through a problem before it answers. It reached the Gemini consumer app for Advanced subscribers at announcement and later became the model serving AI Mode in Search. The vendor page carries the announcement date.
| Vendor confirmed | |
| Core update |
| Google March 2025 core update DetailHide detailGoogle released the March 2025 core update, logged on the Search Status Dashboard with no country or language carve out. Google estimated a rollout of up to two weeks, and the dashboard recorded completion on 27 March 2025 at 05:35 US/Pacific. Both the start and completion timestamps are vendor published.
| Vendor confirmed | |
| Search feature change |
| Google introduces AI Mode as a Search Labs experiment DetailHide detailGoogle introduced AI Mode, a separate tab in Search that answers conversationally and takes follow up questions, as a Search Labs experiment limited to Google One AI Premium subscribers in the United States. The same announcement put Gemini 2.0 behind AI Overviews for harder query classes. The vendor page carries the announcement date and Labs availability opened the same day.
| Vendor confirmed | |
| Search feature change |
| OpenAI launches ChatGPT search with real time web retrieval DetailHide detailOpenAI launched ChatGPT search, letting the assistant pull current pages while composing an answer and attribute the pages it drew on. Access was restricted to Plus and Team subscribers at launch and was widened to all logged in users in February 2025. This row records the initial launch; the later widening is not separately seeded.
| Vendor announced | |
| Search feature change |
| Google expands AI Overviews to more than 100 additional countries DetailHide detailGoogle widened AI Overviews to over 100 further countries and territories and to several more languages, putting the surface at more than a billion monthly users by its own count. The expansion was framed as beginning on this date rather than finishing on it, and the individual countries were never enumerated. Any panel that treats a given market as AI Overviews enabled from this date is assuming something the source does not support.
| Vendor announced | |
| Search feature change |
| Google launches AI Overviews for all United States searchers DetailHide detailAt Google I/O 2024, Google announced that AI Overviews would go to every Search user in the United States, retiring the Search Generative Experience label in the process. Google put the reach at hundreds of millions of US searchers within the following week, so the announcement date and the date a given query started returning an AI Overview are different dates. This row records the announcement date and asserts no completion date.
| Vendor announced | |
| Model release |
| OpenAI releases GPT-4o as the ChatGPT default model DetailHide detailOpenAI announced GPT-4o and made it the model serving ChatGPT for both free and paid tiers. The capability set arrived in stages rather than all at once, with text and image handling ahead of voice and full API access, so the interval over which ChatGPT answers actually changed is wider than this single date. The date recorded here comes from contemporaneous reporting rather than a dateline on the vendor page.
| Vendor announced | |
| Search feature change |
| Google announces Search Generative Experience at I/O 2023 DetailHide detailAt Google I/O 2023, Google announced the Search Generative Experience, which placed a generated answer block ahead of the ranked result list for a subset of queries. Access sat behind a Search Labs waitlist at announcement, so the population able to see it grew over an interval Google never bounded publicly. SGE is the direct ancestor of AI Overviews and is seeded so the AI Overviews rows have somewhere to point back to.
| Vendor announced | |
| Model release |
| OpenAI releases ChatGPT as a public research preview DetailHide detailOpenAI opened ChatGPT to anyone as a free research preview, running on a model in the GPT-3.5 series. This row is the registry origin point, marking the first date on which a general audience could get conversational answers to information questions without going to a search engine. The vendor post confirms the release, and the date recorded here comes from contemporaneous reporting rather than a dateline on the vendor page, so the row is capped at vendor_announced.
| Vendor announced |
The row contract is defined in schema/event.schema.json and enforced in CI on every pull request.
Known limits
What this record cannot see.
The registry is a lower bound on how often these surfaces changed. Each limit below is stated on the repository as well, in METHODOLOGY.md.
The registry does not claim completeness
Vendors change the model behind a consumer surface without announcing it, and routing layers shift traffic between variants continuously. Nothing here detects that, so silent model swaps are unobservable and absence of a row is not absence of an event.
Announcement dates are not exposure dates
Most rows anchor to an announcement. The date a population actually saw a change is later by an interval vendors rarely publish, and no row asserts a completion date it cannot source.
Geographic and language coverage is uneven
Vendor announcements about multi country rollouts usually do not enumerate countries. Rows record what the vendor said, so a market cannot be assumed covered on the announcement date.
The registry does not claim causation
It records events and observed volatility with stated confidence. Overlapping events stay separate rows even when the data cannot separate them, and every causal reading belongs to whoever is doing the study.
Citing and contributing
How to cite
Cite the repository and the commit you read.
The file changes as events are added and as attribution tiers are revised, so a citation without a commit points at a moving target.
serp-event-registry, rampstackco. https://github.com/rampstackco/serp-event-registry
Retrieved at commit <sha>.Released under the MIT license. The citation block on the repository is the same one.
How to propose an event
Open a pull request with primary sources.
Add one line to data/events.jsonl, keeping the file sorted by event_date, and include at least one primary source: a vendor blog post, a status page, or an official changelog. Secondary reporting can support a row without carrying it.
Proposals without sources are closed. Where the date is softer than the row claims, the ask is to lower date_precision or attribution rather than to drop the row.