AI search model updates: Live release log for organic search, AEO & SEO teams
- AI engines changed how they retrieve, display, and attribute sources at least 10 times in 2026, and each change can move your citations within days.
- Google moved inline links higher in AI Mode and AI Overviews on October 15, 2026, which raises the value of pages that answer one task cleanly.
- ChatGPT moved to GPT-6 with Intelligent UI on October 7, 2026, and a link-tagging bug hid ChatGPT referrals in GA4 from October 8 to October 16.
- Perplexity and Anthropic rebuilt retrieval around code that filters results before the model reads them, so dense, well-structured pages carry more weight.
- AirOps, the best AEO platform for enterprise companies, logs a model or feature here only after the lab documents it, so your roadmap tracks shipped behavior.
Latest retrieval and model shifts at a glance
| Date (2026) | Engine | What changed | Enterprise priority |
|---|---|---|---|
| Nov 10 | Gemini and Google AI Mode | Gemini 4 Carbon becomes generally available, including AI Mode for Google AI Ultra | Monitor |
| Nov 9 | Claude | Claude Fable 5.2 and Mythos 5.2 launch; no stated change to web search | Monitor |
| Nov 5 | Perplexity API | Agent API effort levels cost less; a new ultra level and 4K search-result tokens launch | Log only |
| Nov 4 | Perplexity API | Decisions API can ground answers in live web search | Log only |
| Oct 28 | Google AI Mode | Ads expand to 10 more countries; multiturn ads launch in the US | Monitor |
| Oct 15 | Google AI Mode and AI Overviews | Inline links sit higher in responses; official-site section in testing | Act now |
| Oct 8 to Oct 16 | ChatGPT | Links lost utm_source=chatgpt.com, so referrals read as Direct |
Act now |
| Oct 7 | ChatGPT | GPT-6 and Intelligent UI become the default chat experience | Act now |
| Sep 24 | Perplexity API | Fast Search launches on the Photon retrieval service | Log only |
| Sep 22 | Claude | Claude Opus 5.5 launches; no stated change to web search | Monitor |
| Jun 1 | Perplexity | Search as Code becomes the default search architecture in Computer | Monitor |
| May 19 | Google AI Mode | Gemini 3.5 Flash becomes the default model; AI Overviews and AI Mode join into one flow | Act now |
| Feb 17 | Claude API | Web search gains dynamic filtering before results enter context | Monitor |
What enterprise SEO and content teams should watch
A model update matters to your brand when it changes one of three things: which sources the engine retrieves, how it displays links, or whether your analytics can see the visit. Release notes rarely say this in plain terms, so read each update against those questions.
Retrieval. Perplexity's Search as Code and Claude's dynamic filtering both let the model write code that trims results before it reads them. Pages with a clear answer near the top of each section give that filter something to keep.
Link display. Google's October 15 update puts inline links higher in AI Mode and AI Overviews. A page that owns one task, such as a pricing page or a setup guide, has a better shot at the new official-site section.
Attribution. The October 2026 ChatGPT tagging gap showed how fast AI referral data can break. Pair GA4 referral reports with citation tracking so one broken tag does not rewrite your quarter. Our guide to how to measure AI search visibility covers the full measurement setup.
When an update lands, check movement before you change content. Brand presence in AI answers shifts daily even without a model change, as we cover in AI search volatility. Give each release a 7-day read on citations by engine, then decide.
Verified release log
November 10, 2026: Gemini 4 Carbon becomes generally available
What changed. Google made Gemini 4 Carbon available to developers, enterprises, scientists, and consumers. API access is in Google AI Studio and Gemini for business for paying customers, at the same introductory price as Gemini 4 Argon ($2 per million input tokens and $10 per million output tokens; $4 / $20 after November 30, 2026). It is rolling out on November 10, 2026 to all Google AI Ultra subscribers in Google Antigravity, the Gemini app, and AI Mode in Search. Everyday users can try it in the Gemini app and AI Mode only with Google AI Ultra. The announcement does not document a change in how AI Mode or AI Overviews retrieve, rank, or display citations. Argon remains the stated recommendation for most software engineering and knowledge work. Source: Google
Action for enterprise teams. No content change is required. Carbon is available only with Google AI Ultra in the Gemini app and AI Mode, and it is not the default model. Watch AI Mode and Gemini citations for your priority prompts over the next 7 days and log any movement against this date.
November 9, 2026: Anthropic launches Claude Fable 5.2 and Mythos 5.2
What changed. Fable 5.2 is generally available (claude-fable-5-2). Mythos 5.2 is limited to vetted Life Sciences and Cyber Verification Program organizations. Both have a 1M-token context. Fable pricing is $8 per million input tokens and $40 per million output tokens, with optional fast mode at $16 / $80. Anthropic says Fable can be up to 30% faster than 5.1. The announcement does not describe a change in how Claude searches the web or picks citations. Source: Anthropic
Action for enterprise teams. No content change is required. Watch Claude citations for your priority prompts over the next 7 days and log any movement against this date.
November 5, 2026: Perplexity improves Agent API effort levels
What changed. Every Agent API effort level costs less. A new ultra level (ultra@2026-11-05.0) runs on anthropic/claude-opus-5-5, can run in background mode, and uses three parallel research agents plus a merge agent. Perplexity reports medium effort rose from 55% to 77% on BrowseComp/DSQA at almost 90% lower cost. Each search returns up to 4K tokens of results, up from 2K, with a new snippet engine. Fast Search is $1.00 per 1,000 requests. Source: Perplexity
Action for enterprise teams. This is a developer API, so it does not change consumer Perplexity answers today. Log it and watch for the same retrieval pattern in Perplexity's consumer products.
November 4, 2026: Perplexity Decisions API adds web grounding
What changed. Perplexity's Decisions API can now run a web search before it answers. Adding "mode": "web_grounding" to a request runs one Perplexity Fast Search query per question and appends each result's title, URL, and page text to the model's evidence. A request can also point the model at one URL so it answers from that page. Perplexity reports accuracy on its PWS-Decisions benchmark rose from 51% to 86%. Source: Perplexity
Action for enterprise teams. This is a developer API, so it does not change consumer Perplexity answers today. Log it and watch for the same grounding pattern in Perplexity's consumer products.
October 28, 2026: Google expands ads in AI Mode
What changed. Google scaled ads in AI Mode to Australia, Canada, India, Kenya, Malaysia, New Zealand, Nigeria, Pakistan, the Philippines, and Singapore, and launched multiturn ads in the US so ads can appear in follow-up queries. Ads carry a "Sponsored" label. Google says AI Mode queries are, on average, three times longer than traditional searches, and plans Brazil, Mexico, and the UK for Q4 2026. Source: Google
Action for enterprise teams. Paid placements now compete with organic citations inside follow-up answers. Review the follow-up prompts in your category and confirm your pages answer the second and third question a buyer asks.
October 15, 2026: Google moves inline links higher in AI Mode and AI Overviews
What changed. Google said inline links will appear "higher up in responses more frequently." It is testing a dedicated section at the top of responses for the one or two official sites that complete a task, with context about why each link is useful. It also expanded timely-article carousels with Preferred Sources to more locations and languages, and gave end-of-response links image thumbnails. Source: Google
Action for enterprise teams. Audit the pages that answer a single task for your buyers, such as pricing, integrations, and setup docs. Make the answer explicit in the first lines of the page and keep titles specific, so Google can match the page to the official-site slot.
October 8 to October 16, 2026: ChatGPT referrals lost their tracking tag
What changed. After GPT-6 rolled out, links in ChatGPT answers opened without utm_source=chatgpt.com and without a referrer, so analytics tools logged many ChatGPT visits as Direct. Search Engine Roundtable reported the issue on October 14 and updated the story on October 15 to say UTM tags were returning and the issue was a bug. OpenAI's release notes do not document the incident. Source: Search Engine Roundtable
Action for enterprise teams. Annotate October 8 to October 16, 2026 in GA4 and in any board report that shows ChatGPT referrals. Compare that window against citation and mention trends before you draw conclusions about demand.
October 7, 2026: OpenAI ships GPT-6 with Intelligent UI
What changed. OpenAI introduced GPT-6 with Intelligent UI in ChatGPT. Answers can combine text, visuals, and interactive elements, and GPT-6 can start answering while it continues to think or use tools. Plus, Pro, Business, and Enterprise use GPT-6 Sol; Free and Go use GPT-6 Luna from October 8. The Pro reasoning option stays on GPT-6 Astra. Source: OpenAI
Action for enterprise teams. Answers now include comparisons and diagrams built inside ChatGPT. Publish structured comparison data, such as tables with named criteria, that the model can pull into those formats. Track ChatGPT citations by prompt for 7 days after a release like this one, using a setup like our LLM brand citation tracking guide.
September 24, 2026: Perplexity launches Fast Search on Photon
What changed. Perplexity added Fast Search to its Search API, running on Photon, a new Rust-based retrieval and ranking service. Perplexity states Fast Search returns 95% of results in 230 ms or less. Source: Perplexity
Action for enterprise teams. Fast retrieval for developers means more third-party agents will query the web through Perplexity. Keep key facts in crawlable HTML so those agents can read them.
September 22, 2026: Anthropic launches Claude Opus 5.5
What changed. Anthropic introduced Claude Opus 5.5, the first model in the Claude 5.5 family. Anthropic says it performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5. Claude Sonnet 5.5 followed on September 28, 2026. The announcement does not describe changes to how Claude searches the web or picks citations. Source: Anthropic
Action for enterprise teams. No content change is required. Watch Claude citations for your priority prompts over the next 7 days and log any movement against this date.
June 1, 2026: Perplexity introduces Search as Code
What changed. Perplexity introduced Search as Code, which exposes retrieval, ranking, filtering, and fan-out steps as SDK building blocks. The model writes and runs code to build a retrieval pipeline for each request. It is available in the Agent API and is the default search architecture in Perplexity Computer. Source: Perplexity
Action for enterprise teams. Agents can now run hundreds of retrieval steps on one task. Pages with clear headings, short answer paragraphs, and labeled data are easier for that code to filter and keep. Our GEO guide covers the structural basics.
May 19, 2026: Gemini 3.5 Flash becomes the default in AI Mode
What changed. At Google I/O 2026, Gemini 3.5 Flash became the default model for AI Mode globally and for the Gemini app. Google also brought AI Overviews and AI Mode into one AI Search experience, so a user can move from a results page with an AI Overview to a follow-up in AI Mode. Google said AI Mode had passed 1 billion monthly users. Source: Google
Action for enterprise teams. Map the follow-up questions that sit behind your top commercial queries. The AI Overview and the AI Mode follow-up now draw from one conversation, so cover both the first answer and the next step on the same cluster of pages.
February 17, 2026: Claude web search adds dynamic filtering
What changed. Alongside Claude Sonnet 4.6 and Opus 4.6, Anthropic released new web search and web fetch tool versions. Claude writes and runs code that filters search results and fetched pages before they reach its context window. Anthropic reports an average 11% performance gain with 24% fewer input tokens across two benchmarks. A June 11, 2026 tool version added control over which result blocks return in the API response. Source: Anthropic
Action for enterprise teams. Filtering code keeps the passages that answer the question and drops the rest. Put the answer to each section's heading in its first sentence.
How we handle model rumors and codenames
Search logs fill with model names before and after launch. Queries such as "gpt 6 astra" and "gpt 6 sol" appeared in our Bing data this fall. We add a model or feature to this log only when the lab documents it in an official release note, model page, or developer changelog.
That rule applies here. GPT-6 Astra appears in OpenAI's release notes on September 3, 2026, and GPT-6 Sol and Luna appear on October 7, 2026, so both are logged as shipped models with their documented dates. A name that appears only in leaks, social posts, or search query data stays out of the log until the lab confirms it.
For enterprise teams, this keeps planning tied to behavior you can test. Build your response to an update after it ships and after you have a 7-day citation read, and keep codename spikes out of your prompt tracking so they do not distort your share-of-voice baseline.
How AirOps helps you respond to model updates
AirOps gives enterprise marketing teams the intelligence to see where their brand stands in AI answers across ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews, and AI Mode, plus the workflows to act on it. When an engine ships a change, you can compare citations by engine before and after the release and refresh the pages that moved. See how platforms compare in our roundup of AEO and AI search visibility tools, or explore the AEO content hub.
FAQ
How often do AI search engines update their models?
In 2026, the four engines in this log shipped a documented model or retrieval change every few weeks. Google, OpenAI, Perplexity, and Anthropic each shipped at least two between February and November.
Do model updates change which brands get cited?
They can. Changes to retrieval, link display, or answer format shift which pages an engine reads and shows. Measure citations by engine for 7 days after a release before you change content.
Why did ChatGPT referral traffic drop in October 2026?
From October 8 to October 16, 2026, ChatGPT links were missing the utm_source=chatgpt.com tag and a referrer, so many visits appeared as Direct in analytics. Search Engine Roundtable reported the tags returning on October 15.