Google researchers have developed a faster approach to query fan-out, the process that lets AI search explore several related searches from a single question. Called Retrieve-for-Train (R4T), the framework aims to make those searches more varied while reducing the time spent generating them.

Google described the research on September 15. Its tests used fashion and music datasets; deployment in Google Search remains unconfirmed. Search Engine Journal’s Roger Montti covered the announcement on September 24.

Making each branch of the search useful

As SEW’s query fan-out explainer describes, a question about a laptop for college and gaming could involve separate searches for prices, gaming performance, weight and battery life. Each search contributes something needed for the answer.

Google’s researchers describe a problem with that process: a language model can generate several versions of essentially the same search. In its fashion example, searches about festival style become searches about festival fashion and clothes, covering similar ground.

R4T trains the model to produce a useful spread of searches that remain relevant to the request and match material in the database. The resulting examples then train a smaller diffusion model to reproduce that behaviour faster.

How R4T speeds up query fan-out

The smaller model generates retrieval directions directly as embeddings, numerical representations used to find matching items. It produces those directions together, avoiding the need to write out subqueries word by word.

In the paper’s efficiency test, systems generated 10 retrieval directions per query. For a batch of eight queries, the diffusion model took 0.07 seconds versus approximately 1.46 seconds for the autoregressive approach. For 1,024 queries, it took 4.21 seconds versus nearly 50 seconds.

Those measurements cover fan-out generation, rather than the full process of answering a question. The paper also acknowledges that some retrieval-quality assessments relied on an AI judge.

Query fan-out changes what makes a result useful

The interesting part for publishers is how this research evaluates a collection of results. A relevant result can still add little if the other results already cover the same information.

Consider the laptop example. Once a search system has enough information about price, a detailed battery test could help it answer an unresolved part of the question. Another general buying guide might contribute less. That is an illustration of how complementary sources can help answer a question, rather than a finding about which pages Google currently selects.

R4T makes that collection-level usefulness part of its training objective. Its practical appeal is the possibility of exploring distinct aspects of a request without making users wait for a language model to generate every search separately.

The post Google unveils a faster way to power query fan-out appeared first on Search Engine Watch.

Source:: searchenginewatch.com

OpenAI has introduced dots, AI agents designed to continue working between conversations, remember ongoing projects and use connected apps. The rollout began on September 29 for eligible Pro and Business Premium subscribers, with an administrator-enabled beta for Enterprise.

Each dot runs on GPT-6 Astra and has its own cloud computer and browser. OpenAI says its plugin system provides connections to more than 4,000 apps, with users able to reach their dot through ChatGPT, Slack or Teams.

What dots can do while you’re away

OpenAI’s examples include revising launch materials when a product’s scope changes and turning interview transcripts into show notes and social drafts. The company also describes an early tester whose dot spotted a missing invoice, prepared it and sent it after approval.

There are two different kinds of background activity. An assigned task can continue while you’re away. Separately, dots can inspect connected information for useful updates without a fresh request. OpenAI calls this proactive research, and restricts its tools to reading: they cannot send messages, edit connected apps or control a browser or computer. Follow-up actions must pass the usual permission checks.

Who gets access, and what’s included

The access documentation is more specific than the announcement’s reference to eligible markets. Pro access excludes the European Economic Area, Switzerland and the UK. Business Premium is supported across all ChatGPT-supported regions. Enterprise access starts switched off until an administrator enables it.

The first dot is included in Pro and Business Premium at no extra charge. OpenAI says conversations with dots do not count toward ChatGPT usage limits, although tasks started in Codex or ChatGPT Work use those products‘ allowances. Its help page says further plan-specific usage terms will follow after the first month.

Setup requires the desktop app or a desktop browser. Mobile conversations follow where available, and the gradual rollout may take several days to reach an eligible account.

Agencies need to account for what a dot remembers

Persistent context should help with recurring client work, but it also makes offboarding more complicated. OpenAI’s privacy FAQ says disconnecting an app stops new access while leaving information already learned in the dot’s context. Users currently cannot inspect, edit or delete individual dot memories. Clearing that context requires deleting the dot.

For an agency using one dot across several clients, that means removing one client’s connection would still leave previously learned information behind. Deleting the dot would clear its wider context too. Separate files and ChatGPT memories have their own deletion controls, so deleting the agent would only handle part of that cleanup.

The post OpenAI launches dots, AI agents that keep working after you log off appeared first on Search Engine Watch.

Source:: searchenginewatch.com

Google has detailed an AI system that investigates the channels behind mass-produced synthetic videos, using their activity and connections to uncover suspected spam networks.

Called SAFE, short for Scaled Abuse Forensics Examiner, it tackles a weakness in duplicate detection. AI lets spammers produce variations of videos that are different enough to escape matching, while the accounts distributing them may still behave alike.

The authors report early deployment, but give no launch date or confirmation that SAFE operates in Google Search.

Upload times and device clues become evidence

SAFE divides an investigation among specialist AI agents. One examines content for policy violations. Others investigate channel relationships and behavioral signals, including device fingerprints, synchronized uploads and bursts of publishing.

The paper gives an illustrative example of channels using identical operating-system versions and uploading within the same five-second window. A coordinating agent weighs the findings to distinguish coordinated abuse from organic activity.

Google’s researchers say early deployment accelerated investigations. They publish no numerical accuracy or time-saving results.

Changing the video can leave other clues intact

The useful idea is that content variation and operational variation are separate problems for a spammer. Generating a different video may evade a duplicate check while leaving shared account connections or upload patterns available to investigators.

That helps explain the value of examining a group of channels together. Individual uploads can look different while their combined activity supplies evidence of coordination.

The September update connection remains a suggestion

Roger Montti covered SAFE for Search Engine Journal on September 25, suggesting it may be connected to Google’s September spam update.

An author-uploaded copy on ResearchGate records June 22 as its upload date, putting the research in public view months before that coverage. The date establishes when the paper was shared; SAFE’s deployment timing remains unspecified.

The published description focuses on video and channel investigations. Its role in web-search enforcement would need separate confirmation from Google.

The post Google uses AI to hunt video spam through upload times and device clues appeared first on Search Engine Watch.

Source:: searchenginewatch.com

Google AdSense has updated its payment-threshold guidance to require identity verification at $0 for affected publishers. Those accounts must complete the check before they can start showing ads or receiving payments.

The requirement depends on location. Google’s account-setup guide dates the change to June 2026. Barry Schwartz reported the documentation update at Search Engine Roundtable on September 23, describing a reduction from the previous $10 identity-verification threshold.

What publishers need to provide

Google says publishers need a current government-issued photo ID. Depending on location, they may also be asked for a video selfie. Publishers offered that process can opt for manual review, which Google says takes two to three business days.

Its setup guide gives publishers 45 days from the first verification request to complete the process. After that, Google says it will pause account activation.

Address verification remains a separate step, triggered at $10 in earnings or the local equivalent. The payout threshold also remains separate: Google’s table lists $100 for US-dollar accounts, €70 for euro accounts and £60 for sterling accounts.

Verification becomes part of the launch schedule

For someone opening an affected account, identity checks now belong in the preparation for running ads. A publisher choosing manual review needs to allow for that processing time before expecting advertising revenue to begin.

Google says most publishers finish the verification process in five to ten minutes, but that estimate sits alongside the longer manual-review option. The 45-day deadline should therefore be treated as the time allowed to complete the request, rather than a period in which an affected account can run ads without verification.

Separately, existing publishers should also be preparing for Google’s change to how AdSense impressions are counted, scheduled for February 17, 2027.

The post Google AdSense moves ID checks ahead of your first penny appeared first on Search Engine Watch.

Source:: searchenginewatch.com

Claude suffered an outage on Tuesday, September 29, with users encountering failed requests, conversations that would not load and problems signing in.

According to Anthropic’s incident report, errors began at 14:00 UTC. The disruption affected Claude.ai, including its desktop and mobile apps, Claude Code, Claude Cowork and some API requests.

An initial mitigation reduced errors, but a second issue blocked sign-ins, new chats, file uploads and other functions. During the disruption, Anthropic advised users who were already signed in to stay logged in.

At 15:30 UTC, the company said all services had been operating normally since 14:59 UTC. The incident remained under monitoring.

Check the work you sent during the outage

Anthropic warned that some messages sent between 14:00 and 14:59 UTC may not have been saved.

For anyone using Claude for drafts, research or coding, checking that those exchanges appear in the conversation history is the immediate follow-up. A working chat window alone cannot tell you whether earlier work was retained.

The post Claude recovers from outage, but some messages may be missing appeared first on Search Engine Watch.

Source:: searchenginewatch.com