AI Search News: Claude's GEO Rules, Gemini 4 Argon in AI Mode

This week's AI search news starts with a rule, not a feature: Anthropic's updated usage policy bars using Claude to seed AI answers with fake "independent" sources. Google put Gemini 4 Argon into AI Mode for paying subscribers, Bing launched the first citation gap report from an AI search engine, and new Ahrefs data shows free ChatGPT rarely searches the web at all. Here's what changed from October 8 to 13, 2026, and what it means for your brand.

OpenAI had a quiet week for brands: we found no search, shopping, or ads updates in its newsroom or ChatGPT release notes, and no changes to its crawler documentation.

Claude usage policy: no seeding AI answers with fake independent sources

On October 8, Anthropic published an update to its usage policy that takes effect November 12. It gathers rules on influence operations into one new section, "Do Not Engage in Deceptive Campaigns or Artificial Activity," which Anthropic says covers commercial as well as political deception.

The policy text now prohibits using Claude to "Manipulate the sources from which search engines or AI systems draw answers by seeding them with content that misrepresents its origin, authorship, or independence," for example "networks of sites posing as unaffiliated sources corroborating the same claims." It also bans building tools for such campaigns and marketing them "as a product or service." Users are responsible for agents they deploy, including "actions those agents take through tools, browsers, or connected systems."

Background: Anthropic says it already prohibited this conduct, but the rules were scattered across several sections. Mike King of iPullRank wrote that, as far as he can tell, this is the first time a frontier model provider has spelled it out in plain language. Google's spam policies already count "attempting to manipulate generative AI responses in Google Search" as spam.

Why it matters for brands: This governs how people use Claude, not how Claude ranks content. But it draws a clear line for AI visibility work: disclosed promotion is fine, manufactured independence isn't. Our guide to Claude's usage policy and AI visibility sorts common tactics into allowed and prohibited.

What to watch: Whether OpenAI and Google add similar language to their own model policies.

Gemini 4 Argon comes to Google AI Mode

On October 13, Google released Gemini 4 Argon, its newest model, "for Google AI Ultra and Pro subscribers in the Gemini app and AI Mode in Search." In the Gemini app, Google says "Argon answers questions more directly with less hedging," and that Gemini "now connects you to more than 40 apps," like Instacart, OpenTable, Adobe, and Shopify.

Background: Free users are moving the other way. Since October 9, Gemini users without an AI subscription get Auto model selection by default, and Google's help page says most of their prompts go to Flash-Lite, its fastest model, while prompts that need deeper reasoning may go to Flash or Pro.

Why it matters for brands: A paying subscriber who switches AI Mode to Argon may now get a different answer about your category than a free user does. "Less hedging" may also mean shorter shortlists, though that's our guess, not Google's claim. If you track AI Mode or Gemini, note the date: comparisons across October 13 compare different models. Our guide to ranking in Gemini covers how Google's AI picks sources.

What to watch: Whether Argon becomes the default AI Mode model for everyone.

Bing Webmaster Tools AI Performance adds a citation gap report

On October 13, Microsoft added Topics Portfolio (Beta) to the AI Performance report in Bing Webmaster Tools. It "groups the grounding queries where your site appears into broader thematic clusters" and shows each topic's Topic Size ("approximate total citations across all sites"), Sites Cited, and number of queries.

The new part is competitive. Top Opportunity Queries lists "the strongest queries across your whole portfolio where other sites appear and yours does not," with citation volumes shown as ranges. Each topic's detail page shows your own citations alongside opportunity queries for that topic.

Background: The tab appears next to Citations & Pages only if there is enough data for your site, so smaller sites may see a shorter list or none. Bing calls the numbers "directional signals for planning, not as a precise ledger," and warns that branded queries can show up as opportunities.

Why it matters for brands: As far as we know, no other AI search engine gives site owners a first-party list of queries where competitors get cited and you don't. That's a ready-made content brief for Copilot and Bing's AI answers.

What to watch: Whether Google adds anything similar to Search Console's AI reporting.

ChatGPT query fan-out: free answers rarely search the web

On October 12, Ahrefs published an analysis of 66.6 million free, logged-out ChatGPT responses (on GPT-5.6 Luna) and their 15 million fan-out queries from August and September. Only 14.4% of responses triggered a web search, averaging about 1.6 searches each.

When ChatGPT searched a specific site with site:, it was usually a brand: of the roughly 160 most-searched sites, 75.9% of those searches targeted brand sites. "2026" was the most common search term, and "price" rose from 6.0% to 8.9% of fan-outs between August and September. Citations shifted too: about 1 in 5 answers cited Reddit in March, versus about 1 in 43 by late September, while OpenAI's partner publishers went from 7.5% of citations among the top 200 domains in March to 12.5% in late September.

Background: This is vendor data, and it predates GPT-6, which OpenAI says is better at deciding when to search.

Why it matters for brands: For most prompts, ChatGPT answers from what it already knows about you. When it does look you up, it often goes straight to your site, so put prices, specs, and policies in plain text. Our guide to checking if ChatGPT mentions your brand shows how to see which pages it cites.

What to watch: Whether GPT-6 searches more often.

AI Overviews rank tracking data may be wrong since mid-September

On October 10, rank tracking company Nozzle reported that since September 13, 8.0% of the Google results pages it collected came back "poisoned," against 0.17% before, and that on the worst days, September 27 to 29, "it was a quarter of all SERPs." Nozzle describes these as "complete, normal-looking pages whose results have been swapped out." The sample covers 27.5 million results pages for more than 660,000 keywords in 49 countries.

AI Overviews are affected. An AI Overview appeared on 34.9% of poisoned pages versus 54.2% in Nozzle's baseline, and Overview sources carried over from the previous check 42.3% of the time versus 68.8%. Nozzle attributes the poisoning to Google's response to AI scraping. Google hasn't commented.

Why it matters for brands: If a scraping-based tool showed your AI Overview presence or citations dropping since mid-September, the data may be the problem. Nozzle found 77.9% of poisoned pages had their usual sites back within three days, so don't react to a one-day drop, and cross-check with Search Console, which doesn't rely on scraping.

What to watch: Whether other trackers confirm the pattern and how they filter it out.

Google tests pharmaceutical ads in AI Mode and AI Overviews

On October 12, Adweek reported that Google is letting pharmaceutical advertisers, including for prescription drugs, run ads in its AI search features. A Google spokesperson said: "We're testing standard text and Shopping ads in AI Mode and AI Overviews, which we recently expanded to include certain pharmaceutical products." Adweek says the test began this month, is limited to the US, and shows ads in sections labeled "Sponsored."

Background: Search Engine Roundtable's Barry Schwartz wrote that Google told him "there were no recent policy changes here," and that he "did not spot any" of the ads himself.

Why it matters for brands: Ads in AI answers are reaching regulated categories, starting with one of the most tightly controlled. Health brands now compete for paid as well as organic space beside AI summaries. The ads appear in labeled sponsored sections, so a place in the AI answer itself still has to be earned organically.

What to watch: Whether other regulated categories, such as financial services, follow.

Shopify shares sale prices with AI shopping agents

On October 9, Shopify announced that it "now shares your compare-at price with AI shopping agents through Shopify Catalog, so AI agents can see when a product is marked down." Before, agents reading Catalog "only saw the current price." Merchants can turn this off with a toggle in the "Compare-at" section of their Shopify Catalog mapping settings in admin.

Background: Shopify Catalog is the feed through which Shopify shares merchants' product data with AI shopping agents. The changelog doesn't name which agents receive it.

Why it matters for brands: Shopping agents compare prices for a living, and a visible markdown gives them a reason to pick your listing. Heading into holiday sales, check that compare-at prices are accurate before the data reaches agents automatically. Our guide to AI visibility for ecommerce covers what else shopping assistants look at.

What to watch: Which agents show the markdown to shoppers, and how.

Also this week

• Forum posts get a fast lane into Google (Oct 8). The new UGC Fresh Data Program lets approved platforms that "primarily host UGC, social postings, or forum discussions" send posts and engagement data to Google through a dedicated pipeline, "independent of the Google Indexing API and traditional organic crawling." Google says it "doesn't guarantee that content will appear in Search."
• The September 2026 spam update finished (Oct 8). It ran from September 24 to October 8 and "applies globally and to all languages."
• Ask YouTube is now available to US viewers on web, mobile, and TV (Oct 12). YouTube says its conversational search lets people "ask complex queries, refine with follow up questions," and get results "in an interactive format."
• AI visibility differs by country (Oct 13). In Semrush's first UK benchmark, any two AI platforms shared 63 to 70 of their top 100 mentioned brands, versus 70 to 78 in the US. In July, "Tesco ranked as ChatGPT's 25th most-cited UK domain, while Asda ranked 606th."
• Profound now serves AI bots a pre-rendered copy of your pages (Oct 11). Its Dynamic Bot Rendering means "AI and human visitors can see two different versions of the same page." Google's dynamic rendering documentation says this isn't cloaking "As long as your dynamic rendering produces similar content," so keep the two versions in step.

What to do this week

1. Audit your off-site AI visibility tactics. List every review site, roundup, forum account, and microsite you pay for or control. Anything that looks independent but isn't should be disclosed or dropped. Our GEO guide covers the tactics that hold up.
2. Open Bing Webmaster Tools' AI Performance report. If Topics Portfolio appears, export the opportunity queries for your strongest topic and check which pages could answer them.
3. Recheck AI Overview tracking before acting on drops. For any decline since mid-September, compare against Search Console and look for a three-day recovery before changing pages.
4. Shopify merchants: review the compare-at setting. Decide whether agents should see markdowns, and make sure every compare-at price is one you actually charged.