LabraSEO

ChatGPT favours primary sources 📊, Claude bans AI manipulation campaigns 🚫, Google Discover goes AI concierge 🤖

Issue 13, 13 October 2026

What changed in search and AI answers, for marketing teams, in five minutes.

Headlines & Launches

Google plans to turn Discover into an AI concierge (5 minute read)

A job posting for a Principal UX Lead describes transforming Discover from a recommendation feed into a "proactive, hyper-personalized concierge experience" that helps users act on goals, not just read articles, according to Search Engine Journal. If this ships, Discover traffic to publishers could shrink as Google keeps more of the interaction inside its own interface, so treat any Discover traffic you have as less stable going forward.

Google is hiring to generate AI content inside Discover (3 minute read)

Open Google job listings describe engineers building AI-generated content (AIGC) for Discover's recommendation feed, including agentic systems that write and check stories for factuality and tone, Search Engine Journal reports. Search Console already reports when your links appear in Discover's generative AI features, so check that report if Discover sends you meaningful traffic.

Google starts testing prescription drug ads in AI Mode (1 minute read)

Google confirmed to Adweek that it has begun testing text and Shopping ads for branded prescription drugs inside AI Mode and AI Overviews, limited to the US and labelled "Sponsored". If you are in a regulated category like pharma, watch for this test to expand, since it signals Google is opening AI answers to more ad categories generally.

Google Merchant Center: poor image quality no longer blocks listings (3 minute read)

Google updated its help documentation so that products with blurry or low-resolution images will keep showing in Search and Ads, where previously they could be suppressed, Search Engine Roundtable reports. Your products will still appear, but Google implies click-through could suffer, so it is still worth fixing weak product images.

Deep Dives & Analysis

Ahrefs studied 15 million ChatGPT fan-out queries (14 minute read)

Ahrefs analysed 66.6 million free, logged-out ChatGPT responses from August and September and found that only 14.4% of them triggered a web search at all, with the rest answered from the model's own knowledge. For brands chasing ChatGPT citations, this means most questions about you never reach a search step, so winning the minority that do search matters more than publishing volume.

ChatGPT's citations are shifting toward primary sources (4 minute read)

The same Ahrefs data shows PubMed climbing from 64th to 4th among ChatGPT's most-cited US domains between March and September, while explainer sites like WebMD and Investopedia fell sharply; Ahrefs says ChatGPT "seems to be prioritizing original sources over sites that summarize them." If your content mainly repackages other people's research or data, expect your share of ChatGPT citations to keep eroding in favour of original studies, government data and first-hand testing.

Anthropic's new usage policy bans gaming AI answers (14 minute read)

Mike King at iPullRank highlights a new clause in Anthropic's October 8 usage policy that prohibits using Claude to seed content across sites that "misrepresents its origin, authorship, or independence" to manipulate what search engines or AI systems cite, taking effect November 12. King compares it to Google's early, loosely enforced webmaster guidelines, meaning coordinated fake-source networks built to game AI citations now carry policy risk with Anthropic, though enforcement is untested.

Google AI Overviews now answers typos and nonsense queries (3 minute read)

Barry Schwartz at Search Engine Roundtable shows Google AI Overviews responding even to garbled, no-result queries, sometimes guessing the query was a typo rather than returning nothing. Schwartz says this is not new but reflects Google's push to answer any query with AI rather than admit it does not know, so do not assume an AI Overview appearing means your content was specifically relevant.

Tactics & Tools

RankZero's free AI SEO audit (Sponsor)

See whether ChatGPT, Perplexity and Google AI Overviews can find and cite your brand. Free, results in 30 seconds, no signup.

Quick Links

Bing and Copilot are testing a new Microsoft decision model (3 minute read)

Microsoft confirmed Bing Search and Copilot are experimenting with Microsoft-Decision-1, a model built for routing, classification and ranking tasks, Search Engine Roundtable reports. It is not yet used for all queries or users, so no visible ranking change to act on, but worth watching if Bing traffic moves unexpectedly.

Google Ads tests AI-suggested goals and assets (2 minute read)

Google confirmed it is testing a feature where Google AI suggests campaign goals and ad assets based on a short description you type in, rather than manual setup, Search Engine Roundtable reports. If you run Google Ads, this could show up in your account as an experiment worth trying.

From X and LinkedIn

Jake Ward: Check whether your topics match the page type Google already ranks (2 minute read)

On X, Jake Ward describes fixing a client's topic strategy by cutting overly competitive and overly narrow topics, and matching page type (product, collection, blog post) to whatever already ranks on page one. Before publishing new pages, check what format currently ranks for your target keyword in Google and in AI answers, and build that format rather than defaulting to a blog post.

Nicholas Dulait: Pruning weak pages can lift a site's traffic (5 minute read)

On X, Nicholas Dulait describes deleting 20 of about 100 blog pages with zero clicks and under 50 impressions over 90 days, after which monthly clicks rose from 1,500 to 15,000; this is one site's result, not a controlled study. Run a similar Search Console filter on your own site (clicks, impressions, age) and decide for each flagged page whether to delete, merge or rewrite it, rather than assuming more pages always helps.

Cyrus Shepard: Google's new UGC Fresh Data Program (1 minute read)

Cyrus Shepard notes Google's new program invites sites to send fresh user-generated content directly to Google, and sarcastically questions whether it mainly feeds AI training and AI answers rather than helping users find your content.