Search Signal #11
Reddit's share of ChatGPT answers fell from 19.1% to 2.3%. Plus: Meta pauses its crawl terms, Gemini's hidden user agent, bot clicks in Search Console.
Vikas Kalwani
Founder, SAASY LINKS
01. Reddit's share of ChatGPT answers fell from 19.1% to 2.3%
Ahrefs checked all 85 OpenAI publisher-partner domains against Brand Radar's November snapshot of 5.5 million US ChatGPT answers, published 3 November. Reddit's share of ChatGPT answers fell from 19.1% in March 2026 to 2.3% in late September, and Wikipedia's from 14.1% to 2.9%. Partner deals bought little on their own: Vogue holds 42,108 citations, The Atlantic 724, Stack Overflow 130. US-only, a single snapshot, and vendor research.
Why it matters: The Reddit-mention tactic that half the GEO market sells is priced off a citation surface that has shrunk by roughly a factor of eight inside ChatGPT. A playbook built on 2026 screenshots is already out of date.
Action: Run your ten most important buying prompts in ChatGPT this week and record how many citations point at Reddit, with the date, so you hold a measured baseline instead of an assumption.
02. Reddit mentions still correlate hardest with AI visibility, at 0.75
Published the same day, Ahrefs compared seven kinds of off-site brand mention against AI visibility for 10,000 US brands across 260 million non-branded prompts. Spearman correlations ran Reddit 0.75, domain mentions 0.70, webpage mentions 0.70, news 0.67, YouTube 0.64, TikTok 0.60. Returns diminish sharply: 10 times more webpage mentions associated with about 3.7 times more AI mentions. Within search-volume bands the correlations drop to 0.34 to 0.54, so brand size explains much of the headline figure.
Why it matters: One vendor published both this and item 01 on the same day, which is the live tension on this beat: Reddit correlates best with being mentioned while its ChatGPT citation share collapses.
Action: Before funding a Reddit programme, ask the vendor for the within-band correlation rather than the 0.75 headline, and which specific engine their visibility score is measured on.
03. Meta paused the 20 November terms, 15 days before they were due to bite
On 5 November Meta posted a Business Help Center notice saying updates to its Commercial, Product Catalog, Self-Serve Ad and Ad Creative Generative AI terms "are now paused and will not take effect on November 20, 2026." A spokesperson told Nieman Lab Meta is "pausing the changes as we continue to engage with stakeholders." Three legal hits preceded it: the Oslo District Court injunction on 28 October, a Sao Paulo order on 3 November carrying R$1 million a day, and the UK CMA case opened 30 October.
Why it matters: Paused is not withdrawn. There is no new date, no named provision Meta will rewrite, and the clause authorising automated visits to any website linked to a business account is still on the table.
Action: Send the Commercial Terms crawl clause to your legal contact this week rather than waiting for a new date, and confirm robots.txt names Meta-ExternalAgent and Meta-WebIndexer as separate decisions.
04. Gemini fetches your pages as "Google" and AI Mode as "Google-Lens"
Dave Smart of Tame the Bots prompted each service to fetch URLs carrying unique query parameters, then read his server logs. The Gemini app fetches live pages using the bare user-agent string Google, while AI Mode in Search uses Google-Lens. Some requests resolved to a google-proxy hostname inside Google's user-triggered-fetcher range, though not all did. Both pull only the HTML document and render nothing. One of his clients' firewalls was blocking the Google string outright.
Why it matters: A generic firewall rule can cut a site out of Gemini citations silently: no error page, no Search Console message, and nothing visible in any AI visibility tool.
Action: Grep your CDN and WAF logs for user agents exactly equal to Google and to Google-Lens, and check whether either is being challenged, rate-limited or blocked.
05. OpenAI is building a Search Console for ChatGPT
Juozas Kaziukenas spotted a "Claim your business in ChatGPT" flow inside a wider ChatGPT Business Hub and posted screenshots; Search Engine Roundtable wrote it up on 6 November. Per his account it charts daily impressions and clicks through to the site, with no prompt-level data, no user data, no competitor benchmarks and no rankings, gated on verified ownership such as a connected Shopify store. OpenAI has announced nothing, so treat every specific as provisional.
Why it matters: A first-party impressions and clicks feed would reset what AI visibility vendors can credibly charge for estimating the same two numbers.
Action: Avoid signing a multi-year AI visibility contract this quarter, and make sure whoever controls your Shopify store or domain verification can claim the property the day it opens.
06. Google is throttling goto link resolution and your rank data is quietly wrong
SE Ranking reviewed 10,000 Google US desktop keywords on 3 November and found 84% of flagged cases were unrelated listings matching a single word from the query, with 96.5% of those blocks starting at position 9 or lower. DataForSEO says Google now caps how many goto links one results page can resolve per period, leaving its High Priority and Live modes resolving about half. Serper logged the limitation on 30 October; SerpApi's incident opened 3 November.
Why it matters: Any ranking drop to position 9 or lower reported since late September may be a tracking artefact rather than a real loss, and it is flowing into agency reports unchallenged.
Action: Before anyone is held to a ranking decline, validate it against Search Console impressions and average position for the same queries over the same date range.
07. Search Console clicks spiked up to 6.5x on 2 November, and it looks like bots
Marie Haynes compared three anonymised properties against a 26 October to 1 November baseline and found 2 November click multipliers of 6.5x, 4.1x and 2.1x, heavily desktop-skewed. The signatures are implausible: click-through rates of 45% to 100% at positions 11 to 55, and desktop CTR moving from 0.20% to 2.89%. Joy Hawkins reported three affected sites, said she was fairly confident it was bot traffic, and said things were normal again by 4 November. Google has not commented.
Why it matters: Unfinalised Search Console data is about to arrive in November reporting as a fake win, and whoever presents it will have to walk it back in December.
Action: In Search Console, filter the performance report to desktop for 31 October to 4 November and compare clicks against GA4 sessions before that number reaches a board deck.
08. The New York Times sued Brave over a 30 billion page index you cannot block separately
The Times sued Brave Software in Manhattan federal court on 2 November, with the Chicago Tribune, Daily News and Orlando Sentinel filing a parallel case hours later. The complaint alleges Brave copied more than 30 billion webpages without identifying its crawler or offering a way to block it, and resells the content through its Search API to AI developers including Anthropic, Perplexity, Mistral, xAI and Cohere. Brave uses no distinct user agent and says it follows Googlebot's rules. These are allegations with no ruling.
Why it matters: An AI index that identifies itself as Googlebot cannot be refused without refusing Google, and the only removal route Brave documents is noindex, which also removes the page from Google.
Action: Run a site: query for your domain on search.brave.com to size your exposure, and check whether the descriptions it serves match your current pages or a stored older copy.
09. Google moved brand AI Overviews into the right column, displacing Knowledge Panels
Around 2 November Google shifted AI Overviews for many brand searches out of the main desktop results and into the right-hand column, where Barry Schwartz says they are effectively replacing brand Knowledge Panels. Chris Long and David McSweeney flagged the change on X, and Sachin Patel noted Knowledge Panels now carrying an AI-generated label. Screenshots cover Apple, Zillow and Coca-Cola. There is no Google statement, so this is practitioner observation rather than a confirmed rollout.
Why it matters: On a branded query, the description a buyer reads is now generated rather than assembled from the structured sources a brand could influence directly. Brand entity hygiene becomes a visibility task, not a tidiness task.
Action: Search your own brand on desktop, screenshot what the right column says today, and correct any wrong descriptor at source: your about page, your Wikipedia or Crunchbase entry, and your G2 category.
10. The EU set its first test for when an AI answer infringes
Advocate General Maciej Szpunar delivered a non-binding opinion on 3 November in case C-250/25, Like Company v Google Ireland, the first generative AI case to reach the Court of Justice. His test: a chatbot answer infringes press publishers' rights only if it circumvented a paywall or bot block, or reproduces the expression so closely that a reader no longer needs the original. On training he wrote "A trained LLM is not a database." The Court has not ruled.
Why it matters: If the Court follows the opinion, access control rather than copyright becomes the publisher's real lever in the EU, which turns a paywall or a bot block into a legal instrument.
Action: No action needed yet, the opinion is non-binding and a ruling is months away. But if you gate research in the EU, record now whether your paywall and robots.txt actually block AI fetchers, because that record is what the test turns on.
11. Dominant brands get cited 1.4x to 2.8x more, and content rewrites barely move it
A preprint posted 5 November audited generative search across 1,176 market players and found dominant brands cited 1.4 to 2.8 times more often than average ones. The authors locate the bias in the models' parametric memory rather than retrieval, since adding web retrieval does not correct it, and report that fact-preserving content rewrites produce negligible gains. Swapping authentic brand signatures for unfamiliar ones lowered recommendation rank with product attributes held constant. A preprint with no venue, and the engines audited are not named.
Why it matters: If the bias sits in model weights rather than in retrieval, the core GEO premise that rewriting a page buys you into answers is weaker than most vendors are pricing it at.
Action: Move part of your AI visibility budget from on-page rewriting to earning third-party mentions and reviews, which is the brand-familiarity signal this paper says dominates.
Also worth knowing
- Ziff Davis lost about 37% of its Google organic referrals in Q3, on top of roughly 23% a year earlier, and said its content appeared in AI Overviews and AI Mode about 2.7 billion times, close to ten times its Google Search clicks.
- Site reputation abuse penalties no longer apply in the EEA, per Search Console emails Google sent on 4 November for a policy change that took effect on 30 August 2026.
- Cloudflare's CEO called agent optimisation "sort of the new version of SEO" on Q3 revenue of $782.5 million, alongside a position paper listing 14 agent problems and per-agent allow or block controls in AI Crawl Control.
- AI Mode added direct booking for home services through Angi, Thumbtack, Housecall and GetJobber in the US, and one AI Mode response was spotted carrying four ads.
- OpenAI put ChatGPT at roughly one fifth of Google's search volume, counting every web fetch inside a conversation as a search, which is a very different thing from referral value.
- Gary Illyes played down click data arriving in Search Console's AI reports, saying the team wants to give "data that makes sense" but has nothing to announce.
SAASY LINKS
We build the visibility this newsletter reports on
Editorial links, AI citations, and organic pipeline for B2B SaaS. Senior-led, 90-day sprints, no long-term contracts.
Book a Call