Latest Technical SEO News - Page 6
Crawling changes, algorithm updates, rendering fixes, and structured data developments as they happen.
-
Google's CEO says agents will browse sites before protocols mature
Sundar Pichai says agent protocols will scale better but take longer. Google-Agent, which usually ignores robots.txt, is the traffic to handle first.
-
One site's impressions hit zero while live URL tests still pass
A site owner says Search Console's Page Indexing report is stuck on October 12, with checks in server logs and Bing that separate stale data from deindexing.
-
Review searches now trigger an AI Overview 90% of the time
Semrush tracked 400 commercial keywords on US desktop from September 2025 to August 2026. Every modifier group, including pricing, finished above 80%.
-
Two stale blog posts made Google's AI Overview deny a new API
Google's AI Overview said Feature Upvote had no REST API because two old blog posts said so in passing. Correcting those pages and resubmitting them fixed it.
-
Google organic drove 27x more conversions than all LLMs combined
Across 100 sites and 145 million GA4 sessions, LLM referrals were 0.73% of Google organic volume, and ChatGPT visits converted at 13.8% against 3.02%.
-
ChatGPT's searches name their source and a freshness window
Mark Williams-Cook's Search Engine Journal piece names the result_source field and the day-count recency window visible in ChatGPT's Fetch/XHR requests.
-
China's ad industry sets 30 runs per prompt for AI visibility
China's new GEO standard puts an 8-point error bar on 30 runs, excludes branded prompts from visibility rates, and makes vendors list what they cannot promise.
-
Google's robots.txt parser silently drops lines it can't read
Google's robots.txt video confirms the parser accepts only user-agent, allow, disallow and sitemap, and advises percent-encoding for non-ASCII paths.
-
Meta's shopping agent lands as direct traffic from Linux Chrome
Botify's breakdown of Meta Muse: sessions from Cloudflare and Fastly IPs that GA4 counts as human, connector orders with no session, and how to count both.
-
Google Discover drops stale pages before ranking scores them
Google's Supersession-Decay Filtering has pruned superseded and decayed items from Discover for two years, and it decides eligibility before ranking runs.
-
Some AI Overviews lost their desktop label but kept it on mobile
Glenn Gabe spotted AI Overviews for shows and movies rendering unlabeled on desktop but labeled on mobile, which may account for Semrush's lower AI Overview counts.
-
DeepSeek's new search API scores your host, not your pages
A 110-call test found every URL on a host sharing one authority tier, spam PDFs on a US government host at tier 4, and Fetch returning robots.txt-disallowed pages.
-
Google treats ?filter[]= and ?filter%5B%5D= as separate URLs
John Mueller, citing RFC 3986, says the raw and percent-encoded bracket forms are not equivalent, so faceted sites emitting both leave Google to dedupe them.
-
Translated titles missed the phrase local searchers type
A developer running a 13-language site pulled country-level Semrush volume and found Japan's real query drew 2,900 searches a month against the title's 260.
-
Meta's AI agent reads the accessibility tree, not the DOM
Meta says Muse's browsing agent reads the accessibility tree, so facts kept only in JSON-LD and unnamed buttons never reach it. No user agent is published.
-
Desktop merchant impressions ran 3 to 7x high in one SEO's data
A practitioner's Search Console data across several shops shows merchant listing desktop impressions inflating from May 16, stopping on Oct 9, with CTR near zero.
-
DeepSeek's new Search API rates your pages 0 to 4 for authority
DeepSeek's Search API returns an authority tier and a 0 to 3 rerank score per result, accepts site: queries at $0.75 per 1,000, and names no crawler for the index.
-
Nine sites block GPTBot but let OpenAI's search bot crawl
A Reddit test resolved 19 AI user agents against robots.txt on 122 well-known sites. Bytespider drew the most blocks and user-triggered fetchers far fewer.
-
Entity linking scores flip when the API finds the names itself
WordLift tested five entity linking APIs on VoxEL by exact Wikidata ID match. Its own resolve leads only when handed the names; Google leads end to end.
-
Four AI search tools cited an invented concept within a week
Fudan researchers posted about a fictional protocol on nine platforms AI tools already cite; Grok linked to it in 30 of 32 answers, and ChatGPT in 7 of 20.