LinkedIn Tops AI Citations, Google Rewrites Headlines & the March 2026 Spam Update

Google released the March 2026 spam update on March 24 at around 3:20pm ET. It applies globally across all languages and is expected to complete within days — not weeks — based on the rollout language Google used. This targets sites violating Google's general search spam policies: thin content, hidden text, cloaking, structured data misuse, and similar violations. Google described it as "a normal spam update."

The more useful detail is what this update explicitly does not target: link spam and site reputation abuse are both excluded. Those have their own dedicated updates. That distinction matters for triage — if you see movement from this update, you are looking at general spam compliance issues, not your link profile. Google declined to say what share of queries are affected, and recovery from confirmed spam actions typically takes months with periodic refreshes rather than a single clean pass.

Google confirmed to The Verge that it is running a "small and narrow" test of AI-generated headline rewrites in traditional search results. This is not the existing system that pulls from H1 tags or meta descriptions — the AI is generating entirely new titles from scratch. One documented example: an article titled "I used the 'cheat on everything' AI tool and it didn't help me cheat on anything" was rewritten to "'Cheat on everything' AI tool" — a phrase that never appeared in the original content. Google's stated goal is to "identify content on a page that would be a useful and relevant title to users' query."

The Discover precedent is instructive. Google described AI headlines in Discover as a "small UI experiment" in December 2025. By January 2026, after showing good user satisfaction, it was reclassified as a feature. The same framing — "small and narrow" — is now being applied to Search. Publishers have no disclosure when their headlines are rewritten and no opt-out mechanism. Discover already accounts for roughly 68% of Google-sourced traffic for major publishers, meaning AI headline rewrites are already live on the channel that sends most of their referrals.

Two practical technical updates from this week's SEO Pulse worth noting separately. First: Google expanded Personal Intelligence from paid AI Pro and Ultra subscribers to all free US users on personal Google accounts as of March 20. AI Mode can now access Gmail and Google Photos to personalise responses — referencing email confirmations, travel bookings, photo context. This is limited to US personal accounts for now; Workspace accounts are not included.

The crawl limits clarification is the more durable of the two. Technical SEOs have long worked with Google's stated 15 megabyte crawl limit. Gary Illyes and Martin Splitt confirmed this week that the practical threshold in Google Search is closer to 2 megabytes — and that internal teams can override the 15MB figure. The 15MB limit is a default, not a hard ceiling. For most pages this changes nothing in practice, but for sites with heavy JavaScript payloads, large inline scripts, or dense HTML structures, 2MB is the number to work against.

Georgieva's argument here is uncomfortable because it is accurate. While the industry spends considerable energy debating GEO tactics and AI Mode strategy, most SEO programmes in 2026 will fail for reasons that have nothing to do with algorithms: unclear ownership, fragmented data, misaligned KPIs, and the gap between knowing what to do and getting it done. She documents a specific case where cross-functional teams "hadn't even read the strategy document" — which produced a GEO initiative that stalled at the planning stage with no implementation.

The AI over-reliance point is worth pulling out separately. When every team uses the same tools to answer similar questions, content outputs start to converge. That generic output is exactly what large language models are least likely to cite — AI citation systems favour content with specific claims, original data, and clear authorial perspective. The practical recommendations are not glamorous: tie visibility metrics to business outcomes, assign ownership explicitly, run smaller experiments rather than lengthy strategy documents, and train non-SEO teams on why any of this matters to their work.

This is the most data-rich GEO study published this year. SEMrush analyzed 89,000 unique LinkedIn URLs cited across 325,000 prompts in ChatGPT Search, Google AI Mode, and Perplexity. The headline finding: LinkedIn appears in 14.3% of ChatGPT Search responses, 13.5% of Google AI Mode responses, and 5.3% of Perplexity responses for professional queries — ahead of Wikipedia, YouTube, and every major news publisher. Corroborating data from Profound's separate 1.4 million citation analysis shows LinkedIn moving from outside the top 20 on ChatGPT in November 2025 to around position five by February 2026.

The content type findings are where this becomes actionable. 95% of LinkedIn citations go to original posts — reshares are nearly invisible to AI citation systems. Long-form articles between 500 and 2,000 words dominate citation rates for longer content; feed posts perform best between 50 and 299 words. Cited authors post 5 or more times per month in 75% of cases. Follower count does not reliably predict citation frequency — smaller accounts with high content quality do get cited. The content type shift is also significant: profile page citations dropped from 33.9% to 14.5% between November 2025 and February 2026, while posts and articles combined rose from 26.9% to 34.9%. AI systems are moving from static identity signals to dynamic content signals.