Technical SEO 2026: CWV, Semantic Search, and GEO Actions

I treat technical SEO as engineering: budgets, traces, schemas, and deploys. Ranking copy without a crawlable, fast, unambiguous HTML document is theater. Below is the checklist I use on client and own properties, including generative/AI search work that used to live in a separate 2025 SEO guide (now merged here).

1. Core Web Vitals: optimize friction, not vanity scores

INP (Interaction to Next Paint)

INP replaced FID as the interactivity vital. Target: good under 200ms for field data (CrUX), not a single lab run on a quiet cable connection.

• Break long tasks; yield to the main thread so clicks paint soon.
• Audit third-party tags. Chat widgets and tag managers are common INP killers. Move non-critical work to workers where feasible (for example Partytown-style patterns) or load after interaction.
• Prefer fewer synchronous scripts in <head>. Defer analytics until after first meaningful interaction when product allows.

LCP and CLS

• Mark the true LCP element with fetchpriority="high" and a matching preconnect for the image host.
• Reserve space with width/height or CSS aspect-ratio so late ads and embeds do not shove content.
• Avoid inserting banners above existing content after paint without a reserved slot.

Measure with CrUX + your RUM. Lab Lighthouse is a regression gate, not the source of truth.

2. Semantic / entity structure

Engines map topics as entities and relationships, not only keyword strings. Practical moves:

• Build topic clusters with intentional internal links between parent and child URLs (hub pages that answer the class of questions, spokes that go deep).
• Use BLUF (bottom line up front): open major H2 sections with a 40 to 60 word direct answer before the narrative. That block is what answer engines lift.
• Prefer question-shaped H2/H3 when the query is interrogative (“How do I reduce Java allocation churn?”) instead of vague labels (“Tips”).

3. Schema as the machine API for your page

JSON-LD is how you reduce ambiguity for crawlers and summarizers.

• Baseline: Article / BlogPosting with author Person, datePublished, dateModified.
• Add FAQ or HowTo only when the visible page truly contains that Q&A or steps (no hidden FAQ spam).
• Use stable @id URIs for your org and authors; link sameAs to real profiles (LinkedIn, GitHub, Wikidata when accurate).
• If you publish original datasets or benchmarks, Dataset markup helps machines treat you as a primary source.

Validate with Rich Results Test / schema validators after each template change.

4. Mobile-first indexing without AMP nostalgia

The smartphone Googlebot view is the index for most sites. Parity matters: same primary content, same structured data, usable touch targets.

• SSR or solid hydration for content that must rank; do not hide the article body behind a client-only fetch with no HTML fallback.
• Serve appropriately sized images (client hints or srcset), not a desktop hero on 3G.
• Keep DOM depth and CSS weight under control; mobile CPUs punish nested theme wrappers.

AMP is not a ranking requirement in 2026. Fast responsive HTML wins.

5. Indexing speed and bot policy

• IndexNow (or search-console URL inspection + sitemaps) on publish/update so refreshed docs do not wait on crawl luck.
• robots.txt: allow search crawlers you care about; decide deliberately on AI training bots. Blocking training crawlers is a business choice, not a ranking hack. Document the policy for legal/comms.
• Fix soft 404s, redirect chains, and parameter duplicates. Crawl budget still matters on large sites.

6. GEO / AI-search actions (merged from the 2025 guide)

Generative Engine Optimization is not a new meta tag. It is making your page the easiest correct citation.

Concrete GEO checklist

1. Information gain: ship one fact only you have (benchmark, screenshot of your UI, config that failed, before/after latency). If an LLM can invent the paragraph unaided, cut it.
2. Extractable answers: 40 to 60 word definition under the H2 that matches the query; then detail.
3. Author entity: byline linked to an about page with credentials and sameAs links. Anonymous “team” blogs underperform for YMYL and expert topics.
4. Citations outbound: link primary docs (MDN, vendor docs, RFCs). Summarizers prefer grounded pages.
5. Original media: diagrams and tables beat stock illustrations for reuse and trust signals.
6. Track AI referrers: in analytics, segment referrals from known AI chat/search hosts separately from classic organic. Treat them as high-intent even when volume is small.
7. Community corroboration: useful answers on GitHub, Stack Overflow, or niche forums often feed models and humans; do not rely on the blog alone.
8. Stop volume farming: one definitive 800 to 1500 word page beats five thin rewrites of the same SERP.

Threat vs opportunity (short)

7. Implementation order I recommend

1. Fix CWV regressions on money pages (INP first if field data is poor).
2. Correct canonical/indexation bugs.
3. Add honest schema + BLUF on top landing URLs.
4. Run the GEO checklist on the 20 URLs that already earn impressions.
5. Only then expand net-new content.

Related reading: Model Context Protocol guide, AI agent harnesses, Claude Skills series, JavaScript in AI development, and n8n AI workflow.

FAQ: Technical SEO in 2026

What is the most important Core Web Vital right now?

For interaction-heavy pages, INP. Still watch LCP and CLS; a single “good” INP with a 6s LCP is not success.

How does semantic search change keyword research?

Map entities and questions people ask, then cover the cluster with linked URLs. Exact-match density theater is not the job.

Is AMP required?

No. Performance parity on responsive pages is the bar.

Why bother with schema if AI can read HTML?

Schema removes ambiguity (which “Mercury”? which author?). It does not replace clear writing.

What happened to the Ultimate SEO Guide 2025?

Its useful GEO and E-E-A-T tactics are merged into this page. The old URL redirects or points here so you maintain one living technical + AI-search reference.