SEO Tips 2026
SEO in 2026 Means Doing Two Things Simultaneously That Used to Be the Same Thing
AI Overviews suppress organic CTR by 61% on informational queries. Being cited in those overviews restores 35% of that traffic. Rank in traditional search and get cited in AI answers — those goals mostly overlap. The divergences are where the real work is.
Citation in AI Overviews is now the primary SEO goal for informational content. Being mentioned without a link earns zero CTR. Being cited earns 35% organic lift and 91% paid lift on the same queries.
The December 2025 core update made the content type hierarchy definitive: unedited AI output loses 85–95% of traffic; AI-assisted with expert oversight holds or gains.
GEO isn’t replacing SEO. It shares ~70% of the same technical work. The divergences — answer-first structure, entity density, named citations — are additive, not alternative.
When recovering from a core update hit: fix technical issues (INP, Core Web Vitals) first, then rebuild content. Doing it backwards introduces signal noise that prevents quality improvements from registering. This sequencing error is responsible for most “6-month recovery” delays.
01 / THE PROBLEM WITH EVERY “BEST SEO TIPS” LISTBad timing — and a world that got smaller
Most SEO tip articles are written for a Google that existed 18 months ago. They recycle guidance from a search landscape where ranking #1 meant capturing 28–39% of clicks. That world didn’t disappear. It just got a lot smaller.
Advanced Web Ranking data still shows the #1 organic position capturing around 39.8% of clicks — on queries without AI Overviews. The catch: AI Overviews now appear on 13–25% of all searches, and Seer Interactive’s September 2025 study of 3,119 informational queries (25.1M organic impressions across 42 organisations, June 2024–September 2025) found that on those queries, organic CTR dropped from 1.76% to 0.61%.
That’s a 61% drop. Not theoretical. Measured, over 15 months, across 42 organisations. The trend line doesn’t flatten.
And the March 2026 core update just started rolling out. We don’t have full impact data because it takes 2–4 weeks to complete. What the three 2025 updates (March, June, December — each 13–18 days) showed consistently: demonstrated expertise, original content, and topic depth hold. Keyword-stuffed coverage and AI-churned summaries lose. Every single time.
The short version: you now need to do two things simultaneously that used to be the same thing. Rank in traditional search. Get cited in AI answers. Those goals mostly overlap — and the divergences are where the real work is in 2026.
02 / THE SIX THINGS THAT ACTUALLY MOVE RANKINGSNot twenty. Not a framework with phases. Six.
What follows is each one with the mechanism behind it — because that’s the part most tip lists skip, and it’s also the part that tells you whether a tactic applies to your specific situation or not.
There’s a meaningful difference between appearing in an AI Overview and being cited in one. This sounds obvious. It isn’t — because the gap is invisible in every standard analytics setup.
You’ll see zero clicks from an AI Overview that mentions your brand without a link — but GSC will still show it as an impression. Marketing teams measuring “AI visibility” by impression share are measuring the wrong thing. Most organisations currently cannot see the citation/mention gap at all. You need a dedicated AI citation tracker to surface it.
Run your top 20 informational queries through an AI citation tracker (Profound, Seer’s generative AI tracker, or Semrush Enterprise AIO — each uses different sampling methodology, none is definitive). Identify queries where you appear but aren’t cited. Those are your highest-ROI revision targets. The fix is usually structural: add a direct-answer sentence within the first paragraph of the relevant H2/H3 section, then the FAQPage schema. Measure citation rate after 4–8 weeks.
Rewriting page titles and meta descriptions to “optimise for AI.” Title tags are not how AI Overviews select citations. Structured content and topical authority are. Title rewrites are a distraction from the actual problem.
Amsive’s post-update analysis documented major health publishers losing over 20 visibility points after December 2025. Notably, Mayo Clinic and Cleveland Clinic also dropped briefly before recovering — suggesting Google refined how it attributes authority rather than stripping it from established institutions entirely. Sites relying on brand name recognition without on-page experience signals (author credentials, first-person account language, sourced data) took longer to recover. Some didn’t recover at all.
The practical challenge, especially for agencies: E-E-A-T signals are scattered across three teams that almost never coordinate. Technical infrastructure (schema, author entity markup). Content (author bios, disclosure language, source attribution). Off-site presence (author bylines elsewhere, brand mentions, LinkedIn activity). Most clients have some of each, none working together.
Author schema implementation requires coordination between content, dev, and CMS admin — usually three teams with three separate sprint cycles. Build this into the retainer scope explicitly, or it won’t happen. A schema pointing to a LinkedIn with 12 connections is not an authority signal. The schema is a pointer. The signal has to exist off-site first.
Treat E-E-A-T as a product sprint. Audit current state across all three layers (schema, on-page signals, off-site presence). Identify the weakest pillar. Close that gap first. Not a one-time content project — an ongoing infrastructure build that belongs in retainer scope from day one.
Interaction to Next Paint replaced First Input Delay as a core metric in March 2024. By 2026, it’s the hardest Core Web Vital to optimise — unlike LCP (a single load-time event), INP measures the full latency of every user interaction throughout a visit. One slow click handler on a modal that 80% of users never open still tanks your score. Google’s threshold is 200ms for a “good” result.
For agencies managing WordPress sites with 40+ plugins: start with the chat widget and the consent banner. Those two alone cause a significant share of INP failures on marketing sites. INP diagnosis requires browser DevTools Performance recording — that’s not a tool purchase question, it’s a skills gap question for your technical SEO team.
It’s two years out of date. You’re measuring something Google stopped counting in March 2024. CrUX in Search Console is the most reliable source for field INP data, but lags approximately 28 days. PageSpeed Insights provides lab INP via TBT as a proxy. These are different measurements — don’t treat them interchangeably.
Schema still matters — it just matters more as a machine-readability signal for AI systems than as a rich-result trigger. The relevant types for most marketing content: Article (with author and datePublished), FAQPage, HowTo, and Organization with a proper SameAs array linking to LinkedIn, Crunchbase, and Wikipedia where applicable.
FAQPage schema is the most underused type in the stack. It structures content exactly the way AI retrieval systems prefer — question/answer pairs with clean attribution. A page with properly implemented FAQPage schema isn’t just answering user questions. It’s handing the answer engine a pre-formatted citation package.
Validate every page’s schema via Google’s Rich Results Test and the Schema.org validator. Both free. Both consistently underused. If your H3 sections are phrased as questions and you don’t have FAQPage schema on them, you’re leaving citation candidates on the table.
The December 2025 update hit sites behaving like general-interest news hubs harder than those with defined topic ownership. Consistent internal linking, logical content clusters, no gaps in coverage of related subtopics — this architecture outperforms a site ranking for 200 disconnected keywords. The pattern is consistent across every major update since 2023.
Wikipedia — the most-cited source in AI Overviews globally with over 1.1 million mentions — lost 8% of human pageviews in 2025 (Wikimedia Foundation annual report). Authority can increase AI citation share while traffic declines simultaneously. For brands that sell things, citation may still drive commercial-intent searches even when informational traffic shrinks. The evidence doesn’t resolve this cleanly yet. Anyone who tells you it does is overreading the data.
Generative Engine Optimisation is the practice of structuring content so AI systems — ChatGPT, Perplexity, Google AI Overviews, Claude — retrieve and cite it in generated answers. It shares approximately 70% of SEO’s technical work: structured content, authoritative sourcing, clean schema, topic depth. The divergences are real.
ChatGPT processes 2.5 billion prompts per day (OpenAI platform figures via Frase.io, March 2026 — directional, not independently audited). AI-referred sessions grew 527% year-over-year in the first five months of 2025 (Previsible’s 2025 AI Traffic Report). These aren’t the majority traffic sources yet — but they’ll be first for research, product comparison, and B2B vendor evaluation queries.
Early research from Princeton and IIT Delhi suggests content with verifiable statistics and named citations achieves meaningfully higher AI retrieval rates. That research is promising but production evidence is still thin. Treat specific GEO tactic choices as directional, not proven. Platforms change retrieval behaviour faster than researchers can study it — and Authoritas tracking found ~70% of AI Overview citations rotate over 2–3 months. A citation today is not guaranteed tomorrow.
“Zero-click searches rose from 56% to 69% of all queries between May 2024 and May 2025. The traffic SEO built is being absorbed by AI answers. The citation is the new click.”
Editorial synthesis — Similarweb zero-click research (2025), Seer Interactive AIO CTR study (November 2025)03 / DECEMBER 2025 UPDATE LESSONSWhat we learned — and one thing the post-mortems got wrong
The December 2025 core update (December 11–29, 18 days to complete) surfaced clear winners: eCommerce in Apparel, Retailers, Real Estate, Sports & Fitness. And a clear loser: Wikipedia dropped 435 visibility points — the biggest absolute loss of the update — while simultaneously being the most-cited source in AI Overviews globally. That contradiction is real and worth sitting with.
The pattern that holds consistently: pages satisfying transactional or navigational intent hold or gain. Pages trying to serve every intent simultaneously lose. Google is getting demonstrably better at matching intent to page type, and content designed to rank for everything tends to serve nothing well enough to survive that evaluation.
What the recovery timeline literature gets wrong
Multiple sources cite 3–6 months for substantial recovery after a negative core update hit. That’s probably right — for content-quality improvements alone. It’s wrong for cases where technical issues were amplifying the content problem.
An INP score of 600ms combined with thin content doesn’t recover in 3 months once you fix only the INP. The content still needs rebuilding. And fixing both simultaneously creates false recovery readings in GSC because you can’t isolate which change drove improvement.
Three data streams converge on one compound strategy. Seer’s data shows AI-cited brands recover 35%+ CTR versus non-cited. Early GEO research suggests content with verifiable statistics and named citations achieves significantly higher AI retrieval rates. The February 2026 Discover update specifically rewarded in-depth, original, timely content from sites with demonstrated expertise. Together: content that is fact-dense, citation-rich, structurally clean, and author-attributed performs across traditional ranking, AI citation, and Discover simultaneously. One investment, three distribution channels.
04 / THE SEQUENCING FAILURE“You’re trying to measure a cleaner engine through a dirty sensor.”
A senior technical SEO director at a mid-size B2B agency described this in a retainer debrief (shared on condition of anonymity). Site gets hit by a core update. Content team does a quality audit and rewrites 60 pages. Dev is three sprints out from the INP fix and the consent banner rebuild. Four months later, GSC shows minimal recovery. Content team blames the algorithm. Dev team blames the content. Nobody looks at the sequence.
The mechanism: Google’s quality assessment systems evaluate content in context of the full page experience — including Core Web Vitals, crawlability, and real-user behaviour signals. An INP score of 600ms on a page with genuinely good content doesn’t just slightly reduce ranking potential. It introduces quality signal drag that may prevent content quality gains from registering accurately.
“An INP failure doesn’t just slow your site. It introduces signal noise that prevents content quality gains from registering accurately in Google’s evaluation. You’re trying to measure a cleaner engine through a dirty sensor.”
Editorial synthesis — Google Search Central CWV documentation (2024), NitroPack INP guide (2026), Amsive December 2025 core update analysisThe correct sequence
05 / EVIDENCE LEVELS BY TACTICWhat the data actually supports — and what it doesn’t yet
| Tactic | Evidence Level | Time to Impact | ⚠ Adversarial Note |
|---|---|---|---|
| AI citation optimisation (answer-first section structure, FAQPage schema) | Moderate Seer data strong; AI selection mechanism partially opaque |
4–12 weeks per page | Authoritas found ~70% of AIO citations rotate over 2–3 months. Citation today is not guaranteed tomorrow. |
| INP optimisation (interaction latency, third-party script audit) | Strong Google’s CWV thresholds are published and verified |
Immediate to 4 weeks | INP gains don’t produce visible ranking lift in isolation. They remove drag on content quality signals. Hard to attribute directly in reporting. |
| Topical cluster expansion (pillar + supporting content, internal linking) | Strong Consistent across 2023–2025 update cycles |
3–6 months per cluster | Sites with large legacy content inventories face a cleanup cost that can exceed the build cost. Thin legacy pages suppress cluster authority. |
| Author schema + E-E-A-T infrastructure | Directional Correlation with December 2025 recovery, not controlled |
1–3 months to implement; signal build takes longer | Author schema alone does nothing without off-site author presence to corroborate it. Schema pointing to a LinkedIn with 12 connections is not an authority signal. |
| GEO content restructuring (direct-answer leads, stat density, citation hygiene) | Directional Early academic research; thin production evidence |
Unknown — AI training and retrieval cycles vary by platform | ChatGPT, Perplexity, and Google AI Overviews use different retrieval architectures. Platform-specific testing methodology is not yet mature. |
Sources: Seer Interactive (Nov 2025) · Amsive (Feb 2026) · Authoritas AIO study (2025) · NitroPack CWV guide (2026) · Google Search Central documentation. Evidence levels: Strong = consistent findings across multiple sources or published Google threshold; Moderate = solid primary source, mechanism partially opaque; Directional = promising but limited production evidence.
06 / ROLE-SPECIFIC GUIDANCEWhat this means for your actual job this quarter
Reframe the KPI conversation before the March update report lands. In the next 3–5 weeks, the March 2026 core update finishes rolling out, some clients see ranking changes, and they ask what happened. If your reporting still shows organic traffic as the sole primary KPI, you’ll have a hard conversation without good answers.
Before the update settles, introduce AI citation share and zero-click impression share as supplementary metrics alongside traffic. Frame it as: “Traffic is still the business outcome. These new metrics explain what’s driving changes in traffic that were previously invisible.” Don’t replace traffic metrics. Contextualise them.
Stop doing this: Framing AI Overviews as a temporary disruption that will stabilise. The Seer data is 15 months of consistent trend. The Ahrefs comparison spans two years. This is the new baseline. Clients who understand that early make better investment decisions than those surprised by it in Q3.
The content audit you actually need to run this quarter — not a full site audit. Pull your top 30 informational pages by impression count from GSC. For each, note whether an AI Overview appears for the primary query (check manually or via Ahrefs’ SERP features filter). For pages where an AIO appears, check whether you’re cited. That gives you a prioritised list in about 2 hours.
Pages where you rank top 3 but aren’t cited in the AIO are your highest-ROI revision targets. The content exists — Google trusts it enough to rank it. The AI system isn’t pulling from it, which usually means the answer isn’t structured cleanly enough. Fix: add a direct-answer H3 within the relevant section, leading with the answer in the first sentence, then FAQPage schema. Measure citation rate after 4–8 weeks.
Stop doing this: Publishing new content to compete with existing content on the same queries. If you have a page ranking position 2–5 on a target query, a second page on the same topic cannibalises the first and confuses Google’s intent mapping. Revise the existing page. Update the date. Don’t publish a second article.
The 12–18 month attribution problem your CFO doesn’t know about yet. AI Overviews reduce click-through rates by 34.5% on average for searches where they appear. A piece cited in an AI Overview generates brand influence at scale — appearing in front of users who don’t click — while showing traffic decline in Search Console. Teams cutting content budgets based on post-update click data may be defunding their most-cited, highest-influence assets.
Add AI Overview inclusion rate to your measurement stack before making any budget cuts. Then set a specific expectation: there’s a 12–18 month lag between topic-cluster content investment and measurable authority signals. Budget decisions made in Q1 2026 won’t show AI citation ROI until mid-to-late 2027. Build that projection into the initial budget justification — before the first quarterly attribution cycle makes the numbers look flat and someone makes the wrong call.
07 / WHAT I DON’T KNOW YETAnd neither should you, honestly
The March 2026 core update is still in its first weeks. No one has clean data. Anyone publishing “March 2026 update analysis” today is either republishing Google’s announcement or speculating. What’s known: it started rolling out in late March, typically takes 2–4 weeks, Google said it’s designed to surface relevant and satisfying content, and the announcement gave less context than usual. That’s it.
GEO is also genuinely still emerging. The production evidence for specific tactics is thin. Platforms change retrieval behaviour faster than researchers can study it — Authoritas tracking found ~70% of AIO citations rotate over 2–3 months. The current evidence doesn’t resolve how durable any specific GEO tactic is. That’s worth knowing before you build a budget around it.
Whoever tells you they have the definitive 2026 SEO playbook for AI search is working from incomplete data and calling it certainty. The honest position: here’s what the data shows, here’s where it’s solid, here’s where it’s directional, and here’s what’s genuinely unknown. That’s the analysis that holds up when the next update hits in Q2.
08 / REFERENCESSources, with confidence levels
Every figure above has a named source and a disclosed confidence level. Strong = consistent findings across multiple sources or published Google threshold. Moderate = solid primary source, mechanism partially opaque. Directional = promising but limited production evidence.
