January's forecasts have now had six months to meet reality. Some were right, several were badly wrong, and the most useful reading of mid-2026 comes from grading them rather than writing new ones.
The scorecard
Every December the industry publishes its predictions. Almost nobody goes back in July to check them. Here's that check, against what the first half of 2026 actually produced.
| The prediction | What actually happened | Verdict |
|---|---|---|
| "SEO is dead" | Search volume grew; Google still sends enormous absolute click volume. The per-page click rate fell sharply | Wrong, but pointing at something real |
| Zero-click will keep climbing | Around 68% of US searches ended without a click in early 2026, up from roughly 60% in 2024 | Correct |
| Ranking #1 will stop mattering | Position-one CTR drops sharply when an AI answer appears, and fewer AI citations now come from the top ten | Half right — necessary, no longer sufficient |
| A rival engine will take share from Google | Google absorbed the threat instead. AI Mode passed a billion monthly users, roughly ten times a year earlier | Wrong |
| Core updates will matter less | Two broad core updates plus two spam updates in six months, with real ranking consequences | Wrong |
| Small sites are finished | Referral losses were far heavier for weakly recognised sites than strongly recognised ones — a brand gap, not a size gap | Directionally right, wrongly diagnosed |
The pattern in that table is worth naming. The predictions that failed were the dramatic ones — death, replacement, obsolescence. The predictions that held were the boring, incremental ones. That has been true of SEO forecasting for a decade and 2026 did nothing to change it.
What actually happened, in order
A quick chronology, because a great deal of commentary this year has collapsed six months of separate events into one undifferentiated sense of upheaval.
March 27 – April 8: the first core update. A twelve-day rollout. The most consistent finding from analyst post-mortems was a shift in visibility away from intermediaries — aggregators, directories, comparison sites, some user-generated content — toward what several described as destination sources: the brands, institutions and specialists who originate information rather than repackage it. A March spam update ran alongside it, finishing in under a day.
That distinction deserves more weight than it received. A directory that lists twenty providers, a comparison site that summarises twenty reviews, and a roundup post that aggregates twenty opinions were all doing roughly the same job — collecting other people's information and presenting it conveniently. That job is now done better, faster and for free by an AI answer. The March update looks less like a penalty on aggregators and more like Google acknowledging that the middleman layer of the web has been made redundant by its own product.
May 19–20: Google I/O. The largest single announcement set of the year. AI Mode passed a billion monthly active users, roughly ten times its figure twelve months earlier, with AI Overviews reaching a substantially larger audience again. Google redesigned the search box itself — described by its own search leadership as the biggest change to it in twenty-five years — to accept text, images, files and video, and to anticipate intent rather than match keywords. Autonomous information agents were announced for the summer: services that monitor a topic between sessions and push synthesised updates to the user, generating no click and no session along the way.
May 21 – June 2: the second core update. Landing two days after I/O, this was the worst possible timing for anyone trying to diagnose a traffic change. Rankings moved for twelve days while the search interface was simultaneously being rebuilt underneath them. A great many teams attributed algorithmic volatility to interface changes, or the reverse.
June 24–26: a spam update. Not a core update, though plenty of reporting conflated the two. Google framed it as a routine global refresh of existing detection systems rather than a new policy, and specifically indicated it was not aimed at link spam or site reputation abuse.
July: refinement. No broad core update. The month's activity was in the results page rather than the algorithm — AI Overviews gained a live news carousel, generated imagery, and new citation displays, while Search Console gained controls over how content appears inside AI features. Quieter, but arguably more consequential than another ranking shuffle.
Those Search Console controls are the item most worth acting on. Site owners have spent two years asking for some say in how their content is used inside AI surfaces, and this is the first meaningful answer. Whatever you decide about participation, the decision should be deliberate rather than a default nobody in your organisation has looked at.
The number that actually matters
Strip away the announcements and one measurement defines the year. A Similarweb and SparkToro study published in June found that roughly 68% of US Google searches in the first four months of 2026 ended without a click — up from about 60% in 2024.
The follow-through is the part worth sitting with. Of the roughly third of searches that do produce a click, about two-thirds go to the open web, a little over a quarter stay inside Google's own properties, and the remainder go to paid results. Run that arithmetic and something like 276 clicks per thousand searches reach the open web at all.
The real headline Search didn't shrink. The open web's share of it did. Those are very different problems, and only one of them is yours to solve.
This reframes the "is SEO dead" argument that dominated commentary this year. Search demand is healthy. What has contracted is the proportion of that demand which arrives at an independent website — the dynamic we've traced through zero-click content and what it does to success metrics, and in more detail in our look at what AI Mode means for organic traffic.
What didn't change
This section is missing from almost every state-of-SEO piece published this year, which is a shame, because it's where most teams should be spending most of their time.
Technical fundamentals still decide eligibility. Crawlability, indexability, site structure, page speed, clean internal linking, valid structured data. None of this became less important because AI answers arrived — if anything the opposite, since a page that can't be crawled and parsed cleanly can't be cited either. Google's own guidance on appearing in AI features has been consistent and slightly boring: there is no special technical trick beyond doing the standard work properly.
Search intent still governs everything. The redesigned search box handles longer, more conversational, more multimodal queries, but it is still trying to work out what someone wants. Pages built around a genuine understanding of intent still outperform pages built around a keyword, exactly as they did in 2019.
Content quality still separates winners from losers within a core update. The March update's preference for originators over aggregators is not a new principle. It's the same quality bar Google has been raising incrementally for years, applied with more force.
The basics still produce most of the results for most sites. If your on-page SEO is inconsistent, your titles are duplicated, and your site architecture is a decade of accumulated compromise, then citation-share tracking is not your highest-value next action. Fix the foundation first. The teams that panicked into AI-search projects while leaving broken fundamentals in place have had a poor year.
And links still function roughly as they always have. Predictions that AI search would render link building obsolete have not materialised at the halfway mark. Links remain a substantial input to how Google evaluates sources, and the sources AI answers draw on are largely the ones that rank — which means link equity feeds citation eligibility indirectly. What has changed is the relative value of a mention without a link, since entity recognition now matters independently. Both are worth pursuing; neither has been replaced.
The three things that genuinely are new
Having defended continuity, here is what has actually changed, and it is more than cosmetic.
1. Citation share is now a separate outcome from ranking
This is the substantive shift. Ahrefs benchmarks suggest position-one click-through rate falls by roughly a third when an AI Overview sits above it, and analyses through 2026 indicate a shrinking share of AI Overview citations are drawn from top-ten organic results — a marked change from a year earlier. You can rank first and be absent from the answer that most users read.
There is an upside hidden in the architecture, though, and it gets almost no attention. AI Mode decomposes a single user prompt into many simultaneous sub-queries before assembling an answer — reportedly more than a dozen in some cases. Each of those is a separate opportunity to be cited. A site that is the definitive source on one narrow sub-question can appear in an answer to a broad question it would never have ranked for. That's a genuinely new route to visibility for specialist sites, and the opposite of the doom framing most coverage applies to it.
Citations are also unstable: research reported this year suggests a large majority of pages cited in AI Overviews change citation status within two to three months. That makes citation share something to monitor on a recurring cadence rather than audit annually — the practical core of generative engine optimisation, and the thread running through what's been working with AI Overviews all year.
2. Brand recognition became a measurable ranking defence
The most quietly important finding of 2026. Analysis published after I/O suggested strongly recognised sites lost around 22% of their search referrals over a two-year period, while weakly recognised sites lost closer to 60%. Separately, branded queries appear to earn meaningfully higher click-through under AI Overviews rather than lower, and brands cited inside an AI answer appear to earn substantially more clicks than uncited competitors on the same query.
Treat those figures as directional rather than precise — methodologies vary and most come from agency research rather than peer review. But the direction is consistent across independent sources, and the implication is uncomfortable for anyone who has treated brand as a soft cost centre. Brand trust in an era of AI-generated everything has become a technical SEO concern.
3. Measurement broke in a way that can't be patched
Information agents don't generate sessions. Multi-turn AI research stays inside Google. AI-mediated purchase flows can remove the website visit entirely. There is no UTM parameter for a conversation.
The consequence isn't that analytics is worthless — it's that last-click decision-making is structurally broken for anything AI search touches, which is the same conclusion arriving from every direction in the wider attribution problem. Impressions can now rise while clicks fall, which makes impression-based reporting actively misleading. The practical response is unglamorous: report on visits, conversions and revenue; segment Search Console by query type; and stop treating rank tracking as a proxy for business outcome.
What to do in the second half
Five actions, in rough priority order for a team that hasn't yet responded to any of this.
- Establish a citation baseline. Run your top twenty target queries through the major AI surfaces. Record whether you appear, how you're described, who appears beside you, and which sources get cited. You cannot improve what you have never measured.
- Separate your reporting by query type. Informational queries are where AI answers bite hardest; transactional and navigational queries behave much as before. A blended traffic number hides which of your problems is real.
- Audit the fundamentals before the frontier. Crawl the site. Fix what the crawl finds. This remains the highest expected-value work for the majority of sites, and it is a prerequisite for citation anyway — a point that extends to designing sites that AI can actually read.
- Publish the material a model can't invent. Original data, named case studies, first-hand testing, genuine expertise. Content that merely rephrases what already ranks gives an AI nothing to cite and everything to summarise. A deliberate content programme built on real expertise is now the input to citation, not just to ranking.
- Build owned channels. Email, community, direct and branded search. Every interface change this year has moved value toward audiences you reach without an intermediary.
The honest summary
Six months in, SEO in 2026 looks less like a discipline being destroyed and more like one being narrowed. The easy traffic — definitional queries, thin how-tos, aggregated summaries — is being absorbed by the answer layer and is not coming back. What remains is harder to earn and more defensible once earned: original material, recognised brands, genuine expertise, and audiences you own.
That's not a comfortable transition for anyone whose model depended on the easy half. It is, though, a reasonably good outcome for anyone who was doing the difficult version all along.
Not sure whether your traffic drop is a ranking problem or an AI problem?
We diagnose what actually changed and build SEO programmes for how search works now.
Explore Search Engine Optimization →Frequently asked questions
Is SEO dead in 2026?
No, but the job has changed shape. Search volume is higher than ever and Google still sends enormous absolute click volume to the web. What has fallen is the share of searches that produce a visit to any given site — a Similarweb and SparkToro study published in June 2026 found around 68% of US Google searches in the first four months of the year ended without a click, up from roughly 60% in 2024. The discipline has shifted from earning rankings to earning citations, visits, and brand recall, and the sites suffering most are those that relied on being the first blue link for questions an AI can now answer directly.
How many Google core updates were there in the first half of 2026?
Two confirmed broad core updates. The March 2026 core update ran from 27 March to 8 April, and the May 2026 core update rolled out from 21 May to 2 June, completing in just under twelve days. Google also ran two spam updates — one in March that finished in under 24 hours, and one in late June. July brought no broad core update, making it a month of refinement rather than upheaval. Analysts noted the March update in particular favoured original and official sources over aggregators and comparison sites.
Does ranking number one still matter in 2026?
It matters less than it did and still matters more than the loudest commentary suggests. Ahrefs benchmarks indicate position one click-through rate drops by roughly a third when an AI Overview appears above it, and analyses through 2026 suggest a shrinking proportion of AI Overview citations now come from top-ten organic results. But ranking remains a strong input to being cited, and for the many queries where no AI answer appears, position one behaves much as it always has. The accurate framing is that ranking is now necessary but not sufficient.
What changed for SEO at Google I/O 2026?
Google announced that AI Mode had passed a billion monthly active users, roughly ten times its figure a year earlier, with AI Overviews reaching a broader audience still. It also introduced a redesigned search box built to accept text, images, files and video and to anticipate intent, plus autonomous information agents that monitor topics between sessions and push updates without generating a website visit. A core update followed two days later, making ranking changes through late May unusually hard to attribute to any single cause.
What should SEO teams focus on for the rest of 2026?
Three things. Start measuring citation share — how often and how favourably your brand appears inside AI answers for your target queries — because it is now a distinct outcome from ranking. Invest in brand, since research through 2026 suggests strongly recognised sites lost far less search referral traffic than weakly recognised ones over the same period. And rebuild measurement around visits, conversions and pipeline rather than impressions, which can now rise while clicks fall. The technical fundamentals still work and should not be abandoned while chasing the new disciplines.
