Which 2026 Search Trends Are Real? A Sorted Verdict on Every Big Claim

By Matt Griffin, Founder, Formative Digital · Published 2026-07-18. "Trend season is prediction season, and predictions are where agencies hide. Every January and July the same recycled list circulates with no dates and no sources. What follows is the opposite exercise: each claim gets a citation or gets called out for lacking one."

Type the phrase into any engine and a wall of near-identical listicles appears, each numbering ten to fifteen "trends you can't ignore." Reading five of them back to back, a pattern emerges: the same assertions, rarely a publication date, almost never a primary source. So this page runs a different procedure. We collected the claims that dominate the 2026 trend conversation and sorted them into three bins: verified (a dated, checkable source exists), overstated (a real signal stretched past what the data supports), and noise (no evidence located, and we say so plainly). Every verdict below names its source and its date, or names the absence.

How the sorting worked, so you can audit it

The method is deliberately boring. A claim earned verified only when a named organization published measurable data with a date, or when a court, regulator, or the engine itself put the fact on the record. A claim landed in overstated when the underlying signal is real but the popular framing exaggerates its size, timeline, or universality. Noise means we searched for supporting evidence during the week of July 13, 2026 and found none from a primary source, only vendors quoting other vendors. This maps to Vector 11 in our methodology, Measure: track what the surfaces actually do, not what the industry says they will do.

One honesty note before the verdicts. Formative Digital sells AI-search visibility work, which gives us an incentive to inflate the AI trends. The correction for that bias is the same one we apply to client reporting: dates, named sources, and numbers a stranger can re-check. Where our commercial interest and the evidence diverge, the evidence wins on this page.

Verified: AI Overview coverage roughly doubled in a year

AI Overviews appeared on 48% of tracked queries in February 2026, up from 31% twelve months earlier, per BrightEdge's year-over-year panel. Google's own public statements put the figure at roughly half of US queries, and an April 2026 Xponent21 measurement came in higher still at 60.32%. Semrush's broader keyword panel reads much lower, 15-25%, because its query mix skews short and commercial. The spread between studies is itself the useful lesson: coverage depends heavily on query type. Informational, question-shaped searches trigger Overviews constantly; navigational and transactional searches mostly do not.

Roll the trajectory forward and the projection of 70-80% coverage by the end of 2026 stops sounding like vendor talk. The compounding fact underneath: since the May 21, 2026 core update, the content signals that earn a top-ten ranking and the ones that earn an Overview citation have converged into a single quality bar. Our AI search statistics roundup tracks these coverage numbers with their methodology notes as new panels publish.

Verified: the query pie now has more than one slice

ChatGPT reached an estimated 17% of global digital queries in 2026, processing about 2.5 billion prompts a day, while Google holds roughly 80%. Perplexity, Gemini, and Copilot split most of the remainder. That is a genuine structural change: for the first two decades of the industry, "search" and "Google" were interchangeable words, and in 2026 they no longer are. The full comparison of how each engine selects and cites sources lives in our AI search versus Google breakdown.

What makes this verified rather than hype is that the engines disagree with each other in measurable ways. Ask the same question of Google, ChatGPT, Perplexity, and Gemini and you get different citations drawn from different corpora with different freshness windows. A business can be well described on one surface and invisible or misdescribed on another, which is precisely why our audits query all of them rather than sampling one.

Verified: the US v. Google remedies are in effect, and they matter locally

On April 25, 2026, the US District Court for the District of Columbia issued its remedies ruling in United States v. Google. No breakup: the court rejected divestiture of Chrome and Android. What it ordered instead is arguably more interesting for search competition. Google must share aggregated search index and user-interaction data, the click and hover signals that built its quality moat, with "Qualified Competitors," and must offer search and ads syndication on ordinary commercial terms, capped at 40% of a competitor's queries in year one. Exclusive default-placement contracts for Search, Chrome, Assistant, and the Gemini app are prohibited. The DOJ has cross-appealed for stronger relief, with arguments projected for late 2026 or early 2027, so the framework may still tighten.

Why a Brantford business should care about a Washington courtroom: syndicated index access means challenger engines can improve much faster than they could when Google's interaction data was sealed. Every engine that gets better at answering local queries is another surface where your entity data either checks out or does not. We unpack the local-search implications in detail in our analysis of US v. Google and local SEO.

Overstated: "organic traffic is collapsing"

The kernel of truth is real and quantified: when an AI Overview appears, click-through on the top organic result drops by an average of 34.5% according to studies circulating through 2025 and 2026, and informational publishers have absorbed genuine losses. The exaggeration is in the word "collapsing" applied to all organic traffic. Overviews concentrate on informational queries; commercial, local, and navigational searches, the ones that produce customers rather than readers, still route users outward. And pages cited inside an Overview retain visibility that uncited pages lose, which changes the objective from "rank and get the click" to "be the source the answer is built from."

The evidence-check verdict: a redistribution, not a collapse. Traffic is shifting from generic informational pages toward cited sources and toward direct answer surfaces. Businesses whose model was monetizing shallow how-to pageviews are in real trouble. Businesses that sell an actual product or service in an actual place are dealing with a changed funnel, not a dying one.

Overstated: "ChatGPT is replacing Google"

Two clickstream findings puncture this one. First, across large referral panels in 2026, AI assistants collectively drive about 0.3% of tracked referral traffic while Google alone drives 26.4%. Whatever users do inside ChatGPT, it is not yet sending websites meaningful volume relative to search. Second, and more telling: Semrush's 17-month clickstream study found that the share of ChatGPT's own outbound referrals pointing at google.com climbed past 21% by early 2026. A fifth of the time, the destination after an AI answer is the incumbent search engine, because users verify what the model told them.

So the honest 2026 model is layered, not substitutional. AI engines increasingly shape which brands a user considers; Google still carries most of the confirmation and transaction traffic. Ignoring either layer is a planning error. We laid out where this layering likely goes next in our future-of-SEO analysis, written before the remedies ruling and holding up reasonably well against it.

Noise: three claims that failed the source check

llms.txt as a ranking lever. The proposal exists and some sites have adopted the file, but as of July 2026 we could locate no confirmation from Google, OpenAI, Anthropic, or Perplexity that any production crawler consults it for retrieval or citation decisions. Google's John Mueller publicly compared it to the keywords meta tag in 2025. Absence of evidence, named: no engine has documented honouring it. Harmless to publish, dishonest to sell.

"Google penalizes AI content." Google's published spam policy targets scaled content abuse, defined by intent and value, and explicitly not by production method. The 4,237 templated pages this very domain once carried were suppressed for being templated at volume, not for being machine-written; thin human-written content meets the same fate. The trend claim inverts the actual mechanism.

One-click GEO tools. A wave of 2026 SaaS products promise AI-visibility rankings from a URL paste. None we examined publish their sampling methodology, and none disclose that model outputs vary run to run, which makes a single-sample "score" close to meaningless. Measurement across engines is real work, Vector 11 work, done with repeated sampling and dated snapshots, not a dashboard demo.

What actually changed inside the ranking systems this year

Two core updates shipped in the first half of 2026, the second landing May 21, a faster cadence than the historical three-to-four-month gap. The May update matters because it runs Gemini-based quality models against the page corpus, and because it tightened the coupling between conventional rankings and Overview citations. Industry tracking through 2025 had shown the overlap between top-ten results and Overview citations falling as low as 17-38% from an earlier 76%; the post-update pattern pulls the two back toward one standard. In practice that ends the brief period where "GEO" and "SEO" could be sold as separate services. There is one content discipline now, and it is graded by classifiers that read the whole site, not one page.

The site-level grading deserves emphasis because it is the least discussed and most consequential change. A cluster of templated pages drags down the pages that deserve to rank; we carry the scar tissue to prove it, and cleaning up that legacy is ongoing work on this domain. Radical transparency cuts both ways, so there it is.

What an Ontario business should actually do before 2027

Strip the trend list down to actions and five remain, in order of return.

1. Make your entity data agree with itself everywhere. Business name, address, phone, and hours identical across your site's schema, your Google Business Profile, and your directory citations. Every engine that gained index access after the April remedies ruling is another parser cross-checking these fields. The minimum viable markup for most businesses fits in a dozen lines:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "LocalBusiness",
  "@id": "https://www.example.ca/#business",
  "name": "Example Business",
  "url": "https://www.example.ca/",
  "telephone": "+15195550100",
  "address": {
    "@type": "PostalAddress",
    "streetAddress": "100 Example St",
    "addressLocality": "Brantford",
    "addressRegion": "ON",
    "postalCode": "N3T 0A0",
    "addressCountry": "CA"
  },
  "sameAs": [
    "https://www.google.com/maps?cid=YOUR_CID",
    "https://www.facebook.com/examplebusiness"
  ]
}
</script>

2. Ask the four engines about yourself, this month, and write down the answers. Google, ChatGPT, Perplexity, Gemini. Note every hallucinated hour, dead competitor mention, and missing service. That log is your baseline; without it you cannot tell in December whether anything you did worked.

3. Publish fewer pages with more evidence in each. The 2026 classifiers reward pages that contain something the top five results do not: your pricing logic, your before-and-after data, your first-hand answers to the questions customers phone in. One page like that outperforms ten summaries.

4. Keep the informational-traffic expectation honest. If part of your traffic is how-to content, expect the 34.5%-range click erosion on Overview-covered queries and measure whether citations inside those Overviews are compensating with branded search lift.

5. Ignore the noise bin entirely. No llms.txt projects, no AI-content panic, no single-score visibility tools. Budget spent there is budget taken from the four items above.

Where does an agency fit? Only where the arithmetic favours it. A business owner can do items one and two in a weekend with our research library as the manual. Items three through five compound month over month, which is where our production system earns its keep, and the engagement terms include the Results Guarantee: if your existing domain shows no measurable organic search results after 12 months of work with Formative Digital, we work for free until you see them. Results depend on industry, competition, and existing digital presence, which is exactly why the guarantee exists in writing rather than as a slogan.