The year in review

What changed in search between September 2025 and September 2026

Twelve months produced more structural change in search than the previous five years combined. Two broad core updates, four spam updates, a rebuilt Google search box, a second major search ad auction, and the first native reporting on AI visibility all landed inside one year. The timeline below is the short version, and every date on it is publicly documented.

One thing worth saying out loud about core updates

Google shipped only two broad core updates in all of 2026, in March and in May. As of September 1 there was no September core update announced, and most of the year's volatility came from spam updates and from AI surface expansion rather than from changes to core ranking. Teams that blamed every traffic drop on a core update spent the year fixing the wrong thing.

The mechanism

Why clicks disappeared even though searching went up

Clicks disappeared because the answer moved to the results page. When Google renders an AI Overview, the searcher gets a synthesized answer at the top of the page, and the incentive to visit a website in order to find that answer goes away. Pew Research Center measured organic click rate at 8 percent when a summary is present versus 15 percent when it is not, and measured clicks on links inside the summary at 1 percent.

The size of the effect has been measured independently several times, which matters because a single vendor number is easy to dismiss. Ahrefs measured a 58 percent click-through drop for position one on queries with an AI Overview in February 2026, revised upward from 34.5 percent in April 2025. Researchers at ISB and Carnegie Mellon ran a randomized field experiment in 2026 and measured a 38 percent drop in organic clicks. Pew found that 26 percent of sessions ended entirely after an AI summary search, compared with 16 percent without one.

Brand queries behave differently, and this is the most useful nuance in the dataset. Amsive analyzed 700,000 keywords in April 2025 and found brand-query click-through rate rose 18.68 percent when an AI Overview appeared, while non-brand queries fell 37 percent. People who already know your name still come to you. People discovering a category for the first time increasingly never leave the results page.

The rebound is real, and it is not a recovery

Click-through rate on AI Overview queries has partially recovered as Google tunes link placement, and that recovery should not be mistaken for a return to normal. Seer Interactive tracked AI Overview query click-through rate rising from 1.3 percent in December 2025 to 2.4 percent by February 2026. The floor fell hard, then came up slightly.

The honest read is that AI surfaces are still being adjusted month to month, which is exactly why measurement has to be continuous rather than annual. A number captured in December described a world that no longer existed in February. Any strategy built on a single snapshot is already out of date.

Coverage

How often AI answers appear on the queries that actually matter

AI answers appear most often on exactly the searches that precede a purchase. Peec AI analyzed 500,000 business-intent prompts between April 13 and 20, 2026, in one of the largest coverage studies published to date, and found AI Overviews on 86.7 percent of business-intent searches and 88.5 percent of decision-stage queries. Coverage on fact-style queries such as definitions, comparisons, and instructions runs between 93 and 97 percent.

The length pattern in that table is the part to sit with. Coverage climbs from 64.6 percent on two-word queries to 89.1 percent on queries of eleven to fifteen words, which means the longer and more conversational the question, the more certain it is that a machine answers it. Buyers ask longer questions when they are close to deciding. Coverage is highest precisely where the money is.

Buyer behavior

How buyers actually make decisions now

Buyers now form their shortlist inside a system you do not control. Forrester surveyed nearly 18,000 global business buyers and published in January 2026 that 94 percent used AI during their most recent purchase process, up from 89 percent the year before, and that AI answer engines had become the single most meaningful vendor research source, ranking ahead of vendor websites, sales representatives, and product experts.

The use cases are specific rather than vague. Forrester found 55 percent of buyers use AI to compare vendors, 54 percent to research products, and 47 percent to build the internal business case that gets a purchase approved. Gartner's March 2026 survey put generative AI use in a recent purchase at 45 percent, and found that buyers now consult an average of seven information sources before deciding.

Forrester also documented what this does to websites. B2B companies reported traffic declines of 10 to 40 percent as buyer research moved into answer engines during 2026. The traffic did not go to a competitor. It stopped existing as traffic.

Shift one: the shortlist gets built where you cannot see it

The shortlist now forms inside a conversation you never observe. A buyer asks an assistant to compare vendors in your category, the assistant names three, and if you are not one of the three there is no retargeting pixel, no form fill, no attribution record, and no second chance at that buyer. The loss never appears in analytics because it never touched your site.

This is the single hardest thing to explain to a marketing team that measures everything, because the metric for it does not exist in the dashboard. A drop in qualified pipeline with flat website metrics is often this. It looks like a sales problem and it is a visibility problem.

Shift two: buyers arrive informed and confidently wrong

Buyers are better researched than ever and buying outcomes are getting worse at the same time. Forrester's State of Business Buying research found that 86 percent of B2B purchases stall and 81 percent of buyers end up dissatisfied with the provider they chose. Forrester also found in 2025 that 20 percent of buyers, and 28 percent of procurement professionals, felt less confident after encountering unreliable AI information.

The failure mode is confident misunderstanding: firm conclusions built on fragmented or inaccurate synthesis. For your brand, that means an inaccurate AI description is not a vanity problem. It is a lost deal that never announces itself, and the only signal you get is a sales call that starts from a false premise.

Shift three: rep-free until the moment it is not

Buyers want to do most of the process without you, then want a human to check the AI's work. Gartner's March 2026 survey found 67 percent of B2B buyers prefer a partly rep-free experience, and its May 2026 research found 69 percent still turn to a sales representative to validate AI-generated insights. Those two numbers describe one behavior, not two.

The practical consequence is that every sales conversation now starts later, with a better informed and more skeptical counterpart. Your content has to carry the first eighty percent of the selling, and it has to be accurate enough to survive the validation call that follows. Gartner also found that buyers who reach genuine clarity about value are twice as likely to report a high-quality deal, which is the argument for clarity over persuasion.

Shift four: consumers are doing the same thing, faster

Consumer behavior shifted harder and quicker than business buyer behavior. BrightLocal's 2026 survey found consumer use of AI tools for local business recommendations went from 6 percent to 45 percent in a single year, which is the fastest behavior change anywhere in this dataset. The consumer version of the shortlist problem is a person asking their phone which dentist to call.

Same mechanics, shorter cycle, and far less patience. A homeowner with a flooded kitchen is not comparing seven sources. They are asking one question and calling whoever gets named.

The buyer journey, remapped

The old motion versus the 2026 motion

The foundation layer

SEO in 2026: what still works

The technical foundation of SEO matters more in 2026 than it did in 2023, and it produces a different return. AI engines crawl, index, and retrieve from the open web, so a page that cannot be crawled, rendered, or indexed cannot be ranked and cannot be cited either. Google still sends 87.63 percent of all search referral traffic as of May 2026 according to Cloudflare Radar, which means the traditional layer is not optional even in a zero-click environment.

What changed is the return profile. SEO is now the cost of entry rather than the strategy, and the work that used to end at a ranking now serves two systems at once.

Server-side rendering moved from preference to requirement

AI crawlers do not execute JavaScript, which makes server-side rendering a hard requirement rather than a best practice. A client-side rendered page can rank perfectly well in Google and remain completely invisible to the engines that write recommendations, because those crawlers see an empty shell where your content should be. This is the most common and most expensive technical failure we find in audits.

Verify it before you publish anything, and verify it section by section rather than page by page. A page can be server-side rendered while one critical module, often pricing or reviews, loads client-side and disappears.

Information gain replaced content volume

The test is no longer whether content is helpful. The test is whether it contains something the model does not already have. John Mueller stated in November 2025 that Google's systems do not care whether content is created by AI or by humans, and the 2026 spam updates went after content lacking human expertise rather than content with a particular authorship. The August 2026 spam update, which ran roughly two days and sixteen hours and was the longest of the year, was tied in industry reporting to detection of mass-produced synthetic content.

Content that restates what a model already knows adds nothing, gets filtered, and increasingly gets treated as spam. Content carrying original data, firsthand experience, named expertise, and specific numbers gets cited. That distinction is now the whole game on the content side.

Light navigation beats comprehensive navigation

Heavy internal linking, which used to be an SEO virtue, now works against you with AI agents. Peec AI analyzed ChatGPT Deep Research agent logs in June 2026 and found that pages carrying 60 or more links delivered only 33 percent actual content on an agent's first read, versus 78 percent for pages with light navigation. The agent spends its budget parsing your menu instead of reading your answer.

The same research found that a successful in-page find triggered a targeted re-read 95 percent of the time. Give the agent a clean read on the first pass and it comes back for more. Bury the content under navigation and it does not.

Structure content for passage retrieval, not page ranking

Engines retrieve chunks of pages, not whole pages, which changes how a page should be written. A 3,000 word guide that buries the answer to a specific question in paragraph forty is retrievable in theory and invisible in practice. Every section needs a self-contained, quotable answer in its first two sentences.

Keep paragraphs to three or four sentences so a retrieved chunk carries one complete thought. Put the number, the source, and the date inside the same sentence, because a chunk lifted out of context has to make sense standing alone. This page is written to that rule, which is the most honest demonstration available.

The second layer

GEO: the second visibility layer

GEO, Generative Engine Optimization, is the practice of earning mentions and citations inside AI-generated answers from ChatGPT, Gemini, Perplexity, Google AI Overviews, and Google AI Mode. SEO earns a position on a page of results and measures success as a ranking and a click. GEO earns a place inside the answer itself and measures success as being named, described accurately, and recommended, whether or not a click ever follows.

These are not competing disciplines and they are not the same discipline either. They are two visibility layers that share a foundation and diverge at the top. A site that is slow, client-side rendered, thin, or unindexed fails in both. A GEO Audit measures both layers against the same domain so you can see which one is actually costing you.

Ranking number one does not get you cited

The overlap between Google's top results and AI citations is far smaller than most people assume. Ahrefs analyzed 15,000 queries in August 2025 and found only about 12 percent of URLs cited by AI tools also appeared in Google's top 10 organic results. In a larger analysis of 863,000 keywords and 4 million AI Overview URLs, the share of AI-cited pages that also ranked in the organic top 10 fell from 76 percent to 38 percent over eighteen months.

The two systems are diverging rather than converging. A rank report is therefore no longer a visibility report, and a monthly report showing only rankings is measuring one of the two places your buyers look.

The finding that favors the challenger

Generative engines flatten the competitive field, and the evidence for this is the most commercially useful research in the entire GEO literature. The Princeton-led study presented at KDD 2024 remains the only large controlled benchmark on GEO tactics, tested against a 10,000 query benchmark. It found that adding source citations produced a 115.1 percent visibility lift for a fifth-ranked site and a 30.3 percent visibility loss for a first-ranked site.

Read that pair again. The tactics that help a challenger brand most are the same tactics that help the incumbent least. If you are sitting fifth in your category and have spent a decade watching the same three competitors own the top of Google, this is the first structural opening you have been handed in years, and it closes as the category fills up.

What the controlled research says actually works

Four tactics carry measured evidence, and one popular tactic measured worse than doing nothing at all.

Princeton-led GEO study, presented at KDD 2024, 10,000 query benchmark

Most of your citations will not come from your website

The most strategically uncomfortable number in GEO is where citations actually originate. Muck Rack found in May 2026 that 84 percent of AI citations come from earned media rather than from brand-owned pages. An agency working only inside your CMS is working on roughly 16 percent of the surface area that decides whether an engine names you.

The domains engines trust are public and consistent. Peec AI analyzed 30 million cited sources in March 2026 and found Reddit ranked first or second among cited domains on all five major AI platforms, with the top ten reading Reddit, YouTube, LinkedIn, Wikipedia, Forbes, G2, Yelp, Facebook, Medium, and TechRadar. ChatGPT leans on Wikipedia, Reddit, and Forbes. Google AI Mode and AI Overviews favor YouTube, Reddit, and Facebook.

Operationally, this makes GEO a public relations and community problem before it is an on-page problem. Review velocity, forum presence, podcast appearances, third-party listings, and editorial coverage stop being nice-to-have brand activity and start being the primary citation supply.

Citation is winner-take-most, and it is unstable

A small number of URLs capture most citations and most URLs capture none, which means half-measures produce zero rather than half. Peec AI analyzed more than 1 million citations in February 2026 and found that on Perplexity, 64 percent of URLs were never cited at all while the top 6 percent captured just under half of all citations. On Google AI Mode, more than nine out of ten URLs had a citation rate below 1.0.

What gets cited today can also vanish overnight. Reddit was the most-cited domain on every major engine in March 2026, and Promptwatch reported its share of ChatGPT citations falling from roughly 3.8 percent to roughly 0.5 percent on August 14, 2026. Build presence on platforms you do not own, and never build your visibility on them.

Two things almost everybody sells you that the evidence does not support

This is the part of the page that costs us money to publish, and it is here because you deserve the finding rather than the pitch.

The four mechanical changes GEO forces on SEO practice

The paid layer

SEM in 2026: the paid search rebuild

Paid search got more valuable and less controllable in the same year. Manual control is being removed on a published schedule, a second major search ad auction opened and scaled to a billion-dollar run rate in seven months, and the buying model on that new auction moved from impressions to outcomes inside six months. Every one of those changes has a date attached and none of them is speculative.

The September 1 migration that Google announced by email

Any Google Ads account with campaign-level Broad Match or legacy Automatically Created Assets is being changed whether or not anyone logs in. Google closed creation of new campaign-level Broad Match and ACA entities on August 3, 2026, and began auto-migrating those Search campaigns to AI Max on September 1, 2026. This was communicated by email rather than by blog post, which is why most advertisers do not know it happened.

The performance change will surface in October reporting with no obvious cause. Audit every Search campaign for campaign-level Broad Match and legacy ACA settings, decide the configuration on purpose, and use the built-in one-click experiment to measure the change rather than assuming it. Dynamic Search Ads follow in February 2027, delayed from the original schedule, and standalone Display campaigns migrate to Demand Gen by January 2027.

What AI Max is, and what Google claims for it

AI Max is an optimization layer inside existing Google Search campaigns, not a new campaign type. It bundles search term matching, text customization, and final URL expansion. Google reports an average 7 percent more conversions or conversion value at similar cost per acquisition or return on ad spend when the full suite runs versus search term matching alone, and that figure is vendor-reported rather than independently verified.

Treat it as a configuration decision with a measurable outcome, not as a philosophy. The advertisers who lose money on this migration will be the ones who never checked what changed.

Why your cost per click went up

Auction pressure is concentrating into fewer, more valuable searches. As AI Overviews absorb informational and research-stage queries, exploratory traffic stops producing clicks and budget shifts toward revenue-dense bottom-funnel terms, which raises competition in exactly the auctions that matter most. WordStream and LocaliQ benchmarks put average Google Search cost per click at $2.96 in Q1 2026, up 12 percent year over year from $2.64.

Accounts weighted toward educational or research-stage traffic feel this first, which is common in B2B, professional services, and lead generation. Segment by query intent and stop paying informational prices for traffic that no longer converts. Rebuild negative keyword lists against the new search term patterns AI Max surfaces, and audit Quality Score before concluding the market alone is to blame.

The second auction: ChatGPT Ads

A new search ad channel went from zero to material in under seven months. OpenAI began testing ads in the US on February 9, 2026, and reported a $1 billion annualized revenue run rate in under 200 days on August 31, 2026, with self-serve Ads Manager available in 52 countries as of September 3, 2026. The pricing model moved fast: it launched at a $60 CPM with a $200,000 managed minimum, and by August 2026 cost-per-click and outcome-optimized bidding accounted for the majority of campaigns, with the $50,000 self-serve minimum removed.

Ads appear as a labeled block beneath the answer and are shown to free and Go tier users, with sensitive-topic guardrails retained. Perplexity took the opposite path and pulled all advertising in February 2026, citing user trust. Anthropic's Claude has no ad program as of 2026.

One caution belongs in the budget conversation. An Ipsos survey in 2026 found 63 percent of US adults say ads in AI results make them trust those results less, which means brand-sensitive categories should weigh the placement carefully rather than chasing early inventory on principle.

The local layer

Local search in 2026: from ranking to recommendation

Local search stopped being a ranking contest and became a recommendation contest. BrightLocal's 2026 survey found consumer use of AI tools for local business recommendations went from 6 percent to 45 percent in one year, and Google launched Ask Maps, a conversational AI layer inside Google Maps, on March 12, 2026. The question is no longer where you appear in a list. It is what the assistant says when somebody asks for a plumber, a dentist, or a church.

AI Overviews actually appear on far fewer local queries than commercial queries generally, in the range of 7 to 13 percent according to multiple 2026 research compilations. That is why the local contest is not the Overview. It is the recommendation, and that answer gets assembled from your Google Business Profile, your website, your reviews, and third-party mentions read together as a single data stream.

The three moves that carry the weight locally

1. 1

   Consistency first. Name, address, phone, hours, and services have to match everywhere they appear. Inconsistency is the leading cause of an engine describing a business inaccurately, and BrightLocal found only 35 percent of small businesses even have a Google Business Profile in its 2025 SMB report.

2. 2

   Review velocity second. BrightLocal's 2026 survey found 97 percent of consumers read reviews before choosing a local business and 68 percent will only consider a business rated four stars or above. Reviews are now a marketing input that feeds the next buyer's AI answer, not a customer service afterthought.

3. 3

   Problem-state pages third. Generic service pages are almost never cited. Problem-state pages are, because they match the sentence a person actually types when something breaks. "HVAC services in Menifee" loses to "why is my AC blowing warm air," and the second page is the one an engine reaches for.

Behavior on local AI searches is more commercial than most people expect. A Sagapixel local services study found 39 percent of ChatGPT users click through to a business website when searching locally, that the average user reviews 3.7 local businesses with only 27 percent looking at just one, and that 75 percent still type keywords rather than natural language for local services.

Measurement

How to measure AI visibility honestly

AI visibility can be measured well enough to manage and not well enough to attribute perfectly. Anyone presenting a clean attribution model for AI traffic is overselling, and that is worth knowing before you buy one. What is defensible is a baseline, a trend, and a join to pipeline.

Capture a baseline immediately, because there is no backfill

Search Console's Generative AI reports carry no historical data, which means your baseline starts the day you capture it and every day without it is permanently lost. This is the rare piece of advice with an expiration date attached to it. A brand that starts measuring in January 2027 will never be able to show what happened in 2026.

Change what you count

Sessions are now a misleading primary metric, because the traffic that survives is worth more per session than it has ever been. Adobe Analytics found in April 2026 that AI-driven traffic to US retail sites rose 393 percent year over year in Q1 2026, that AI-referred traffic converted 42 percent better than non-AI traffic in March 2026, and that revenue per visit from AI referrals ran 37 percent above non-AI traffic. Adobe's January 2026 holiday analysis found AI-referred visitors spent 45 percent more time on site, viewed 13 percent more pages, and were 33 percent less likely to bounce.

Both halves of this matter and most coverage only reports one. AI referral traffic is small and it is valuable. Cloudflare Radar measured all AI chatbots combined sending 0.29 percent of search referral traffic in May 2026 while Google sent 87.63 percent. A brand whose AI referrals are 1 percent of sessions and 8 percent of revenue has a channel worth building, and a brand measuring only sessions will conclude the opposite and defund it.

Referral share and influence share are not the same measurement. AI answers cover 87 percent of commercial queries while AI chatbots send under a third of one percent of referral traffic. Anyone presenting one of those numbers as the other is either confused or selling something.

Answers

The 20 questions business owners are asking about SEO and SEM in 2026

Twenty questions, answered with a named source and a date on every number. Each answer is written to stand alone if a machine lifts it out of this page.

The engagement

The Three-Phase GEO Roadmap

Ninety days is enough to move a GEO score, and the order of operations is what makes it work. Content published on a broken foundation does not get cited, which is why infrastructure comes first and content comes second in every engagement we run.

Phase 1  |  GEO Foundation  |  Days 1 to 30

Phase 1: GEO Foundation

Infrastructure first. Server-side rendering verified section by section, crawlability and indexation confirmed for both search and AI crawlers, entity consistency corrected across your site, your Google Business Profile, third-party directories, review platforms, and any press coverage carrying your name. Schema implemented once and correctly as hygiene. Navigation reduced so an agent reading your page finds content rather than menu.

The output of this phase is a baseline. Search Console Generative AI impressions, a GA4 AI channel that does not depend on Google's undisclosed referrer list, and a citation share reading across engines, all captured before anything changes so the next ninety days are measurable.

Phase 2  |  GEO Authority  |  Days 31 to 60

Phase 2: GEO Authority

Citation-worthy content built on the foundation. Passage-level structure with a self-contained answer at the top of every section, original data where you have it, direct quotations from named sources, visible source citations, and problem-state pages replacing generic service pages. Earned coverage work begins here rather than at the end, because 84 percent of citations come from outside your website.

The output of this phase is content an engine can lift and a growing presence on the third-party domains engines already trust.

Phase 3  |  GEO Amplify and Measure  |  Days 61 to 90

Phase 3: GEO Amplify and Measure

Distribution, re-testing, and a decision. Visibility is re-tested at Day 45 and again at Day 90, with a before and after GEO score comparison rather than a claim. Paid budget goes behind content that proved itself organically first, never ahead of it.

The output of this phase is a quarterly business review showing what moved, what did not, and what the next ninety days should target. The loop then resets at a higher authority baseline.

Start here

Where to start

Get your GEO Score and the citation gaps behind it

See what an AI visibility audit covers

No obligation. Delivered as a scored report.