SEO in 2026: how search engine optimisation works now

SEO in 2026 still means earning visibility in Google, but the prize has changed. Google now answers many queries itself through AI Overviews and AI Mode, so a top ranking earns about 58% fewer clicks than it did before those features, according to Ahrefs. The work that wins today is being the source those answers cite, while protecting the clicks that remain.

How Google ranks pages in 2026

Google’s ranking process has not been replaced by AI. It has been wrapped in it. The crawler still fetches pages, the index still stores them, and the core systems still score them on relevance, quality, usability and links. What changed is the layer on top. For a growing share of queries a Gemini model breaks the question into sub-queries (Google calls this query fan-out), pulls candidate pages for each, and writes an answer with citations. Ranking for the sub-queries is what gets a page into that answer.

The scale is enormous. In its June 2026 update for website owners, Google said AI Overviews have more than 2.5 billion monthly users and AI Mode has passed one billion. At Google I/O on 19 May 2026 the company made Gemini 3.5 Flash the default model behind AI Mode worldwide, rebuilt the search box to accept images, files and open Chrome tabs, and said AI Mode queries had more than doubled every quarter since launch.

And yet the old signals decide who gets picked. Google’s May 2026 guidance on optimising for generative AI in Search says there are no extra requirements to appear in AI Overviews or AI Mode; standard SEO fundamentals qualify a page. That is convenient for Google to say, but the Seer and Ahrefs data below supports it. Pages that rank are the pages that get cited.

Google also still owns the market. StatCounter puts Google at 91.1% of worldwide search in August 2026, Bing at 4.5%, Yahoo at 1.23%, Yandex at 0.99% and DuckDuckGo at 0.7%. ChatGPT is a real referral source now (OpenAI told TechCrunch it had 900 million weekly users in February 2026), but for most businesses it remains a fraction of Google’s volume. Optimise for Google first, then make sure the same pages are readable by every other engine.

AI Overviews, AI Mode and what happened to clicks

Every serious study since mid-2025 points the same way: when Google shows an AI answer, fewer people click through to websites. The only disagreement is how much. These are the studies that hold up.

The Pew behavioural study

Pew Research Center tracked 900 US adults who installed a browsing monitor for March 2025. Across 68,879 unique Google searches, 12,593 showed an AI summary. With a summary, users clicked a traditional result 8% of the time; without one, 15%. Clicks on links inside the summary were 1% of visits. People were also more likely to end their session on a results page with a summary (26%) than without (16%). Cite this one when someone claims AI Overviews send traffic back.

The Ahrefs Search Console study

Ahrefs compared aggregated Search Console data for 150,000 keywords with AI Overviews against 150,000 informational keywords without them, in December 2023 and December 2025. The top result’s click-through rate on AI Overview queries came out at 1.6% against a forecast of 3.7% without the feature, a 58% cut. For every 100 clicks the number one result used to earn, Google now keeps 58.

The Seer Interactive brand panel

Seer’s April 2026 update tracks 53 brands, 5.47 million queries and 2.43 billion organic impressions, which makes it the best source for trend. Organic click-through rate on queries with an AI Overview fell from 3.19% in January 2025 to 1.31% in December 2025, then recovered to 2.36% in February 2026. Queries without an AI Overview rose from 2.75% to 3.82%. The most useful finding is on citation: a brand cited inside the AI Overview earned about 120% more organic clicks per impression than an uncited brand on the same query, though it still trailed a no-AI-Overview query by 38%.

How often the AI answer appears

Coverage swings, and Google adjusts it. Semrush tracked more than 10 million keywords through 2025 and saw AI Overviews on 6.49% of queries in January, 24.61% in July, then 15.69% in November after Google pulled back. The mix shifted too: in January 2025, 91.3% of triggering queries were informational, but by October commercial queries were 18.57% of the total and transactional 13.94%. Treat commercial queries as exposed, not only how-to content.

Measuring AI visibility in Search Console

Until June 2026 no site owner could see how often Google’s AI features showed their pages. That changed on 3 June 2026, when Google introduced Search generative AI performance reports in Search Console, first for a subset of UK sites. Google’s announcement page was updated on 31 August 2026 to confirm worldwide rollout.

The report gives you impressions inside AI Overviews, AI Mode and the generative features in Discover, split by page, country, device and date. It does not give you clicks, and Google has not said when it will. So the AI report tells you where you are shown, and the standard performance report tells you what you earned. Track the gap monthly.

Two controls arrived alongside it. Google is testing a toggle that lets a site opt out of generative AI features without affecting normal rankings, with the warning that opted-out sites “will not receive traffic or impressions from our generative AI features.” For almost every commercial site that is the wrong trade, because Seer’s data shows the cited page still beats the uncited one by a wide margin. Publishers with licensing power may see it differently, a debate covered in our content and publishing news for September 2026. Google also refreshed its best-practice guidance: unique content, clear structure, good page experience, quality images and video.

Technical SEO and Core Web Vitals

Technical SEO is the part of the job AI search made more important, not less. An AI answer can only cite a page Google has crawled, rendered, indexed and understood, and query fan-out means Google fetches candidates for several sub-queries at once. Pages that render slowly or are blocked by mistake do not get a second chance.

Crawl and index hygiene

Start with the boring checks. Robots.txt should allow Googlebot, and blocking Google-Extended should be a deliberate choice rather than a plugin default. Every page you care about needs a self-referencing canonical, a 200 status, a place in the XML sitemap and at least one internal link. Use the Search Console page indexing report to find “crawled, currently not indexed” clusters; they are usually thin templates, not technical faults. On a JavaScript framework, confirm the rendered HTML contains the body text, because Google’s rendering still adds delay and the AI systems work from indexed text.

Core Web Vitals in 2026

The thresholds have not moved: Largest Contentful Paint under 2.5 seconds, Interaction to Next Paint under 200 milliseconds, Cumulative Layout Shift under 0.1, at the 75th percentile of real Chrome users. The Chrome UX Report release notes for March through July 2026 record no methodology changes, only a small pass-rate regression in May 2026, mostly on Android, that Chrome’s team called seasonal. Ignore any article claiming a 2026 “Core Web Vitals update” changed the thresholds.

Page experience is a tie-breaker, not a headline factor. It still earns attention because INP failures usually come from third-party tags, consent scripts and chat widgets, the same scripts that slow rendering for crawlers. Fixing INP tends to fix crawl waste as a side effect. Our conversion rate optimisation guide covers the revenue side of the same speed work.

Site architecture for fan-out

Because AI Mode splits a question into sub-questions, one long page that answers everything loosely loses to a cluster of pages that each answer one thing precisely and link to each other. Build a hub for the head topic, a spoke per sub-question, and use descriptive anchor text between them. Our content marketing guide explains topic clusters; our digital marketing hub is the model this site follows.

On-page SEO, content and E-E-A-T

On-page SEO in 2026 has one new discipline: write so the answer can be lifted. The first paragraph under each heading states the finding, the number or the recommendation in plain words; the rest supports it. Google’s AI systems and ChatGPT search quote passages, not pages. A passage that makes sense alone is a passage that can be cited.

The basics still count

Title tags that lead with the topic. One H1. Descriptive H2s. Internal links whose anchor text describes the destination. Alt text that describes the image. Meta descriptions that answer the query in 150 characters. None of this is new, and all of it still separates pages that get crawled from pages that get understood. Keyword stuffing died years ago; entity coverage replaced it. A page about SEO that never mentions Search Console, Core Web Vitals or AI Overviews reads as incomplete however often it repeats “SEO services”.

E-E-A-T is a standard, not a score

Experience, expertise, authoritativeness and trust come from Google’s search quality rater guidelines, the document human raters use to grade results. The current edition dates from September 2025, expanded the “your money or your life” category to civic and government topics, and added guidance on rating AI-generated content. There is no E-E-A-T number in the algorithm. But rater judgements train the systems, so the guidelines describe what those systems reward: named authors with real credentials, first-hand evidence such as screenshots and original data, clear sourcing, and an about page that says who runs the site.

AI-assisted content

Google does not penalise content for being AI-assisted. It penalises content generated at scale without adding anything, under its scaled content abuse policy, and the August 2026 spam update was widely reported to have hit that pattern. The safe workflow is the one good editors always used: research first, write the original parts yourself (data, opinion, experience), let a model help with structure, then edit hard. Our AI in digital marketing guide lays out a full production process with quality gates.

“Sites that opt out will not receive traffic or impressions from our generative AI features.” Google, announcement for website owners, June 2026.

Links, authority and the forum problem

Links still matter, and anyone saying otherwise has not looked at who gets cited. Google has spent years cutting the weight of raw link counts in favour of relevance and source quality, but the effect for a site owner is the same: earned mentions from real sites in your field move rankings and move AI citations. Bought links, private blog networks and link exchanges remain spam policy violations, and the 2026 spam updates enforce them.

The uncomfortable fact is who Google trusts most. Ahrefs analysed 145 million US search results in August 2026 and found the “Discussions and forums” feature on 11.7% of them. Reddit appeared in 83.9% of those features worldwide and 87.8% in the US, Facebook in 38.3%, Quora in 32%. Pew adds a related point: 6% of sources linked in AI Overviews were .gov sites, three times their 2% share in standard results. Google’s AI leans on institutions and communities it already trusts.

You cannot out-Reddit Reddit. But you can be present where your customers ask questions, under a real name, because those threads rank and get cited. You can publish original data, because that is what forum posters and journalists link to. And you can fix the internal links on your own site, the one link source you fully control. Local businesses have an extra route through citations and reviews, covered in our local SEO guide.

Structured data after the FAQ rich result shutdown

Google quietly ended FAQ rich results in 2026. A notice on the FAQPage structured data documentation says the feature stopped appearing in Google Search results from 7 May 2026, and on 15 June 2026 Google removed the feature’s documentation. Most sites had already lost eligibility in August 2023, when Google restricted the display to government and health sites.

Should you strip FAQPage markup? No. It remains a valid schema.org type, Google still reads structured data to understand a page, and the question-and-answer format serves AI answers well. This guide keeps its FAQ block and markup for that reason. What you should stop doing is treating schema as a ranking lever. It is a comprehension aid.

Getting cited by AI answers (GEO)

Generative engine optimisation is the practice of earning citations inside AI answers: Google’s AI Overviews and AI Mode, ChatGPT search, Perplexity, Gemini and Copilot. The evidence says it is ordinary SEO done with more discipline. Google’s May 2026 resource says there are no separate requirements. Seer’s data shows cited pages are usually the ones already ranking. What GEO adds is a set of writing habits that make a page easy to quote.

What gets a page cited

Answer-first sections with a specific number or recommendation in the opening sentence. Headings phrased the way people ask questions. Tables for comparisons, because models extract them cleanly. Dates on every statistic, since recency is a visible tie-breaker. A named author and an organisation the model can resolve to an entity. And original data, because an answer built from ten pages saying the same thing has no reason to cite the eleventh.

What does not work

Padding pages with “according to experts”. Publishing hundreds of AI-written question pages, which is the scaled content abuse pattern the August 2026 spam update targeted. Blocking AI crawlers in robots.txt while hoping to be cited; ChatGPT cannot quote what its crawler cannot read. Allow the crawlers that send referrals, measure them in analytics, and revisit quarterly.

Measurement is the hard part. Search Console shows AI impressions but no clicks, ChatGPT and Perplexity referrals appear in analytics only when a user clicks through, and no tool sees inside every engine. Our AI marketing guide compares the AI visibility trackers, and the SEO news for September 2026 covers the latest changes to how each engine cites sources.

Core updates, spam updates and penalties

Google confirmed three ranking updates in the first eight months of 2026, and the last was the sharpest. Knowing the dates lets you match traffic changes to causes instead of guessing.

The August spam update deserves a closer look. SE Ranking compared US rankings for 100,000 keywords across 20 industries on 17 and 22 August, then against a quiet week in late July. During the update 16.71% of top-10 URLs dropped beyond position 100, against 9.2% in the control period, an 82% increase. Volatility ran from 74.64% in real estate to 85.55% in fashion and beauty. Search Engine Journal noted that Google’s May 2026 clarification extended spam rules to attempts to manipulate generative AI responses, though Google did not say whether this update enforced that.

If you lost traffic in a core update, the path is unchanged: find which page types fell, compare them honestly with what now ranks, improve or consolidate, and wait for the next core update. If you lost traffic in a spam update, look for the pattern (scaled thin pages, expired domain reuse, third-party content on your subdomain, paid links) and remove it. Manual actions still appear in Search Console. And if the loss matches none of these dates, check whether AI Overviews started showing on your money queries; that is the most common cause of unexplained 2026 declines.

SEO tools worth paying for

The free layer got better this year and the paid layer got more expensive. The free layer is Google’s own: Search Console (now with AI impressions), the Chrome UX Report and PageSpeed Insights for Core Web Vitals, the Rich Results Test, and Google Trends. Add Bing Webmaster Tools, because Bing’s index feeds Copilot.

The paid layer breaks into four jobs. Crawling and technical auditing: Screaming Frog, with Sitebulb or Lumar for large sites. Keyword and competitor research: Ahrefs or Semrush, both of which now sell AI visibility modules that estimate how often a domain is cited in AI Overviews and chat engines. Rank and SERP-feature tracking: SE Ranking, Advanced Web Ranking, or the tracker inside your research suite, set to record whether an AI Overview appeared. Log analysis: a log-file tool or an edge dashboard, so you can see which AI crawlers fetch your pages. A single-site business needs Search Console, one crawler and one research suite. An agency needs all four jobs covered. Anything beyond that is procurement theatre. Our digital marketing guide includes a tool stack by team size.

SEO KPIs that still mean something

Organic sessions alone is now a misleading KPI, because it can fall while visibility rises. Report these instead, together.

Set targets against your own baseline, not industry averages. A 2.4% click-through rate on an AI Overview query is respectable in 2026 by Seer’s numbers; the same figure would have been a failure in 2023. Pair this scorecard with the framework in our marketing analytics guide so SEO is judged on incremental revenue, the way paid search is in our PPC guide.

A 12-week SEO plan

This plan assumes one site, a small team and no penalties to clear. Stretch the content weeks if you can only improve pages properly rather than quickly.

The single change worth making first is rewriting the opening paragraph of every page that already ranks so it states the answer in two sentences. It costs nothing, needs no developer, and it is the format both Google’s AI features and ChatGPT lift from.

Common SEO mistakes in 2026

Reporting organic sessions without AI impressions, then concluding SEO “stopped working” while the site is cited more than ever. Chasing the FAQ rich result after it was removed. Publishing hundreds of AI-drafted pages that repeat what already ranks, and meeting a spam update. Blocking every AI crawler on principle and then wondering why the brand never appears in ChatGPT answers. Ignoring forums as beneath the brand while Google shows them on 11.7% of results. Building one giant page per topic instead of a cluster that answers the sub-questions AI Mode runs. And diagnosing a traffic drop on day three of a 12-day core update.

Most of these share one root error: measuring SEO the way it was measured in 2021. The engines changed how they present answers. The measurement, and the writing, have to change with them.

Last researched and updated: 7 September 2026.