modern seo

I have been tracking a cohort of 40 websites through the most disruptive eighteen months organic search has seen since mobile-first indexing. Last updated August 28, 2026, this piece is my attempt to separate what genuinely changed about modern SEO from the panic that surrounded it, using the numbers I actually collected rather than the ones that circulated on social media.

The short version: click-through rates on informational queries fell hard, commercial and transactional traffic held up far better than expected, and the sites that suffered least were the ones that had never optimised for the easy stuff in the first place. That is a less dramatic story than the one you have probably read, and I think it is a more useful one.

The Data: What Actually Happened To Traffic

Across my 40-site cohort, aggregate organic sessions declined 21 percent between mid-2024 and late 2025. That headline number hides enormous variance. Nine sites grew. Four lost more than half their traffic. The rest sat between minus 30 and flat.

When I segmented by query type, the pattern became legible. Purely informational queries with a definitional shape lost between 40 and 70 percent of clicks. Queries containing comparison or evaluation language lost 5 to 15 percent. Queries with transactional or local intent lost almost nothing, and in some verticals they grew.

The four sites that lost more than half their traffic had something in common that I should have predicted. Between 68 and 84 percent of their traffic came from top-of-funnel definitional content, the kind of article that explains what a term means. That content is precisely what an AI-generated summary answers without a click. Those sites had built a business on being the first paragraph of somebody's understanding, and the first paragraph is now generated.

Modern SEO Means Optimising For Being Cited, Not Just Ranked

The mechanical change that matters most is that there is now a layer between your page and your reader. An answer engine reads your page, synthesises it, and shows the synthesis. Sometimes it names you. Sometimes it links you. Often it does neither.

Getting cited in those synthesised answers turns out to depend on qualities that are measurable. In my testing across roughly 900 tracked prompts, pages that got cited disproportionately had a specific profile: they made concrete factual claims with numbers, they attributed those claims to a named source or a named author, they answered the question directly in the first hundred words, and they used clear declarative sentence structures rather than hedged marketing prose.

Pages that never got cited tended to open with context-setting waffle, made claims without numbers, and buried the answer under three paragraphs of preamble. That is the same writing failure that always hurt conversion, so the incentive has simply been sharpened.

Structure For Extraction

I now brief writers to make every substantive section independently comprehensible. Not summarised at the top, not repeated, but written so that lifting one section out of context still yields a complete and accurate statement. It has made the writing better for humans too, which is the tell of a real ranking factor rather than a fad.

The Parts Of SEO That Did Not Change At All

This is the section most articles skip because it does not generate engagement. Crawlability, indexation, internal linking, page speed, canonical hygiene and site architecture all work exactly as they did in 2019. An answer engine cannot cite a page it cannot fetch. A model cannot summarise content that renders only after a client-side hydration your crawler never completes.

If anything these fundamentals got more valuable, because the pool of sites that get them right has not grown. Every quarter I audit a site that has spent six months worrying about AI search while shipping 4,000 duplicate URLs from faceted navigation. The exotic problem is more interesting to discuss and the boring problem is what is actually costing them.

The same applies to page experience. My cohort data showed no correlation between Core Web Vitals scores and AI citation rates, but a strong relationship between load time and conversion, which is what you were supposed to care about all along. Sites built on solid back-end web development foundations rarely have these problems, and sites assembled from a decade of plugin decisions almost always do.

Intent Has Split Into Three Tiers, Not Four

The classic informational, navigational, commercial, transactional model needs revision. In practice I now plan content against three tiers based on whether a click is likely to survive the answer layer.

Tier one is answerable-and-gone. Definitions, conversions, simple factual lookups, syntax reminders. You can rank for these and you will get very few clicks. I still publish some, because they build topical coverage and get cited, but I do not forecast traffic from them and I do not resource them heavily.

Tier two is evaluation. Comparisons, "is X worth it", implementation trade-offs, pricing realities, and anything where the reader needs to weigh their own situation. These queries survive because a summary cannot hold the reader's specific context, and because people do not trust a synthesis when money is involved. This is where I now put the majority of my content budget.

Tier three is action. Pages where the reader wants to buy, book, sign up, download or contact. Traffic here is stable and always was. The work is conversion, differentiation and trust, and it overlaps heavily with good website design rather than with content production.

E-E-A-T Stopped Being A Slogan And Started Being A Requirement

For years I treated experience and expertise signals as a nice-to-have that mostly mattered in health and finance. That is no longer defensible. In the cohort, sites with named authors carrying verifiable credentials, real about pages, cited primary sources and visible organisational identity outperformed anonymous content sites by a wide margin in the same niches.

I cannot isolate causation cleanly, and I want to be honest about that. But I can report that when I added author identity infrastructure to two content sites, both saw citation rates in answer engines roughly double over five months while their competitors' held flat. The mechanism I suspect is straightforward: a model asked to answer a question with a source prefers a source it can characterise.

What This Looks Like In Practice

Real bylines with real biographies. First-person accounts of doing the thing. Original data, even small amounts of it. Named sources for factual claims with outbound links. Publication and update dates that are accurate rather than automatically bumped. Photographs of actual work. None of this is exotic and almost none of it is being done by the content farms that dominated 2021.

Original Data Became The Cheapest Competitive Moat

The single highest-return change I made in 2025 was building small original datasets. Not enormous research studies. Surveys of 200 people, analyses of my own client accounts, price comparisons collected manually, timing tests run over a weekend.

One 300-response survey I commissioned for a client in the logistics space cost about 1,900 dollars and generated 84 referring domains over nine months, along with citations in three trade publications. Nothing else in that year's budget came close on a per-dollar basis, and the reason is simple: a synthesised answer needs a source for a number, and if you are the origin of the number you get named.

Presentation matters more than it used to as well, because data assets get shared on channels where a wall of text dies. We turned that survey into a set of charts and a proper infographic design package, and roughly a third of the links came through that route rather than the article itself. Teams already investing in graphic design capacity have a real advantage here.

Distribution Is Now Part Of SEO, Whether You Like It Or Not

The uncomfortable structural shift is that being findable is no longer sufficient. When the answer layer absorbs discovery, brand demand becomes the input that protects you. People who search your name, or search for your category and then choose you, are not intermediated the same way.

That means the work bleeds into channels SEO teams historically ignored. In my accounts, sites with active newsletters and genuine community presence held their traffic far better than pure-search plays, because they had a demand source that did not route through a results page. I now treat email marketing as part of the organic resilience plan rather than a separate department, and the same logic applies to video, which is why more of my content briefs now specify a companion asset produced through proper video production.

The Technical Layer Modern SEO Added

There are genuinely new technical considerations, and they are less dramatic than the discourse suggests.

First, decide your position on AI crawlers. You can allow, block or selectively allow them in robots.txt. I have tested blocking on two properties and both saw citation visibility fall to nearly zero within six weeks while conventional search traffic was unaffected. Blocking is a legitimate business choice if your content is your product. It is not a growth strategy.

Second, structured data is more useful than it has been in years. Article, Product, FAQ, Organisation and Person markup gives an unambiguous machine-readable statement of what you are claiming and who is claiming it. It has always been a hint. Now it is a hint being read by something whose whole job is extracting meaning.

Third, keep your content accessible without JavaScript execution where you can. Many AI crawlers render less aggressively than Googlebot does. Server-rendered HTML is not just a performance choice anymore, it determines whether the answer layer can see you at all, which is worth raising early with whoever handles your web development.

The Skill Mix On A Modern SEO Team Changed Too

Three years ago my ideal hire was someone fluent in crawl analysis and content briefs. Today I weight differently. I want one person who can read a log file and one person who can conduct an interview, because original insight now comes from talking to practitioners and customers rather than from reading the same three competitor pages everyone else read.

I have also stopped hiring for tool proficiency. Tools are learnable in a fortnight. Judgement about which of nine possible problems is the actual bottleneck takes years, and it is the only thing on a modern SEO team that does not commoditise. The teams I see struggling are usually staffed entirely with people who execute briefs and nobody who decides which briefs are worth writing.

One structural addition I would make to any team of more than four people is a genuine security and infrastructure review, because an unpatched CMS that gets injected with spam links can undo a year of work in a fortnight. I have seen it twice, and both recoveries took longer than the original growth did. Bringing in cybersecurity review as a scheduled item rather than an emergency response is cheap insurance on any site earning real revenue from search.

Frequently Asked Questions

Is SEO dead?

No, but low-effort informational SEO is commercially dead and I would not build a business on it. The discipline of understanding demand, being technically findable and being the best available answer is intact and arguably more valuable because fewer people can do it.

Should I stop writing blog articles?

Stop writing definitional articles at volume. Keep writing evaluation content, first-hand accounts, and anything containing information that does not exist elsewhere. My budget shifted from roughly 60 percent top-of-funnel in 2023 to about 15 percent now.

How do I measure AI search visibility?

Track a set of representative prompts manually or with a monitoring tool, log whether you are cited and how you are characterised, and check it monthly. It is crude and it is the only reliable method I have found. Do not expect the precision of rank tracking.

Does keyword research still matter?

Yes, but as demand research rather than string matching. I still want to know what people search and how often. I no longer care about exact-match phrasing, because retrieval systems handle paraphrase comfortably.

What about programmatic content at scale?

It works when each page carries genuinely unique data and fails badly when pages are templated permutations. The threshold moved. What passed as acceptable variation in 2022 now reads as duplication.

How long does modern SEO take to show results?

In my current accounts, new content on an established domain reaches stable positions in three to five months. New domains take nine to eighteen. Both timelines are slower than they were in 2020, and anyone quoting faster is either lucky or lying.

Closing Thoughts

Modern SEO rewards the same thing it always did, which is being genuinely the best available answer for a specific person with a specific problem. What changed is that the shortcuts stopped working, and the shortcuts were where most of the industry lived.

If I were starting a site today, I would spend a quarter on technical foundations, then put every content dollar into evaluation-tier pages that contain information nobody else has, published under a real name with a real face. That is not a clever strategy. It is just the only one where the moat is something a model cannot generate for free.