Three Years Later: What Generative AI Actually Did to SEO

In February 2023, I wrote about the potential impact of chat on paid search revenue and SEO. Three months later, I wrote a retrospective on the impact of chat on SEO and publishers.

At the time, my basic argument was:

AI would make content cheaper and faster to produce
Organic traffic would get harder because chat interfaces reduce clicks
The return on SEO would likely decline across many keyword categories
Unique perspectives would matter more because mediocre content would flood the system

Three years later, I think that direction was broadly right.

But it was incomplete.

What I did not fully understand in 2023 was how quickly content production itself would become commoditized, how real localization at scale would become, and how much SEO would turn into something slightly different: not just ranking in Google, but being retrieved, extracted, cited, and trusted by AI systems.

Over the last few months, this stopped being theoretical for me.

I rebuilt my entire site and blog backend, published new long-form guides with SEO/AEO baked in, implemented a proper structured data and llms.txt stack, and then translated my entire blog archive — 493 posts, roughly 3.9 million words, across 10 languages in 4 days. Not perfectly. In fact, some of the failures taught me more than the successes.

So this post is not "here are my predictions." It is: here is what I think now, after watching the machine actually do the work.

What I Got Right in 2023

Let me start by being fair to 2023 me.

I still believe these parts were directionally correct:

1. Content got dramatically cheaper and faster to produce

That is no longer debatable.

It is now possible for one person with the right AI workflows to produce, update, reformat, interlink, and localize content at a scale that used to require a team.

A few weeks ago, I wrote 6 new mega guides in 4 days. Then I translated my entire archive into 10 languages in 4 more days.

If your thesis in 2023 was "the supply of content is about to explode," yes. That happened.

2. Organic clicks became less secure

Also true.

If a user gets a satisfying answer directly in ChatGPT, Perplexity, Claude, or Google's AI layer, fewer clicks reach the original publisher. This was the big fear in 2023, and I think it was a valid one.

The click is no longer guaranteed just because your content helped answer the question.

3. Mediocre content lost value faster than many people expected

When AI can generate "pretty good" content for everyone, "pretty good" stops being a meaningful advantage.

The floor rose.

That matters because a lot of old SEO strategy was really just industrialized adequacy: enough volume, enough formatting, enough optimization, enough authority signals to rank for a query and capture traffic.

AI made that game much easier to play.

Which means it also made it much harder to win.

What I Missed

Here is what I think I underestimated in 2023.

1. "Unlimited content generation" is not the moat

Back then, the threat felt like volume.

Now I think volume is the least interesting part of the story.

The real question is not whether AI can generate unlimited content. It can.

The real question is: what happens after that?

Who decides what is worth publishing? Who checks whether the structure is right? Who catches the hallucination that is subtle enough to survive a casual review? Who makes the content sound native in another language instead of translated? Who turns a pile of articles into a coherent information architecture? Who gives the AI systems clean, trustworthy signals about what this page is actually for?

That is where the moat moved.

Production became abundant. Judgment did not.

2. Localization is both a superpower and a trap

In 2023, I mostly thought about content generation. I was not yet thinking deeply about localization at scale.

Now I am.

When I translated my blog archive into 10 languages, the technical achievement was real. So were the failures.

Korean went badly wrong on the first attempt. A huge portion of the posts were not translations at all. They were summaries. Cantonese required a completely different voice discipline from standard Mandarin. Internal URLs broke. Sign-offs drifted. Frontmatter got malformed. Some content looked grammatically correct but felt culturally dead.

That experience changed how I think about "unlimited localization."

Yes, AI makes it possible to localize at astonishing speed.

No, that does not mean localization is now trivial.

If anything, it means the gap between translation and localization matters more than ever. I wrote in January that translation preserves words, localization preserves meaning. I believe that even more strongly now.

If you scale low-quality localization, you do not create a moat. You scale mistrust.

3. SEO is now SEO plus AEO plus retrieval design

In 2023, I was still mostly thinking in the language of search engines.

In 2026, that frame is too narrow.

The new game is not just "how do I rank?" It is also:

how do I become easy to extract?
how do I become easy to cite?
how do I become easy to trust?
how do I make my answers legible to both humans and machines?

That is why, when I rebuilt my site, I did not stop at titles, meta tags, and a sitemap. I added structured data, FAQ schema, llms.txt, answer-first sections, question-format headings, and formatting patterns designed for AI extraction and citation.

This is not a cosmetic shift. It changes how you write.

The old SEO mindset often asked: "What keyword do I target?"

The newer mindset asks: "What question am I answering, how clearly am I answering it, and what signals make that answer retrievable and trustworthy?"

That is a different operating model.

What SEO Looks Like Now

If I had to compress my 2026 view into one sentence, it would be this:

AI did not kill SEO. It changed where the value lives.

Here is where I think the value has moved.

1. From content volume to information architecture

If everyone can generate articles, the winning system is not the one with the most pages. It is the one with the clearest structure.

The architecture matters:

pillar pages and supporting pages
internal links that actually help navigation
logical content clusters
URL consistency
metadata that means something
schema that matches the page reality
archives that are maintainable, not just large

A messy 5,000-page site is less impressive than a tightly organized 200-page site that machines and humans can both understand.

2. From keyword stuffing to answer design

The pages that will keep winning are the ones that answer real questions cleanly.

That means:

direct openings
clear headings
strong definitions early
comparison tables where helpful
less throat-clearing
less generic filler
more actual signal

I noticed this in my own recent posts. The more answer-first and structured they are, the more useful they become not just for Google, but for Sydney, for AI systems, and for readers who scan before they commit.

3. From "publish more" to "build better evals"

This is the part I think many people still miss.

When AI raises the floor, your advantage comes from your evaluation layer.

What do I mean by that?

Style guides are evals. Editorial standards are evals. Localization checklists are evals. Schema validation is an eval. Brand voice rules are evals. Your definition of "this page is good enough to publish" is an eval.

Without those, AI output looks productive while quietly degrading quality.

I wrote recently that depth is how you win. I think that applies to content too. The people and companies that outperform will not be the ones with the most aggressive generation pipeline. They will be the ones with better taste, better QA, better judgment, and better systems for defining what "good" actually looks like.

4. From English-only scale to selective multi-market advantage

Localization is now a real strategic lever.

Not because "more languages" is automatically better, but because high-quality localization opens markets that English-only sites leave untouched.

But this only works if the localization is credible.

Literal translation is not enough. Cultural tone matters. Examples matter. Native phrasing matters. Internal links matter. Formatting conventions matter.

If your Spanish page reads like a machine, you do not have a Spanish content strategy. You have a trust problem.

So, Is SEO Still Worth It?

Yes, but not in the lazy way many people still mean it.

If by SEO you mean: "Can I use AI to produce large volumes of low-cost content and get traffic?"

That edge is disappearing fast.

If by SEO you mean: "Can I build a trustworthy, well-structured, deeply useful body of content that is easy for search engines, AI assistants, and humans to understand and cite?"

Then yes. Very much yes.

In fact, I would argue the opportunity is still significant, but the bar has changed.

The old edge was production. The new edge is system quality.

The old edge was speed. The new edge is speed plus discernment.

The old edge was publishing. The new edge is publishing, structuring, localizing, validating, and distributing.

That is a harder game. It is also a more defensible one.

What I Would Tell Publishers and Marketers Now

If I were updating my 2023 advice into a 2026 operating principle, it would be this:

1. Assume content abundance

Do not build your strategy around the idea that publishing more is rare or defensible. It is not.

2. Invest in structure

Fix your site architecture, schema, internal links, content relationships, and retrieval signals.

3. Treat localization as a product problem, not a translation task

If you go multilingual, do it properly. Style guide, QA, native review where possible, and market-specific judgment.

4. Write for retrieval, not just ranking

Think about how an AI system will parse your page, summarize it, and decide whether to cite it.

5. Build direct audience relationships

This part from 2023 still matters. Brand, newsletter, repeat readership, and direct trust matter more when clicks are less guaranteed.

6. Develop taste and editorial courage

AI can draft forever. It cannot tell you what should not exist.

That is still human work.

My Bottom Line, Three Years Later

If I compare what I thought in 2023 with what I think now, the biggest update is this:

I used to think generative AI would mainly change SEO by making content cheap and reducing traffic.

Now I think the bigger change is that generative AI made production abundant, which forced the real sources of advantage into the open.

Those sources are:

trust
structure
depth
voice
QA
localization quality
direct audience relationship
answer design
retrieval readiness

So yes, unlimited content generation and localization changed SEO.

But not because "now everyone can publish a lot."

It changed SEO because once everyone can publish a lot, the market stops rewarding volume and starts rewarding the things volume cannot solve on its own.

That is the update I wish I could send back to myself in 2023.

The old posts still reflect what I honestly thought at the time. I am keeping them up for that reason.

But this is where I am now.

Cheers,

Chandler