SEO in the age of AI: what stays, what dies, what gets weirder.

Every SEO post in 2026 is panicking about AI. Most are wrong about which parts of SEO are dying. The fundamentals — technical health, helpful content, real authority — became more important, not less. The grift around those fundamentals got harder. Here is the honest mapping.

To be clear up front: we do use AI tooling in our SEO work. Schema generation. Content audits. Pattern detection across thousands of pages. Light-touch programmatic assistance for repetitive technical fixes. What we do not do is what destroyed a generation of SEO companies in 2023-2024: dump AI-generated content into a CMS at industrial scale and pray that nothing breaks.

The line between the two is small in the diagram and enormous in practice. Almost everything in this article is about where that line should sit.

What stays exactly the same

Some things did not change at all. They are just more important now.

Site speed and technical health. Core Web Vitals, mobile usability, crawlability, indexability. Google still ranks on these. AI search engines (Perplexity, ChatGPT, Gemini) still need to crawl and parse your pages before they can cite them. A site that loads slowly does not get cited by AI for the same reason it does not rank in Google. web.dev remains the canonical reference.

Original content with real authority. Google’s helpful content guidance already required this. AI search engines reinforced it. They cite the sources that look authoritative — real bylines, real publication dates, real citations of their own. They do not cite AI-spun thin content. They are trained on the same web we are.

Internal linking and information architecture. The structure of your own site still matters. It tells crawlers (search and AI) what is important and how to find it. We see sites with brilliant individual pages and zero internal linking. They underperform every time.

Schema markup. Schema.org structured data was useful for Google. It is essential for AI. LLMs use schema to disambiguate what a page is about. If you sell shoes, your schema should say so. If you are a local business, your schema should include your address. The five minutes it takes to add this pays off across every discovery surface.

What is dying (and was already dying)

These were broken before AI. AI just made them obviously broken.

Keyword stuffing. Google’s 2022 Helpful Content Update was the death blow. AI search engines were the burial. Pages that read like they were written for crawlers do not get cited.

Programmatic content at scale. The strategy of generating 5,000 pages from a database table and hoping they rank is dead. We watched it die in real time across 2023-2024. AI made it worse because (a) Google can detect AI-generated bulk content more reliably, and (b) AI search engines actively penalize bulk-similar pages when picking citations. Search Engine Land has tracked the collateral damage.

Link farms and PBNs. Paid link networks, private blog networks, “guest post” rings on irrelevant sites. The detection has gotten too good. The penalty for getting caught is too steep. Honest backlinks from real publications are the only sustainable path.

Doorway pages. Service-in-city, service-state, industry-state pages built by mail-merging a template. Killed by both Google and the AI engines. We have written about this in our SEO playbook piece.

Everything that worked in 2018 and felt slimy is now an active liability. The slime detection just got really good.

What gets weirder

This is the interesting category. The changes are real, the playbook is half-written, and what works is going to keep shifting through 2026 and 2027.

Citation distribution, not just click distribution.

The old SEO scoreboard was clicks. Position 1 got the most clicks. Position 10 got far fewer. In AI search, the equivalent metric is: did the engine cite you? You either get cited (and a click) or you do not (and the user reads the AI summary without ever visiting your site). The shape of the distribution is different. More content competes, but the citation pattern rewards genuine authority and depth.

llms.txt and the AI-discovery layer.

The llms.txt protocol emerged in late 2024 and became a quiet standard through 2025. It tells LLMs what your site is about, what pages matter most, how they relate, and which content carries the authority signals worth citing — the AI-era equivalent of robots.txt and sitemap.xml, but more editorial than either. What separates an effective llms.txt from a useless one is the judgment behind it: which pages get priority, how each is framed, what to leave out, and how the file stays accurate as the site evolves. We build and maintain one for every client now, as part of how we think about AI discoverability end-to-end.

Generative Engine Optimization (GEO).

The new acronym you will see in 2026. GEO is the practice of optimizing your content for being cited by generative engines, not just ranked by Google. Ahrefs has been publishing decent research on what factors correlate with AI citation: clear content structure, frequent use of headings, explicit answers to common questions, brand mentions, and verifiable claims with sources. Most of these are good SEO anyway. GEO is mostly a relabeling, but the relabeling matters because the optimization targets are slightly different from a traditional SERP.

Brand mentions as ranking signal.

The older SEO mantra was “backlinks are the currency.” In the AI era, brand mentions — even unlinked — are part of the signal. If credible publications write about you, AI engines pick that up. The implication: PR work and brand-building become more directly tied to discoverability. You can rank without backlinks if your brand is talked about. (You still also need the backlinks.)

Zero-click answers and the long tail.

AI engines will answer the question directly, often without sending a click. For simple informational queries, the click is gone. But for high-intent queries — “best AI agency in Denver for cannabis brands,” for example — the AI cites and the user clicks the citation. The strategic implication: stop writing for “what is X” queries. Start writing for “which X is best for my specific situation” queries.

What we actually do for SEO clients in 2026

Concrete and unglamorous. Here are the activities in an honest SEO engagement — the cadence, volume, and scope scale to the project, the goals, and the budget:

1. Technical audit and remediation — recurring, because the broken stuff is never done.
2. Original content production, hand-edited, with real author bylines and citations. Some clients want a piece or two a month. Others run a full editorial calendar. Both work, as long as the standard holds.
3. Internal linking and information architecture review — what new pages need to be linked from what existing ones.
4. Schema and llms.txt maintenance.
5. Brand mention monitoring and outreach — light-touch through full PR program, sized to the goals.
6. Reporting on a cadence that fits the engagement, written to distinguish “we did this” from “this happened.”

That is the whole list. There is no secret seventh item. Anyone selling you a secret seventh item is selling you a story.

What this is really about

AI did not kill SEO. It killed the dishonest version of SEO. The honest version — build real, helpful, well-structured content on a fast, accessible website with clear authority signals — just became more valuable. The agencies still selling 2018 are about to have a very hard time. The agencies that adapt are quietly doing the best work of their careers.

If you are evaluating an SEO partner in 2026 and they cannot explain how AI search changes their playbook, that is your answer. The future is too obvious to ignore. The agencies ignoring it are the ones already losing your traffic.