Will AI Replace SEO?

Shane HellmrichFounder, PropeloSEO. Reviewed September 2026. Sources cited where a claim needs one.

The short answer

Edward Sturm says it flatly. LLMs get their information from search engines, so the search index is still the source.

The thing everyone calls the replacement is sitting on top of the thing it is meant to replace. Ray's data lands in the same place, because it is almost always ranking number one on Google that gets you more citations.

What did change, and it is not small

Search used to have one outcome. You ranked, or you did not.

AI answers split that in two. The answer text names some brands, and the footnotes cite some sources.

Ray's finding is that Google appears to have decoupled those two lists. You can be the source a model reads and still not be the brand it names.

That is the new failure mode, and it is an ugly one. Your page does the work and your competitor gets named in the sentence the buyer reads.

So what actually wins now

Same two jobs as before, plus one. Rank on page one, because that is still what gets you into the source list, and keep the site fetchable, because a model that cannot read the page cannot cite it.

The new job is being talked about somewhere other than your own site. Recommendation is an off-site outcome, driven by how often other people mention and link to your brand.

That is why most of what is sold as AI SEO misses. It is on-page work aimed at an off-site problem.

What people get wrong

Just format your content for AI. Chunking, schema and readability scores help a machine parse a page, and none of them make a model prefer your brand.

Add an llms.txt file. SE Ranking checked around 300,000 domains and found no citation benefit, and ALLMO found one llms.txt page among 94,614 cited URLs, so it is a five-minute job with no downside that is not a plan.

Write a listicle naming yourself the best. It worked, which is why it is being cleaned up, and Ray tracked companies that scaled these into heavy declines from January 2026.

SEO is dead, pivot everything. The people saying this are usually selling the pivot, and search is the input to the answer engine.

Use AI to write all your content

Google says AI-written content is not penalised. That is a much lower bar than worth publishing.

Shared training data plus shared incentives plus fast iteration loops means everyone sounds alike. Research backs this up, with LLM agents shifting choices toward popular options and generative AI raising individual creativity while reducing collective diversity.

Use LLMs where average is free, like alt text, meeting summaries and data wrangling. Refuse them where average is fatal, like positioning, headlines, hooks and anywhere a buyer picks between you and a competitor.

Invest in inputs models lack. Proprietary data, first-hand interviews, your own experiments and unpublished opinion, because if a competitor can extract your insight from a public scrape, it is wallpaper.

What would change this view

If a controlled test showed an on-page-only change moving a brand from cited-but-unnamed to recommended, with nothing changing off-site, the mechanism described here would be wrong.

If the gap between mentions and citations turned out not to predict anything a business cares about, like calls or bookings, then it is a vanity metric and needs replacing.

If Google or OpenAI re-couple citation and recommendation, this whole page needs rewriting rather than patching. None of that has happened yet.