For two decades, SEO meant one thing: rank in Google’s ten blue links. That model is breaking down. ChatGPT, Claude, Perplexity, and Google’s own AI Overviews are now intercepting queries before users ever reach a results page. If your content isn’t structured to be retrieved, parsed, and cited by large language models, you’re losing visibility you used to take for granted.
This piece is a practitioner’s guide to where SEO is heading, what’s changed under the hood, and the specific tactics worth your time right now.
Table of Contents
The shift from ranking to retrieval
Traditional SEO optimized for a ranking algorithm. You picked a keyword, matched intent, built authority through backlinks, and waited to climb the SERP. The mechanism was deterministic enough that you could reverse-engineer it.
LLM-based search works differently. When a user asks ChatGPT or Claude a question, the system either pulls from its training data, runs a retrieval-augmented generation (RAG) lookup against a live index, or both. The output is a synthesized answer that may cite three sources, may cite none, and rarely sends the user clicking through. Zero-click answers are no longer an edge case. They’re the default.
That means visibility now has two surfaces: the traditional SERP, and the AI interface where your content either gets cited, paraphrased, or ignored. The discovery process for LLMs runs on high-dimensional embeddings rather than exact keyword matches, so the old game of stuffing target phrases into H2s is less useful than it used to be. What matters is whether the model can find your page, understand what it’s about, and decide it’s worth surfacing.
This is the territory now called AEO (Answer Engine Optimization), GEO (Generative Engine Optimization), or LEO (Language Engine Optimization). The acronyms vary. The job is the same: make your content readable to machines that read for meaning, not for keywords.
How LLMs actually read your content
LLMs don’t crawl the way Googlebot does. They interpret. When a model encounters your page, it’s looking for substance, depth, and frontier concepts (ideas or framings that aren’t already saturated across the training corpus). Generic content that restates what every other page in the niche says is statistically redundant. The model has already seen it a thousand times. Yours adds nothing.
A few things matter more than they used to:
Semantic depth and originality. If your page on “alcohol withdrawal timeline” reads like a paraphrase of WebMD, an LLM has no reason to cite it. If it includes original data, a clinician’s framing, or a structural insight the corpus hasn’t seen, that’s frontier content. That’s what gets pulled.
Authentic citations. LLMs evaluate authority partly through who you cite and who cites you. Linking to peer-reviewed sources, government health data, or recognized clinical guidelines signals you’re operating in a high-evidence space. Vague references to “studies show” are noise.
Clear explanations. Models reward clarity. If your content can be summarized in one sentence per paragraph without losing meaning, it’s well-structured for retrieval. If it meanders, the model will either skip it or pull a competing source.
Structured content. Semantic HTML, clean heading hierarchies, and consistent terminology help machines parse what’s on the page. A H2 should describe the section. A H3 should describe a subsection. Don’t use headings for styling.
What to actually do: content optimization for AI
Here’s where the practitioner work happens. These are the levers that move the needle for both traditional SEO and AI retrieval.
Build topic clusters with real interlinking
A single page on “outpatient detox” won’t carry the topical authority you need. You want a cluster: a pillar page that defines the topic broadly, then ten to twenty supporting articles covering specific facets (timelines, medications, insurance coverage, comparison with inpatient, what to expect on day one). Each supporting article links back to the pillar and laterally to siblings.
LLMs map relationships between concepts. A site that demonstrates depth across a topic cluster looks like a domain authority on that topic. A site with one orphaned page looks like a content mill.
Write for high-intent, decision-stage queries
LLMs are increasingly handling informational queries directly. Where they still send traffic is on decision-stage and commercial intent queries: “best outpatient detox in Toronto,” “does Aetna cover Suboxone,” “what to ask before choosing a treatment center.” That’s where users want specifics, comparisons, and trust signals the AI won’t fabricate.
Audit your content portfolio. The pure informational top-of-funnel pages are increasingly cannibalized by AI Overviews. Shift weight toward middle and bottom of funnel where the click still has value.
Use schema.org markup aggressively
JSON-LD schema markup tells search engines and increasingly LLM crawlers exactly what your content is. For healthcare and addiction treatment content, the relevant types include MedicalWebPage, MedicalCondition, MedicalProcedure, Physician, MedicalOrganization, FAQPage, and Article with proper author and reviewer fields. Mark up your medical reviewers. Mark up your authors. Tie author entities to their credentials with sameAs links to LinkedIn, ORCID, or professional registry profiles.
This is the connective tissue that lets a model understand “this article was written by a licensed clinician at a credentialed facility” rather than “this is text on a website.”
Format for parseability
LLMs love content that’s easy to chunk. That means:
- Tables for comparisons (medication dosages, treatment options, insurance tiers)
- Numbered lists for sequential processes (what happens during intake, the seven days of withdrawal)
- Code blocks or callouts for key metrics and definitions
- Quoted experts with attribution
- Diagrams with descriptive alt text and surrounding context
A page that reads as a single 2,000-word block of prose is harder to retrieve in chunks. Same content, broken into scannable sections with clear headings, is dramatically more retrievable.
Anchor text and internal link strategy
Anchor text still teaches both Google and LLMs what a target page is about. Generic “click here” or “learn more” anchors waste the signal. Descriptive anchors that match the target page’s primary topic build semantic coherence across the site.
For agency clients especially, audit your internal linking quarterly. Most sites have a handful of pages doing all the heavy lifting and dozens of orphaned pages no one links to. Fix that distribution.
Keep the technical foundation clean
Indexability in Google and Bing still matters because Bing powers ChatGPT’s web search and Google powers AI Overviews. If your robots.txt is blocking the wrong paths, if your sitemap is stale, if your canonical tags are wrong, none of the content work above compensates for that.
Core Web Vitals also still apply. Slow pages get demoted, and slow pages also create timeout issues when AI agents try to fetch them in real time.
Refresh cycles matter more now
A page published in 2022 about treatment trends is, by 2026 standards, stale enough that an LLM is likely to prefer a newer source even if the older one ranks well in Google. The training corpus refreshes. The retrieval indexes refresh. Your content needs to refresh too.
Build a quarterly review cycle for cornerstone content. Update statistics, add recent developments, refresh citations to current sources, and modify the publish date when meaningful changes are made (not just a date swap, which both Google and reviewers will catch).
Measuring what’s happening
This is the hardest part of AI SEO right now, because the measurement infrastructure is still catching up.
Google Search Console and Bing Webmaster Tools still tell you what’s happening in traditional search. AI Overviews appearances are now showing up in GSC’s performance reports for some accounts. Worth checking weekly.
Beyond that, you’re piecing it together:
- Watch referrer traffic from chat.openai.com, perplexity.ai, claude.ai, gemini.google.com, and copilot.microsoft.com in your analytics. The volume is small but growing, and the conversion rates are often surprisingly high because users arriving from these sources have already been pre-qualified by the AI’s answer.
- Test your own queries. Ask ChatGPT, Claude, and Perplexity questions in your client’s niche and see what gets cited. If competitors are showing up and you’re not, that’s your roadmap.
- Use observability tools that specifically track LLM citations. The category is young but tools like Profound, Otterly, and Peec AI are worth piloting on key client accounts.
Don’t expect clean attribution. Expect signal, not certainty.
What hasn’t changed
Backlinks still matter. Authority still matters. E-E-A-T still matters, arguably more than ever for YMYL niches like healthcare and addiction treatment, because LLMs are explicitly trained to weight credentialed sources more heavily.
The fundamentals didn’t get replaced. They got reweighted. Technical SEO is still table stakes. Original content is still king, just with a higher bar for what counts as original. Link building is still valuable, just with more emphasis on contextual relevance and less on volume.
What changed is that you’re now optimizing for two audiences simultaneously: the search engine that ranks pages, and the language model that synthesizes answers. The good news is that the work overlaps significantly. Content that performs well for LLMs (clear, structured, original, authoritative, well-cited) also performs well in traditional search. The bad news is that the bar is higher than it used to be, and the patience for thin content is gone in both systems.
Get the fundamentals right, then layer on the AI-specific work. That’s the playbook.
Published on: 2024-02-14
Updated on: 2026-05-30
