Why ‘it’s just SEO’ misses the mark in the era of AI SEO

Most of what people are calling “AI SEO” right now is just legacy SEO with new buzzwords.

AI SEO is different.

When you look at how AI Overviews, ChatGPT, Perplexity, and AI Mode actually source and compress information, there are things you can do that literally did not exist in the Google 10-blue-links world.

This article walks through those AI SEO-only tactics, backed with real data, not wishful thinking.

The context: clicks are collapsing, but answers are not

You already feel this in your numbers.

A few facts to anchor the conversation:

• Studies on Google’s AI Overviews showed click-through rates to top organic results drop by roughly 30 to 35% when an AI summary appears, with some categories and publishers reporting losses of 40 to 80% of search traffic on affected queries.
• One analysis using Similarweb data found news traffic from Google fell from about 2.3 billion to under 1.7 billion visits in a year as zero-click searches grew from 56 to 69% once AI summaries were rolled out.
• A Semrush study across 10 million keywords showed AI Overviews now appear on a meaningful share of queries, skewed heavily toward informational intent, and are already reshaping how visibility works by consolidating multiple sources into a single AI-generated answer.

At the same time, the AI market is compounding at north of 30% CAGR, with estimates taking total AI spend into the low trillions by the early 2030s.

So the demand for answers is not going away; it’s moving.

Traditional SEO optimizes for documents that win clicks on a results page.

AI SEO optimizes for facts, entities, and evidence that win a place inside the answer itself.

Below are 12 tactics that exist only in that second world.

1. Prompt graph coverage

Traditional SEO treats a query as a single unit that maps to one page.

Generative engines do something very different. They break a query into a graph of sub-tasks, fetch information for each node, then recombine it. Google has been explicit about AI Overviews using “multi-step reasoning” to handle complex questions in one shot.

Academic work on AI SEO supports this: experiments with AI search functions demonstrate that they fan-out queries into sub-questions and then synthesize information across multiple sources, particularly for comparisons and multifaceted tasks.

AI SEO-only move: you model that graph yourself.

• Map the parent query into its predictable sub-questions.
• Build self-contained sections that fully resolve each sub-task.
• Make each section independently “liftable” for that micro-intent.

You are not just writing for “best project management software.” 

You are writing for:

• “criteria for agencies”
• “comparison vs spreadsheets”
• “pricing breakdown by seat”
• “implementation timeline”

Each of those needs its own atomic, well-titled, well-structured passage.

SEO can target a long-tail keyword cluster. It cannot explicitly optimize for the model’s internal reasoning graph.

2. LLM seeding

Search engines do not absorb your content into their ranking brain.

LLMs do.

Research and practitioner studies on AI search show a consistent bias toward earned, neutral sources like Wikipedia, government and standards bodies, and community documentation, far above brand-owned marketing pages.

Backlinko’s AI SEO and AI optimization work calls this out bluntly:

“if you want to be cited, you have to show up in the surfaces models actually scrape and trust, not just your own site.”

AI SEO-only move:

• Publish definitions, glossaries, and FAQs in neutral, public locations.
• Contribute to documentation, standards, and community Q&A where models learn their base facts.
• Seed Q&A style content in forums and public repos that are heavily used for training.

You are not asking “how do I rank this URL?” You are asking “where will the model learn the canonical version of this concept, and how can I be that source?”

3. Passage-level retrieval optimization

Classic SEO mostly ranks at the URL level.

Generative engines retrieve at the passage level.

Empirical audits of AI answer engines using frameworks like AI SEO-16 show that what gets cited is not an entire page but specific chunks scoring well on structure, metadata, and semantics.

AI SEO-only move:

• Treat every H2/H3 as a self-contained answer that can be ripped out of context.
• Include the full claim, the qualifiers, and the proof in that one passage.
• Avoid forcing the reader (or model) to scroll around the page to rebuild the logic.

The goal is to be the cleanest “reference paragraph” on the internet for a given micro-question.

SEO can help that page rank. AI SEO is about making the one paragraph the model wants to quote.

4. Citation-ready evidence packaging

Generative engines need to justify their answers.

When researchers looked at what kinds of pages AI answer engines cite most often, they found strong positive signals for: structured data, semantic HTML, clear headings, and explicit evidence like tables and stats.

At the same time, hallucination studies show that models are more likely to invent details when they lack concrete, verifiable facts in their retrieval set.

AI SEO-only move:

• Package numbers, ranges, and timelines in tight, machine-readable formats: tables, bulleted comparisons, glossaries, checklists.
• Always pair a strong claim with a concrete stat and a source.
• Make it trivial for the model to lift three sentences and a table as “the proof block” inside its answer.

It is not enough to be true. You have to be structured in a way that makes truth easy to reuse.

5. Neutrality engineering

Models are aggressively tuned to avoid “salesy” language and unsupported claims.

There is growing evidence that generative engines overweight neutral, non-promotional sources and downweight obviously commercial copy, especially in early stages of answer construction.

On top of that, Google explicitly broadened its definition of “spam” to include shallow content that brings no unique perspective or depth, especially in the context of AI Overviews.

AI SEO-only move:

• Strip product copy of marketing fluff in pages you want cited.
• Lead with facts, comparisons, and third-party validation.
• Put opinion and positioning into separate layers that are not competing to be the “neutral evidence paragraph.”

You can still sell. Just do not try to be both the lawyer and the judge inside the same paragraph.

6. Brand-entity memory alignment

Search engines mostly care that your page matches the query and passes quality thresholds.

LLMs care about whether your entity is understood and consistently described across the corpus.

Academic and practitioner studies on AI search show large variance in how different engines frame the same brand, with systematic bias toward well-known entities and clean, consistent profiles.

AI SEO-only move:

• Decide what the canonical facts are for your brand: who you are, what you do, where you operate, who you serve.
• Make those facts consistent across high-authority surfaces: your site, Wikipedia if you have one, Crunchbase, major directories, partner listings, and media profiles.
• Correct outdated or conflicting descriptions wherever you can.

You are training the model on who you are, not just what your title tag says.

7. Competitor co-occurrence hijacking

Comparative prompts are where a lot of real buying intent lives.

AI answer engines handle those by pulling in multiple entity clusters and then synthesizing an answer. AI SEO research and observational data across tools like ChatGPT and Perplexity show that the brands that consistently appear in “vs” and “best for X” answers tend to have rich, neutral coverage in earned media and comparison-style content.

AI SEO-only move:

• Intentionally place your brand into objective, third-party comparison content that is likely to be used as training or retrieval data.
• Publish high-quality comparisons where you and key competitors are both treated seriously, with real trade-offs, not hit pieces.
• Encourage analysts, reviewers, and power users to include you in “shortlist” style content that will be scraped as category context.

Traditional SEO lets you fight for competitor keywords and maybe win a slot. AI SEO lets you become part of the default peer set that the model thinks about when anyone asks for options.

8. Source blending strategy

In AI search, the “SERP” is not a page.

It is a blend of surfaces:

• Brand sites
• Documentation
• Q&A threads
• Academic papers
• Government or standards bodies
• News
• Product reviews

Semrush’s AI Visibility Index work and independent AI SEO analyses both show that generative engines pull from a much more diverse domain set than classic search, with a bias toward community and documentation sources for many categories.

AI SEO-only move:

• Design your presence as an ecosystem, not a website.
• For each topic you care about, identify the top non-Google surfaces that influence LLMs in that niche and get credible representation there.
• Use consistent phrasing and facts across those surfaces so the model sees a clean pattern, not noise.

You are optimizing for corpus composition, not one index.

9. LLM-friendly specification publishing

One thing models are genuinely good at is snapping to structure.

Across multiple AI optimization case studies, content that performs best in generative answers tends to include:

• Explicit definitions
• Parameter lists
• Formulas or frameworks
• Stepwise instructions
• Constraint and edge-case handling

AI SEO-only move:

• Expose your core frameworks as specifications: “To qualify as X, something must satisfy A, B, C”.
• Turn fuzzy positioning into explicit decision trees that the model can re-use.
• Document your methodologies in public, detail-rich formats.

You are giving the model something better than marketing copy. You are giving it a reusable schema.

10. Training-surface expansion

An entire industry is forming around AI SEO, with market estimates projecting tens of billions in spend over the next decade as brands realize that AI search is not a sideshow.

That spend is not going to a single index.

AI SEO-only move:

• Identify training-adjacent surfaces in your vertical: open datasets, public PDFs, GitHub repos, standards, academic or quasi-academic reports.
• Place your best explanations and evidence there in permissive formats that are likely to be ingested or at least retrievable.
• Treat every public artifact as a potential training seed, not just a lead magnet.

This is not about “letting models train on your content” blindly. It is about choosing where and how they encounter your version of reality.

11. Anti-hallucination engineering

Hallucination is not theoretical.

Benchmarks (e.g., Vectara’s hallucination leaderboard) and academic work (e.g., HalluLens and similar studies show) that even top models still produce fabricated details in a noticeable share of responses, especially for low-coverage or ambiguous topics.

For brands, the risk is simple: when the model does not know enough about you, it will make something up.

AI SEO-only move:

• Publish short, high-signal fact sheets about your brand, products, pricing models, and policies in multiple neutral locations.
• Eliminate contradictory public claims where possible.
• Track how AI systems currently describe you and correct the most harmful inaccuracies with targeted content and outreach.

You are not going to drive hallucination risk to zero. But you can move from “the model guesses in the dark” to “the model snaps to one of several consistent, well-documented descriptions.”

12. Mention vs. citation optimization

In AI search, there are three very different states:

1. You are not mentioned at all.
2. You are mentioned in the narrative, but not cited.
3. You are both mentioned and cited as evidence.

Backlinko’s AI optimization research, Semrush’s AI search studies, and new academic work on AI answer engine citation behavior all converge on the same point: citation patterns are systematic and correlate with specific on-page and cross-site quality signals.

AI SEO-only move:

• Engineer pages for both narrative suitability and citation quality: clear purpose, tight scope, strong metadata, structured data, and external corroboration.
• Build out earned media so that third-party sites can be cited even when your own domain is used more in the narrative.
• Measure where you currently sit in each state across engines and design campaigns that explicitly try to move you “up the ladder.”

Traditional SEO had “impression versus click.” AI SEO has “mention versus citation,” and it is at least as important.

The uncomfortable balance

A few realities worth stating clearly:

• AI summaries are driving up zero-click behavior and compressing traffic to publishers, with documented CTR declines ranging from ~15 to 80 % depending on query type and vertical.
• Google and other platforms continue to claim “higher quality clicks” and greater satisfaction, while simultaneously rolling these features out more deeply into search.
• LLMs still hallucinate, and there is no credible path to complete elimination, only mitigation via better grounding and evaluation.

You cannot fix those macro forces as an individual brand.

What you can do is play the game that actually exists:

• Stop treating AI answers as a curiosity layered on top of SEO.
• Start treating AI SEO as its own channel, with its own levers, its own measurement, and its own content patterns.
• Design content not just to rank, but to be retrieved, trusted, and reused by generative systems.

Traditional SEO is not dead. It is just no longer the whole funnel.