What Elements Are Foundational for SEO With AI?
The elements foundational for SEO with AI are: technical accessibility for AI crawlers, content structured for chunk-level retrieval, machine-readable meaning through structured data, entity clarity, demonstrable E-E-A-T, off-site authority and brand mentions, and ongoing freshness and measurement. Together these seven pillars determine whether AI answer engines can find, trust, extract, and cite your content.
Why the Mechanics Changed: Retrieval, Not Just Ranking
Classic search ranks pages. AI search retrieves passages. That is the core shift, and it changes what you optimise for.
When someone asks an AI assistant a question, most systems do not just run that one query. They use query fan-out — expanding one question into many related sub-queries and running them in parallel. Google defines this in its own documentation as a set of concurrent, related queries the model generates to fetch additional relevant results. Ask about fixing a weedy lawn and the system quietly also searches for herbicide options and chemical-free removal methods.
For each sub-query, a retrieval-augmented generation (RAG) pipeline kicks in. Google describes RAG as a grounding technique that leans on core Search ranking systems to pull relevant, up-to-date pages from the index. Behind the scenes, documents get split into chunks, each chunk becomes a vector embedding, and the system pulls the chunks whose meaning sits closest to the query. Those chunks — not whole pages — get fed to the model that writes the answer.
Two consequences follow, and both matter more than any tactic in this article.
Ranking #1 no longer guarantees a mention
Ahrefs found 76% of AI Overview citations came from top-10 pages in July 2025, but only 38% in a later analysis of 863,000 keywords and 4 million AI Overview URLs. A 2026 synthesis of 680 million citations put the overlap below 20%.
Your content has to survive being pulled out of context
If a passage cannot stand alone once it is lifted from the page, it loses. Retrieval happens at the chunk level, so every section needs to make sense with nothing around it.
There is a traffic reason to care too. A Seer Interactive study published in November 2025, covering 3,119 informational queries across 42 organisations and 25.1 million organic impressions, found organic click-through rate on AI Overview queries fell from 1.76% to 0.61% — a 61% drop — while paid CTR fell 68%. Ahrefs put the position-one CTR decline at 58% in its December 2025 data. Being cited is how you earn attention on those queries now: the same Seer study found brands cited inside an AI Overview earned 35% more organic clicks.
And those visitors are worth having. Similarweb's 2025 research put generative-AI referral conversion at roughly 7% on transactional sites, about two points above Google search, while a June 2025 Semrush study found AI-referred visitors converted at 4.4 times the rate of traditional organic traffic.
Each Pillar Answers One Question
an AI System Asks About You
Can I reach you? Can I extract you? Do I know what you are? Do I know who you are? Should I trust you? Does anyone else vouch for you? Are you still current? Get all seven right and you have given AI systems everything they need.
Technical Accessibility and AI Crawler Access
What it is: Whether AI crawlers can reach your content and read it in the form they actually receive it.
Why it matters for AI
This is the pillar most teams get wrong without knowing. Googlebot renders JavaScript with a headless Chromium engine. Most AI crawlers do not. A joint Vercel and MERJ analysis of over 500 million GPTBot fetches found no evidence of JavaScript execution at all. GPTBot downloaded JavaScript files in around 11.5% of requests but never ran them. ClaudeBot downloaded JS in roughly 23.8% of requests and also never executed it. PerplexityBot behaves the same way.
If your main content is painted in by client-side React, Vue or Angular after load, those crawlers see an empty shell — no matter how good the writing is. The result is a split-visibility problem: a site ranks fine on Google, which renders, while staying invisible to ChatGPT, Claude and Perplexity, which do not. Google Gemini is the exception, since it rides on Googlebot's rendering infrastructure.
🛠 Fix it this week
- Open an important page and view source (Ctrl/Cmd+U — not the DevTools Elements panel). Search for a sentence of your body copy. If it is not there, it is client-rendered and invisible to AI crawlers.
- Move critical content to server-side rendering with Next.js, Nuxt or Angular Universal, or switch to static generation. Pre-rendering is a workable fallback.
- Check robots.txt and decide deliberately whether to allow GPTBot, ClaudeBot, PerplexityBot and Google-Extended. Blocking them keeps your brand out of AI answers.
- Confirm schema and metadata live in the raw HTML rather than being injected by script after load.
Common mistake: Blocking AI crawlers by accident, often through aggressive CDN or WAF rules, then wondering why you are never cited. Note the distinction: blocking a training crawler like GPTBot opts you out of future model training, which is separate from blocking the retrieval bots that assemble live answers.
Content Structure Built for Chunk-Level Retrieval
What it is: Organising pages so a machine can lift a clean, self-contained answer out of any single section.
Why it matters for AI
RAG systems retrieve at the passage level. If your first 150 words are throat-clearing, that chunk contains no extractable claim, gets discarded, and you have wasted your highest-value real estate.
The founding academic work here is the Princeton GEO study (Aggarwal et al., KDD 2024), which tested nine content tactics across roughly 10,000 queries. Its strongest tactics — citing sources, adding quotations, and adding statistics — produced relative improvements of 30–40% on the position-adjusted word count metric and 15–30% on subjective impression, with a 115% visibility lift for sources sitting fifth in the SERP. Keyword stuffing did nothing. A March 2026 GEO-SFE study reported that structural changes alone lifted citation rates 17.3% across six generative engines without altering a single word of content.
Length, by contrast, barely matters. Ahrefs found content length had essentially zero correlation (0.04) with AI citation probability, and 53% of AI Overview citations went to pages under 1,000 words. Concision beats padding.
🛠 Fix it this week
- Put a direct, self-contained answer in the first 40–60 words after each heading. Lead with the conclusion, then explain it.
- Use descriptive, question-based H2s and H3s. Every headed section should answer one distinct question and stand alone.
- Convert comparative information into tables and processes into numbered lists. Each becomes a discrete, liftable chunk.
- Replace vague claims like "many studies show" with a specific, sourced number and a year.
Common mistake: Over-fragmenting. More headings is not the same as better chunks — thin, repetitive sections do not carry enough substance to be worth citing. Aim for complete thoughts, not shorter text.
Structured Data and Machine-Readable Meaning
What it is: Schema.org markup in JSON-LD that tells machines what your page and your business actually are.
Why it matters for AI
Here is the honest version, because the industry oversells this one. Google states plainly that there is no special schema.org structured data you need to add to appear in AI Overviews or AI Mode. Correlation data backs the scepticism: Ahrefs found adding schema produced a change indistinguishable from zero for AI Mode and ChatGPT visibility.
That is not the whole story though. Schema removes ambiguity about who you are, and Google still recommends keeping it as part of overall SEO because it keeps you eligible for rich results. The most useful types for AI work are Organization (with sameAs), Article, FAQPage, Product and Person. Treat schema as infrastructure that prevents a failure mode, not a button that causes citations.
🛠 Fix it this week
- Add Organization schema sitewide with name, logo, url, description and a complete sameAs array pointing to LinkedIn, Crunchbase, Wikidata and your social profiles.
- Add Article schema with a real author to posts, FAQPage schema to genuine Q&A sections, and Product schema where it applies.
- Validate everything, and make sure the markup describes content that is actually visible on the page.
Common mistake: Marking up content that is not on the page, or leaving dead URLs sitting in your sameAs array. Both undermine trust instead of building it.
Entity Clarity and Brand Consistency
What it is: Making sure AI systems understand your brand as one unambiguous entity rather than three vaguely related ones.
Why it matters for AI
AI models build internal representations of entities: companies, people, products. If your name is inconsistent — "Acme Corp" in schema, "Acme Corp." on LinkedIn, "ACME" on Wikidata — you look like three different nodes, and none of them accumulates full authority.
Entity recognition also feeds directly into whether you show up in brand comparisons and category answers. Semrush's work on so-called ghost citations shows that being cited as a source and being mentioned as the answer are different signals that can diverge — and mentions are what users act on. Research published in Kevin Indig's Growth Memo found 74% of users picked the top-mentioned brand as their final choice.
🛠 Fix it this week
- Pick one canonical brand name string and enforce it everywhere, down to the punctuation.
- Build or update your Wikidata entry, and keep name, address and phone identical across every directory.
- Link your Organization entity to those external profiles using sameAs.
- Search your brand on Google. No Knowledge Panel is a signal that your entity grounding needs work.
Common mistake: Treating entity work as a one-off schema task. Consistency across the whole web does the work; the markup only declares it.
E-E-A-T and Demonstrable Expertise
What it is: Experience, expertise, authoritativeness and trustworthiness — shown through the content, not claimed in an about page.
Why it matters for AI
Google's generative features are judged by the same quality and E-E-A-T systems as organic search. The single most useful line in Google's own guidance is this: creating content people find unique, compelling and useful will likely influence your presence in generative AI search more than any other suggestion in the guide.
Google explicitly contrasts commodity content, using "7 Tips for First-Time Homebuyers" as its example, with non-commodity content that demonstrates real first-hand experience. This matters even more in YMYL categories — healthcare, finance, legal services — where trust signals carry extra weight and thin content gets filtered out early.
🛠 Fix it this week
- Add real author bios with verifiable credentials, linked to author pages and marked up with Person schema.
- Publish original data, first-hand testing or documented case results that competitors cannot copy. Information gain is the point.
- Cite your sources inline and quote named experts — this aligns with the Princeton GEO findings and with how models weigh credibility.
Common mistake: Publishing generic, anyone-could-have-written-this content at scale. It is exactly what Google's guidance singles out as low value.
Off-Site Authority, Brand Mentions and Third-Party Consensus
What it is: What the rest of the web says about you when you are not in the room: reviews, forums, publications, video.
Why it matters for AI
This is where the data is most striking, and it upends two decades of link-first thinking. Ahrefs' December 2025 study of 75,000 brands, running Spearman correlations across ChatGPT, Google AI Mode and AI Overviews, found the strongest correlates of AI visibility are all off-site brand signals: YouTube mentions (0.737), branded web mentions (0.664), branded anchor text (0.527) and brand search volume (roughly 0.392). All of them outpaced backlinks (0.218) by two to three times.
Correlation is not causation, and the researchers say so themselves. But the pattern holds across datasets. Muck Rack's May 2026 analysis of more than 25 million links across ChatGPT, Claude and Gemini found earned media accounted for 84% of all AI citations, with paid or advertorial content at just 0.3%.
Platform concentration matters too. Reddit and Wikipedia dominate: a June 2025 study of over 150,000 citations found Reddit cited in 40.1% of cases, and Profound put Reddit at roughly 46.7% of Perplexity citations with Wikipedia at 47.9% of ChatGPT's top-ten share. These numbers swing hard — ChatGPT's Reddit share dropped from around 60% to 10% within weeks in late 2025 after a Google parameter change — but the direction is stable. Third-party consensus feeds AI answers.
🛠 Fix it this week
- Get named, not just linked, in credible third-party coverage: industry publications, review platforms like G2, and relevant forums.
- Build a genuine YouTube presence with reviews, tutorials and expert commentary, since brand mentions in titles, transcripts and descriptions correlate most strongly with AI visibility.
- Pursue Wikipedia only if you genuinely meet notability guidelines. Otherwise focus effort on Wikidata and earned media.
- Encourage authentic customer reviews and real community discussion in the places your buyers already read.
Common mistake: Manufacturing fake mentions. Google's guidance warns directly that seeking inauthentic mentions across the web is not as helpful as it might seem, and its spam systems are built to catch exactly this.
Freshness, Maintenance and Measurement
What it is: Keeping content current and actually tracking whether your AI visibility is moving.
Why it matters for AI
RAG explicitly prioritises fresh, up-to-date pages, and citation sets are far more volatile than rankings. Authoritas found roughly 70% of pages cited in AI Overviews change over a two-to-three-month window, and those changes were not tied to traditional ranking movements.
If you are not maintaining content and measuring outcomes, you are flying blind inside a system that reshuffles itself monthly.
🛠 Fix it this week
- Set a quarterly review cycle for your most important pages and refresh statistics, dates and claims.
- Stand up measurement before you need it, so you can tell which changes moved the needle.
- Re-run citation tracking after major model releases or platform updates, since a single change can reshuffle who gets cited.
Common mistake: Publishing and forgetting. AI citation is not a set-and-forget outcome; it decays.
Score Your Own AI SEO Foundations
Tick every statement that is already true of your site. Nothing is saved or sent anywhere — the score updates as you go.
Foundation score
0/12
Start ticking to see where you stand.
Under 8 and the foundations are the bottleneck, not your content volume.
Not sure which pillar is holding you back?
We run the crawler, rendering, schema and citation checks and send you the gaps in plain language.
What to Fix First: A Prioritisation Sequence
Foundations before tactics. There is no point earning a citation-worthy mention if AI crawlers cannot read the page it points to. Work in this order.
| # | Fix | Effort | Why it sits here |
|---|---|---|---|
| 1 | AI crawler access + server-side rendering (Pillar 1) | Medium, one-time | If bots cannot read you, nothing else on this list matters |
| 2 | Content structure and extractability (Pillar 2) | Low–medium, ongoing | Highest-leverage on-page change, and it works on pages you already have |
| 3 | Entity clarity + schema (Pillars 3–4) | Low, mostly one-time | Removes ambiguity cheaply. Insurance, not a growth lever |
| 4 | E-E-A-T depth and original data (Pillar 5) | High, ongoing | The durable moat. Slow to build, compounds once built |
| 5 | Off-site authority and mentions (Pillar 6) | High, ongoing | Strongest correlation with citations, slowest signal to move |
| 6 | Freshness + measurement (Pillar 7) | Low, continuous | Keeps the other six honest and shows you what worked |
Do items one to three in your first month — they are mostly technical and move fast. Treat four to six as the long game, because they are the ones competitors cannot copy in a quarter.
How to Measure AI SEO (and Where the Numbers Lie)
You cannot manage what you cannot see, and AI visibility is genuinely hard to see. Here is what actually exists, and where each method fails.
AI referral traffic in GA4
Build a custom channel group with regex matching AI source domains, placed above the Referral rule.
Blind spot: A large share of real AI sessions arrive with no referrer, so much of this traffic lands in Direct and gets undercounted.
Share-of-voice and prompt tracking
Tools run fixed prompt sets and report how often you are mentioned or cited versus competitors. This is the impressions layer above the clicks layer.
Blind spot: Results are volatile and platform-specific, and no third party can see inside the AI systems' ranking.
Server log-file analysis
Logs capture GPTBot, ClaudeBot and PerplexityBot hits, showing whether AI systems are reading you at all.
Blind spot: Tells you about crawling, not citing. GA4 cannot see any of it.
The honest framing: referral traffic is the click stage, being mentioned is the impression stage, and most AI answers name a brand without any clickable link at all. Pair click-based analytics with visibility tracking and report both alongside your Google numbers.
Five Things That Do Not Work
Every new channel grows a folklore faster than it grows evidence. These five claims circulate widely and do not survive contact with the documentation.
“Add an llms.txt file and you will get cited”
No evidence supports this. Google's Gary Illyes confirmed in July 2025 that Google does not support llms.txt and has no plans to, and John Mueller compared it to the discredited keywords meta tag. Google's May 2026 guidance states you do not need to create machine-readable AI files or Markdown to appear in Search, including its generative features. SE Ranking analysed roughly 300,000 domains, found about 10% adoption, and found no measurable citation effect — removing the file from their model actually improved its accuracy.
“Write for the LLM by stuffing keywords”
The Princeton GEO study tested this directly. Keyword stuffing did not improve AI citation. Clear, fluent, sourced writing did.
“Publish AI-generated content at scale”
Google's scaled content abuse policy targets exactly this pattern, and its guidance emphasises non-commodity content. Volume does not make a site more relevant to a retrieval system.
“Buy an AI SEO guarantee”
Be wary. Google states plainly that no third-party tool has access to its internal ranking or AI systems. Anyone guaranteeing citations is guessing with your budget.
“Schema is a magic AI lever”
Covered in Pillar 3. It is useful infrastructure that removes ambiguity, not a button that produces citations.
⚖️ One claim worth holding at arm’s length
In an August 2025 post on Google’s blog, Search head Liz Reid said average click quality had increased and that Google was sending slightly more quality clicks to websites than a year earlier. That is Google’s position, offered without supporting data in the post, and several independent studies including work from Pew Research and Ahrefs dispute the traffic picture. Treat it as a stance, not a settled fact.
Questions People Actually Ask About This
What elements are foundational for SEO with AI?
Is AI SEO different from regular SEO?
Do I need an llms.txt file to get cited by ChatGPT?
Why do I rank on Google but never get mentioned by AI?
Does structured data help AI visibility?
How do I measure whether AI is citing me?
What is the single highest-impact fix I can make this week?
The Bottom Line
AI did not kill SEO fundamentals. It reordered them. Get the foundations right — crawlable pages, extractable structure, a clear entity, real expertise, genuine off-site reputation — and you hand AI systems everything they need to find, trust and cite you.
Chase hacks like llms.txt or manufactured mentions and you will spend a quarter on things the platforms have already said they ignore. Foundations first, every time.
📚 Sources and studies referenced in this article
- Google Search Central — AI Features and Your Website (updated Dec 2025)
- Google Search Central — Optimizing your website for generative AI features on Google Search (May 2026)
- Liz Reid, Google — AI in Search is driving more queries and higher quality clicks (Aug 2025)
- Aggarwal et al. — GEO: Generative Engine Optimization, ACM SIGKDD 2024 (arXiv:2311.09735)
- Ahrefs — 75,000-brand AI visibility correlation study (Dec 2025)
- Ahrefs — AI search citation overlap study and AI Overviews CTR study (2025)
- Seer Interactive — AIO Impact on Google CTR: September 2025 Update (Nov 2025)
- Vercel + MERJ — The Rise of the AI Crawler
- Muck Rack — What Is AI Reading? (May 2026 edition)
- Semrush — most-cited domains study, ghost citations study, AI referral conversion research (2025–2026)
- Profound — AI platform citation patterns
- Yu et al. — GEO-SFE structural optimisation study (Mar 2026)
- SE Ranking — llms.txt adoption study (~300,000 domains)
- Similarweb — Generative AI Landscape report (2025) and AI search statistics (2026)
- Authoritas — AI Overview citation volatility research
- Kevin Indig — Growth Memo research on brand mention preference
A note on reliability: the Google documentation items are primary sources. The citation-correlation figures are correlation studies, and the researchers themselves stress that correlation is not causation. The Princeton GEO figures are peer-reviewed but were tested on systems mimicking Bing Chat and Perplexity rather than every live engine. Conversion and traffic-growth statistics come from vendor datasets that skew toward marketing and technology verticals, so read them as directional. Where studies disagree — the top-10 overlap ranges from 17% to 54% depending on methodology — that spread is noted in the text rather than resolved to one tidy number.
Next Step
Find Out Which Pillar Is
Costing You Citations
We check rendering, crawler access, chunk structure, schema, entity consistency and current citation share across ChatGPT, Perplexity, Gemini and AI Overviews. You get the specific gaps and the order to fix them in.
Apply this to your industry
See every industry we work with or read more on the Intelligence Hub.