How to Rank for GEO and AEO in 2026—The Two-Engine Playbook for AI Citations

To rank for GEO (Generative Engine Optimization) and AEO (Answer Engine Optimization) in 2026, you optimize for two different engine types at once. These are Google’s index-based AI (AI Overviews and AI Mode), where strong SEO fundamentals still decide visibility, and independent answer engines (ChatGPT, Perplexity, Claude), which retrieve and cite content at the passage level using semantic entity matching. The work that wins on both is the same: clear, self-contained passages that answer a specific question, backed by statistics and named sources, published on a crawlable, entity-clear site. Tactics that no longer work include using llms.txt as a ranking factor, relying on FAQ schema for rich results, and content “chunking.” These approaches should be retired.

What “ranking” even means now

In AI search, you are no longer ranking a page, you are getting a passage cited. Modern AI search engines do not retrieve whole pages. They break documents into semantic passages of roughly 200–500 tokens, convert each into a vector embedding, and select the passage that best answers a specific sub-question. A passage gets cited because it directly supports a point in the generated answer, not because its parent page holds a top-ten position.

This is the single most important shift to internalize, and the data is blunt about it. A Surfer SEO analysis of 173,902 URLs found that 68% of pages cited in AI Overviews were not in the top ten organic results for the same query. An Ahrefs study across 15,000 prompts found only 12% of cited links appeared in Google’s top ten. The link between traditional rank and AI citation has largely broken.

For smaller brands, this is the opening. Citation analysis shows that sites with a Domain Authority between 20 and 80 account for 63.6% of all AI citations while the DA 20–40 tier alone earns a larger share than the DA 80–100 tier. You do not need to outrank Forbes. You need the single clearest passage on a specific question.

The two-engine reality (the part most guides get wrong)

There is no single “AI search algorithm” to optimize for. There are two distinct engine types, and conflating them is why most GEO advice underperforms.

Engine 1 — Google’s index-based AI (AI Overviews, AI Mode). These run on Google’s existing Search index and ranking systems. In its official guidance published in May 2026, Google states that optimizing for generative AI search is optimizing for the search experience — and is therefore still SEO. AI Overviews and AI Mode use retrieval-augmented generation (grounding) and query fan-out to surface passages from the same index that powers regular Search. The practical implication: if your content is not technically sound and high-quality enough to rank in traditional Search, it will not appear in Google’s AI answers either. (Read the primary source: Google’s Guide to Optimizing for Generative AI Features.)

Engine 2 — independent answer engines (ChatGPT, Perplexity, Claude). These build their own indexes through their own crawlers and weight signals differently. ChatGPT search runs on OAI-SearchBot and the Bing index; Perplexity cites aggressively, typically pulling 3–8 sources per response; Claude retrieves through its own search crawler. These engines reward passage clarity and entity consistency, and they will cite a page that ranks nowhere on Google if its passage is the cleanest answer to a sub-query.

The winning strategy is not to pick one engine. It is to recognize that the same passage-and-entity layer feeds both — and to build that layer once, well. Where Google says “it’s still SEO,” the independent engines say “it’s still about being the clearest, most citable source.” Those two statements point at the same work.

The semantic-entity layer: how AI decides which brand to name

AI engines do not match keywords, they extract and reason about entities (people, organizations, products, concepts) and their relationships. This is the mechanism that decides whether an AI names your brand or a competitor’s when answering “what’s the best X.”

Three concepts matter most:

Entity salience. Google’s Natural Language API scores how central an entity is to a page, from 0 to 1. A page that mentions a topic once scores low; a page that defines it, gives its history, names its components, and connects it to related concepts scores high. Practitioner analysis suggests pages where the primary entity exceeds a salience score of roughly 0.7 gain a measurable advantage. Put your primary entity in the title, the first sentence, and your H2s and the structural positions where salience signals are strongest.

The Knowledge Graph as citation eligibility. Google’s Knowledge Graph now holds an estimated 500 billion facts about 5 billion entities, and Gemini is trained on it. If your brand has no consistent entity representation, AI systems have nothing stable to attach citations to. The fixes are unglamorous and underused: claim a Wikidata entry, implement sameAs schema linking your official profiles, and make your company name identical across Crunchbase, LinkedIn, and your own About page. Inconsistency is read as untrustworthiness.

Entity density and corroboration. Content that connects 15 or more related entities tends to read as semantically complete, and analysis of large AI-response samples has linked entity-optimized content to roughly 40% more AI citations than keyword-density content. AI assesses authority holistically which means consistent information about you across Wikipedia, reviews, industry press, and your own site strengthens citation likelihood.

Engineer your content for query fan-out

When someone asks an AI a question, the system rarely searches that one phrase. It generates multiple sub-queries behind the scenes. Google AI Mode runs roughly 8–12, ChatGPT generates anywhere from 4 to 20 depending on complexity and retrieves passages for each, and synthesizes them into one answer. To be cited often, your page must answer many of those sub-queries.

Three rules follow directly:

Cover five or more sub-intents per hub page. Research indicates pages addressing five or more fan-out sub-intents have substantially higher citation probability than single-intent pages. Treat five as a floor. Map the sub-queries your topic triggers (tools like Wellows and Surfer’s fan-out extension expose them; Google’s Gemini API returns the actual webSearchQueries array when grounding is enabled), then give each sub-query its own H2 or H3.

Write self-contained passages of 134–167 words. A December 2025 analysis of 15,847 AI Overview results found passages in this word range achieve the highest citation rates. Each section should be independently answerable — one idea per section, opening with a direct answer, written so it makes sense if lifted out of the page entirely. Do not bury the answer three paragraphs in.

Maximize semantic relevance, not keyword overlap. Engines compare the vector of your passage to the vector of the query. The same analysis found passages with cosine similarity above 0.88 to the query earned 7.3x higher citation rates. You hit that by writing in the user’s actual language about the specific thing they asked — plainly, factually, without promotional framing. Marketing-toned passages are selected less often than passages that read like clear explanations.

The research-backed techniques that actually move citations

The foundational study here is GEO: Generative Engine Optimization by Aggarwal et al., published at ACM SIGKDD 2024 (read it on arXiv). The researchers built GEO-bench, a benchmark of 10,000 queries, and tested nine optimization methods against two visibility metrics. Three techniques won decisively:

1. Statistics Addition — replacing vague claims with specific numbers.
2. Quotation Addition — incorporating credible quotes from relevant sources.
3. Cite Sources — including citations to reliable outside sources within your content.

These three delivered 30–40% improvements on the position-adjusted word-count metric and 15–30% on subjective impression — with minimal content change. Note the paradox in technique 3: citing other sources makes AI more likely to cite you, because it signals thoroughness. This is why the article you are reading links out to Google and Princeton.

Two more findings are worth acting on. Stylistic improvements — making content more fluent and easier to understand (Fluency Optimization) — produced a 15–30% lift on their own, and the best-performing combination paired fluency with statistics. And critically, keyword stuffing — the old SEO cornerstone — actively failed, reducing visibility in generative responses.

The technical foundation after the 2026 resets

Two technical resets in 2026 made a lot of older advice obsolete. Get these right and you clear the floor that everything else sits on.

The crawler split: block training, keep citations. The major AI vendors now run separate crawlers for model training versus search indexing, with different user-agent strings. At OpenAI, GPTBot is the training crawler while OAI-SearchBot builds the ChatGPT search index. At Anthropic, ClaudeBot trains while Claude-SearchBot indexes for search. Google-Extended governs Gemini training, separate from Googlebot. The implication: you can disallow training crawlers in robots.txt while keeping the search crawlers allowed — opting out of training without forfeiting the citations that send qualified traffic. The most common and costly misconfiguration in 2026 is a robots.txt that quietly blocks the search crawlers and erases your AI visibility. Audit it first.

Schema after FAQ’s death. Google permanently removed FAQ rich results from Search as of May 7, 2026, and its official guidance confirms no special schema or markup is required to appear in its generative AI features. Schema is not dead — Organization, Product, and Article markup still earn rich results in regular Search, which indirectly feed AI Overviews — but treat it as table stakes, not a citation lever. Stop building your AI strategy around FAQPage schema.

Beyond those: keep pages fast and crawlable, use semantic HTML (real H1–H3 hierarchy, lists, tables), and refresh content on a quarterly cycle with a visible update date — analysis suggests 89.7% of ChatGPT citations go to recently updated pages.

Authority and trust across engines

Google’s clearest single piece of 2026 guidance is its contrast example: a commodity post titled “7 Tips for First-Time Homebuyers” will not get cited, while a first-hand piece like “Why We Waived the Inspection and Saved Money” will. The differentiator is non-commodity, experience-based content with a genuine point of view. Recycled, summarized industry content is exactly what AI synthesizes instead of citing.

Two authority moves are specific to 2026:

Preferred Sources. Google now lets users designate sites as Preferred Sources, making that content more likely to appear for them in AI Mode and AI Overviews. Prompt your existing audience — newsletter subscribers, returning readers, social followers — to add you. They are the people most likely to do it.

Earned media over manufactured mentions. AI engines weight independent, authoritative corroboration heavily — press placements, genuine reviews, and citations from trusted publications act as anchor points during retrieval. But Google has warned that inauthentic mention-building is unlikely to help, because its generative features rely on the same spam systems as core ranking. Pursue real authority, not synthetic citation farms.

Measure it — the genuinely new skill

The hardest part of GEO/AEO in 2026 is not optimization; it is measurement, because AI answers are invisible in standard reporting. Build a measurement loop around four signals:

• Citation tracking. Use a citation monitor (Profound, Otterly, and similar) to track how often each engine names your brand for your target prompts across ChatGPT, Perplexity, Gemini, and AI Overviews. Being indexed is not the same as being cited — track the gap.
• The AI Overview signature in GSC. Pages ranking in positions 1–5 with anomalously low CTR (well below the expected curve) on queries with meaningful impressions are very likely losing clicks to an AI Overview. AI Overviews can absorb an estimated 47–60% of clicks on affected queries, and Google does not label which queries are hit — so this signature is your detection method. (Note: a GSC impression-logging bug affecting May 2025–April 2026 data was fixed in 2026 but historical data was not corrected; re-baseline accordingly.)
• Fan-out coverage. Run your hub topics through a fan-out tool periodically and track how many surfaced sub-queries your page actually answers. Rising coverage predicts rising citations.
• AI-referral traffic. Watch for referral upticks from AI platforms in analytics, and confirm your search crawlers are being served by checking CDN/server logs by user agent.

The dead tactics to stop doing in 2026

Half the “AI SEO” advice still ranking on page one is now wrong. Retire these:

• llms.txt as a ranking lever. Google has said on record it ignores llms.txt, and studies across 300,000+ domains and 500M+ AI bot events show no measurable citation lift from the file alone. It is harmless to ship for the agentic web and for some smaller AI surfaces, but do not pay for it as a Google/AI-Overview tactic and do not prioritize it over crawler access and content quality.
• Content “chunking” for AI. Google states there is no requirement to fragment content into tiny pieces; its systems extract the relevant passage from a normal, well-structured page. Write self-contained sections for clarity — not artificial micro-chunks.
• FAQ schema for rich results. Removed from Search in May 2026. It no longer generates a SERP feature.
• AI-specific keyword rewriting. AI understands synonyms and meaning; you do not need to capture every long-tail variant or write in robotic “AI-friendly” phrasing.
• Keyword stuffing. It reduced visibility in the Princeton testing. It is worse than useless for AI.

Your 30-day action plan

1. Days 1–3 — Audit access. Check robots.txt isn’t blocking OAI-SearchBot, Claude-SearchBot, PerplexityBot, or Googlebot. Confirm pages are indexable and fast.
2. Days 4–7 — Fix your entity home. Standardize your brand name everywhere, add/clean sameAs and Organization schema, and create or update your Wikidata entry.
3. Days 8–14 — Map fan-out and restructure. Run your top hub pages through a fan-out tool, add a dedicated H2 (opening with a direct 134–167 word answer) for every uncovered sub-intent until each hub covers five or more.
4. Days 15–21 — Apply the Princeton techniques. Add specific statistics with named sources, credible quotes, and outbound citations to authoritative sources on every priority page.
5. Days 22–26 — Strengthen authority. Publish at least one first-hand, non-commodity piece; line up one genuine earned-media placement; prompt your audience to add you as a Preferred Source.
6. Days 27–30 — Stand up measurement. Connect a citation tracker, flag low-CTR/high-rank pages in GSC, and set a quarterly refresh cadence with visible update dates.