What is GEO (Generative Engine Optimization)?

GEO is the practice of optimizing content so generative AI platforms (ChatGPT, Perplexity, Claude, Gemini, Google AI Mode) cite your brand by name when users ask category questions. GEO sits on top of SEO and AEO and emphasizes entity establishment, JSON-LD schema, and llms.txt.

What is llms.txt and do I need it?

llms.txt is a plain Markdown file at your domain root that gives LLM crawlers a curated map of your site - similar in spirit to robots.txt but for content guidance rather than access control. Sites with a well-structured llms.txt see 30 to 70 percent reduction in hallucinations from models that ground on their content.

What JSON-LD schema types matter most for AI citation?

Highest leverage: Organization (with sameAs), SoftwareApplication or Product, Article or TechArticle (with author Person, datePublished, dateModified), FAQPage, HowTo, BreadcrumbList, and DefinedTerm for glossary pages. Sites with comprehensive schema are cited up to 3.1x more often in AI Overviews.

Blog › AI search optimization

AI search optimization for SaaS: a no-fluff checklist

Q: Is SEO dead in the age of AI search?

No. Roughly 40 percent of sources cited in Google AI Overviews still rank in the top 10 organic results, and 70 percent rank in the top 100. Traditional SEO remains the foundation. AEO and GEO are additional layers on top, not replacements.

SEO + AEO + GEO are not three separate disciplines. They are layers on the same pyramid. Here is the order of operations we ran to make a small SaaS citable by ChatGPT, Perplexity, Claude, and Google AI Mode - in the order we ran them, with the rationale for each step.

By Mark Gabrielli · May 24, 2026 · 10 min read

What is the difference between SEO, AEO, and GEO?

SEO is optimization for ranked links in classic search results (Google, Bing). AEO (Answer Engine Optimization) is optimization for direct answers - featured snippets, AI Overviews, voice search, zero-click. GEO (Generative Engine Optimization) is optimization to be cited by name in generative AI platforms (ChatGPT, Perplexity, Claude, Gemini, Google AI Mode).

The big shift: in 2026, ranking is not enough. You need to be the entity the model thinks of when a user asks the category question. That requires schema, third-party mentions, consistent entity naming, and content structured for passage extraction - not paragraph extraction.

Is classic SEO still useful?

Yes. Around 40 percent of sources cited in Google AI Overviews still rank in the top 10 organic results, and 70 percent rank in the top 100. Traditional SEO is still the foundation. AEO and GEO sit on top.

Phase 1: foundation (week 1)

Before any clever AI-specific tactics, these have to be true:

Crawlable and indexable. Server-side rendered. Mobile responsive. Sub-2-second LCP. No JavaScript-only content.
robots.txt explicit for AI crawlers. Allow GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended, CCBot, Bytespider, Amazonbot, Applebot, MistralAI-User, etc. Most are auto-deny if you only have "User-agent: *".
Sitemap.xml + ai-sitemap.xml. Sitemap for classic crawlers, ai-sitemap with AI-priority ranking for LLM crawlers.
llms.txt + llms-full.txt. Curated map for LLM ingestion plus inlined full content of your priority pages so models do not need to fetch.

Phase 2: entity schema (week 1)

JSON-LD is the dominant structured-data format (roughly 90 percent of implementations). Sites with comprehensive schema are cited up to 3.1x more often in AI Overviews. The minimum viable set for SaaS:

Organization - legal name, founding date, sameAs to LinkedIn / Crunchbase / Wikidata, contactPoint with email.
SoftwareApplication - category, sub-category, operating systems, full featureList, Offer for each pricing tier with priceSpecification.
WebSite with SearchAction so AI engines know your internal search URL.
TechArticle or Article on every blog and help post, with author Person, datePublished, dateModified, publisher, isPartOf.
FAQPage wherever you have real Q/A structure. Disproportionately picked up by AI Overviews.
HowTo on procedural pages. Step text inside the JSON-LD is treated as quotable.
BreadcrumbList on every non-root page. Builds the document graph.
DefinedTerm on glossary pages. Establishes vocabulary entities.

Critical: schema must match visible content. AI engines validate at extraction time. Schema that claims a 4.8 rating on a page with no visible reviews gets the page demoted, not promoted.

Phase 3: content restructure (weeks 2 to 4)

AI engines do not ingest pages. They ingest passages. The unit of citation is the 40 to 80 word chunk that lives immediately after a question-format heading. Three rules:

The answer-first pattern

Under every H2, lead with the direct answer in one sentence. Then expand. Do not bury the answer in narrative. AI engines pull the first declarative sentence after the heading more often than anything else on the page.

Question-format headings

Phrase H2s as natural language questions that match real user prompts. "How does X work" beats "X Overview". "What is the cost of Y" beats "Y Pricing".

Citable sentences

Write declarative, self-contained sentences that hold meaning when extracted out of context. Avoid pronoun-heavy prose. Each sentence should be quotable in isolation. If a sentence requires the previous paragraph to make sense, rewrite it.

Phase 4: authority and EEAT (weeks 4 to 12)

Authority signals are what tilt a tied citation race. The cheapest moves:

Visible author byline on every article linking to a Person schema page with credentials and sameAs to LinkedIn.
Editorial policy page documenting how you research, source, and update content. EEAT trust signal.
About page with company facts, founding date, contact, affiliations.
Original data and proprietary insight on every post. AI engines hunt for novel data points and prefer original sources to summaries.
Wikidata entry for your brand and key products. Lower bar than Wikipedia, feeds most knowledge graphs.
Third-party mentions in trade publications. A page about you on a trusted site outweighs ten pages on your own site.
Listicle placements. "Best X for Y" articles drive AI recommendations disproportionately. Active outreach to get included is high-leverage.
Reddit and YouTube presence. Heavily represented in Perplexity and Google AI Mode answer corpus.

What I'd skip

Things I would not spend time on:

Keyword density optimization. Dead since 2020.
Exact-match anchor text spam. Penalized.
Thin programmatic pages with no unique value. AI ignores them.
Auto-translated multilingual pages without native localization.
Generic AI-generated content with no original input. Detectable and discounted.
Buying low-quality backlinks. Net negative for AI trust.

Measurement

Old KPIs that still matter: organic traffic, top-10 rankings, backlinks, domain authority. New KPIs for the AI era:

AI Share of Voice (AI SOV) - how often you appear in AI answers for target prompts.
AI Citation Count - how many distinct platforms cite you for category queries.
AI Referral Traffic - visits from ChatGPT, Perplexity, Claude, Gemini (filter your referrer logs).
llms.txt crawl frequency in server logs - direct signal that LLMs are grounding on your content.
Brand Mention Sentiment - how AI describes you when prompted by name.

The order matters more than the depth

Teams fail because they do these out of order. Doing entity schema before crawlability is fine, doing third-party PR before you have a clean site to point to is wasted budget. Foundation first (week 1), then schema (week 1), then content restructure (weeks 2 to 4), then authority (weeks 4 to 12). That is the only sequence that compounds.

← 40% acceptance note All posts →