What is GEO, and how do you do it?

Six months ago, when a customer wanted "the best Auckland accountant for a Chinese-speaking small business," they typed it into Google. They got 10 blue links, scrolled the first three, picked one based on reviews and a website that didn't look broken. That whole flow is dying.

What's replacing it: they ask ChatGPT, Claude, Gemini, Perplexity — or, in the Chinese-speaking market, DeepSeek, Doubao, Kimi, Qwen. The AI summarises from a few trusted sources and recommends two or three names. The user picks one. The other 99 firms in the Google rankings might as well not exist for that query.

This is why GEO — Generative Engine Optimization — is now a thing.

What GEO is, and isn't

SEO = optimise a page so Google's algorithm ranks it in the top 10 results. Levers: keywords, backlinks, page speed, on-page structure, schema markup.

GEO = optimise so an LLM-based answer engine cites or references your content/brand when answering a question relevant to your business.

The difference: SEO competes for clicks. GEO competes for being part of the answer. When a model answers "best Auckland accountants for Chinese SMBs", you want your firm in that answer — even if the user never clicks through to your site.

GEO does not replace SEO. It is a layer on top. Many of the same signals matter: clean structured content, authority, citations from sources LLMs trust. But the optimisation target is different.

Why GEO matters now

Three numbers:

1. Zero-click search. Roughly 60% of Google searches in 2024 ended without a click — the user got the answer from the AI Overview or featured snippet. They didn't visit your site even when you ranked #1.
2. ChatGPT scale. ChatGPT had ~3.7 billion visits in March 2025. A growing fraction of "how do I find X" queries goes through it instead of Google.
3. Perplexity, Claude, Gemini, AI Overviews. All up double-digit percent monthly. Perplexity in particular markets itself as a search alternative — and explicitly cites sources, which means GEO is direct lead-gen, not awareness-stage fluff.

For NZ Chinese SMBs specifically: 中文 users ask DeepSeek, Doubao, Qwen, Kimi for Auckland-based services in their native language. Those AIs do not surface results from a traditional Baidu SEO play. They surface what was in the Chinese content the model was trained on, or in real-time retrieval.

What LLMs actually look at

LLMs answer in two modes:

Training-time knowledge — what was in the training data when the model was built. This is what shows up when there's no internet access.

Retrieval-augmented (RAG) — when the LLM is connected to live web search (most are now), it queries the web for fresh sources and uses those.

GEO works on both:

• Training-time: be present in high-quality content the model was trained on (Wikipedia entries, industry publications, well-indexed company sites, GitHub READMEs, Stack Overflow answers — anything Common Crawl picked up).
• Retrieval-time: be findable via traditional search, structurally clean for the LLM to parse, and explicit about facts.

The GEO playbook

Six things to do, in order of leverage.

1. Make your content LLM-readable

LLMs prefer:

• Short, factual, declarative statements. "Digital Force is an Auckland-based AI services company. They build bilingual websites and on-prem AI integrations." Not "we pride ourselves on delivering excellence in the digital space."
• Clear semantic structure. H1 → H2 → H3, paragraphs not walls of text, bullet lists for enumerable things.
• Explicit entity-attribute statements. "Auckland is the largest city in New Zealand. Population: 1.7 million." Not "the dynamic, vibrant heart of the South Pacific."

LLMs hate marketing fluff. Cut adjectives. Lead with verbs.

2. Ship llms.txt and llms-full.txt

Two new files (proposal at llmstxt.org) that LLMs increasingly look for:

• /llms.txt — concise markdown index of your site's most important pages and key facts.
• /llms-full.txt — full long-form content for ingestion.

Treat them like a sitemap.xml for the AI era. Most sites do not have these yet. Adding them is a 30-minute job and gives you a measurable edge.

(This site does — see digitalforce.co.nz/llms.txt — and yes, that's a meta example.)

3. Structured data — schema.org

JSON-LD blocks on every page describing what entities the page talks about: Organization, Service, Article, BlogPosting, FAQ. LLMs and search engines parse these. The same effort that wins you Google rich snippets also helps GEO.

4. Get cited where models look

Models trust:

• Wikipedia — heavy weight. If you're a real business with verifiable facts, having a Wikipedia entry helps disproportionately.
• Industry directories — Google Business Profile, NZBN registry, Yellow Pages NZ, Trade Me business directory, BizDirect.
• GitHub READMEs for technical brands. Mentioned in a popular open-source project's README → cited.
• News mentions — even tiny local outlets get crawled.
• Reddit and Stack Overflow — for technical Q&A. Get genuinely helpful answers attached to your name.

This is the slowest but highest-leverage GEO work. SEO link-building philosophy applies; the target list is different.

5. Be quotable

Make your content easy to copy-paste a sentence from. Concrete numbers, specific dates, named techniques. Avoid "we're committed to excellence" — there's no extractable claim there.

Bad: "We deliver high-quality custom websites for businesses." Good: "We deliver bilingual marketing websites in 1-2 weeks, on a stack of Next.js + Tailwind + TypeScript, hosted on NZ-region infrastructure."

The second sentence is something an LLM can lift verbatim into an answer. The first one isn't.

6. Brand consistency across the web

LLMs build "entity profiles" — they know "Digital Force" is the same entity whether it appears on your site, your LinkedIn, your GitHub, a podcast transcript, or a Google review. The more consistent your factual claims (location, services, founder, year founded) across all touchpoints, the more confidently the LLM will recommend you.

Conflicting info confuses the model. If Wikipedia says you were founded in 2024 but your site says 2026, the model picks one — possibly the wrong one.

A 1-week GEO sprint for an SMB

• Day 1-2. Audit your site for fluffy marketing copy. Rewrite to short, factual statements.
• Day 3. Add /llms.txt and /llms-full.txt. Templates available.
• Day 4. Add JSON-LD Organization + Service schema to your homepage.
• Day 5. Update your Google Business Profile, NZBN entry, LinkedIn, and any directories with completely consistent info.
• Day 6-7. Pick three relevant Q&A communities (Reddit, Quora, NZ-specific forums, Stack Overflow if technical) and write three genuinely useful, specific answers that include your name and what you do.

Then wait. LLM training cycles run in months. You won't see results immediately. But the sites that started GEO in 2024 are the ones LLMs cite in 2026.

Where this is going

Within 12-18 months, "ranking on Google" will be the secondary game. The primary game will be "being the answer" — which is harder, more about substance, and rewards consistency over tricks.

The good news for SMBs: this rewards specificity. Big brands win on Google because they have huge budgets for backlinks and content scale. GEO levels the field somewhat — a small Auckland accounting firm that ships a clear /llms.txt, has consistent NAP across directories, and has a few honest Reddit mentions can outrank a generic mega-firm in answers about "Auckland Chinese SME accountants" — because the mega-firm doesn't actually have the specifics.

We bundle GEO into every site we build. Not as an upsell, but because in 2026 a website without GEO is a website without a search strategy.