GEO Optimization: How to Get Cited by ChatGPT in 2026

Every B2B buying decision in 2026 starts the same way. Someone types a question into ChatGPT, Perplexity or Gemini. They read the answer. If one company gets a confident recommendation, they click that company's site. Google gets opened only if the AI answer was vague or hedged.

That is not a forecast. It is already how half the internet shops for considered purchases, and the share is climbing. Here is the uncomfortable part — most Latvian websites, and most websites globally, are invisible to AI search. Not because they are bad. Because they were built in an era when nobody was optimizing for language models.

GEO — generative engine optimization — is the work of making AI tools actually mention you. This post covers the five signals that matter, a ten-prompt audit you can run on yourself in ten minutes, and when GEO is a waste of your money.

SEO and GEO, in one breath

SEO was about getting Google to show you. GEO is about getting ChatGPT to cite you.

When Google shows a page, the user clicks. When ChatGPT answers, the user doesn't click anywhere — they get a summary and maybe one company name. If you are that name, you win. If you are not, you were not in the conversation.

AI engines don't grade the same signals Google does either. There is a new checklist, and it is short.

Five signals AI engines actually reward

1. llms.txt — if you don't have one, you don't exist

A markdown file at /llms.txt that tells AI agents who you are, which pages are canonical and where the structured content lives. Think robots.txt, but written for models instead of crawlers.

Spec is at llmstxt.org. Implementation is a fifteen-minute job. The payoff — when ChatGPT or Perplexity looks at your domain, it understands your business instead of guessing from scraped HTML.

Quick test — type yourdomain.com/llms.txt into a browser. If you get a 404, you don't have the baseline. Most Latvian sites fail this test. Most sites globally fail this test.

2. JSON-LD Organization schema, filled out completely

Not a stub with a name and a logo. A full ProfessionalService or LocalBusiness object with address, phone, opening hours, service catalog, expertise, service area. Machine-readable facts about your company, in a package a model can consume in a single pass.

Two fields do most of the work — knowsAbout and hasOfferCatalog. The first declares your areas of authority. The second declares what you sell. Without them, an AI model classifies you as generic company in Latvia. With them, you are the firm that does Next.js, Laravel and KNX in the Cēsis area.

Which version gets recommended when somebody asks — who does KNX in Latvia?

3. Article and FAQPage schema on content pages

Every blog post needs Article schema with author, date and category. If the post contains question–answer pairs, add FAQPage schema. These are the exact structures Perplexity and Gemini prefer when deciding whether to cite you.

On bitbox.lv we generate them automatically — any H3 that ends in a question mark followed by a paragraph becomes an entry in the FAQPage structure. Readers see a normal article. AI sees a clean, parseable Q&A document. One source, two consumers.

4. E-E-A-T signals — real author, real address, real phone

Google's E-E-A-T framework — Experience, Expertise, Authoritativeness, Trustworthiness — is now an AI search signal too. Before citing you in an answer, a model checks for several things.

A named author with visible expertise — not Editorial Team. A real street address that matches your legal registration. A phone number answered by a human. A tax ID or registration number. Even a founding date. These are the details a thin operation can't fake in a weekend.

It is not just trust theater. When a model is asked — recommend a trustworthy X in Latvia — it actively discounts pages with no author, no address and no credentials. Even if the content itself is strong.

5. Citable sentence structure

AI models don't quote whole pages. They lift sentences. If your page contains no standalone, citable sentence, you literally cannot be quoted — even if the model wants to.

Good: A typical KNX installation in a four-room Latvian apartment costs €8,000 – €14,000. Bad: Smart home is an exciting technology that can do many things for your home.

The first has a number, a geography and an application. The second is a blog intro with no payload. Models simply drop sentences like the second.

The ten-prompt visibility audit

These are the prompts we run against every client's domain before a GEO engagement. Try them yourself — open ChatGPT (or Perplexity, or Gemini), paste them in, see whether your company shows up.

1. Recommend a [your niche] company in Latvia.

2. Which [your niche] in Latvia works well for small businesses?

3. [Your niche] in Riga / [your region] — top five.

4. Compare [competitor A] versus [competitor B] in Latvia.

5. Where in [Latvia / Riga / your city] can I get [your specific service]?

6. Recommend an expert on [your specialization].

7. Which [your niche] companies operate outside Riga?

8. Which Latvian [your niche] works with [specific technology]?

9. [Your company name] — reviews and track record.

10. How should I pick a [your category] vendor in Latvia?

Appear on fewer than three? You have a GEO problem and you should fix it this quarter. Appear on eight or more? You are already in the minority of Latvian businesses ready for AI-first search — and the compounding starts now.

When GEO is a waste of money

The honest section. GEO doesn't fit everyone, and we will tell you to skip it if your situation is one of these.

Local business with a physical storefront — hairdresser, auto shop, bakery, café. Google Maps and Google Business Profile will return ten times more than GEO. Start there.

Extremely small niche with thin global search volume. If only twenty people worldwide are looking for what you do, optimizing for AI won't create scale that wasn't there.

Impulse purchases. If customers buy because of an Instagram ad rather than a research session, GEO adds nothing on top.

GEO works when the customer researches before buying. B2B services, technically complex solutions, considered B2C purchases above €500, medical, legal, financial services, technology firms. Also the smart-home and KNX-integration space — buyers spend weeks comparing options and reading experience reports. In those categories, GEO investment pays back inside three to six months.

What we built on our own site

Want to see it in practice? Open bitbox.lv/llms.txt. That is our live llms.txt, regenerated hourly from the CMS. Then open any story and view page source — you will find three JSON-LD blocks: Article, BreadcrumbList and FAQPage (when the content contains FAQ structure).

It is all generated automatically. The editor writes a normal post. The reader sees a normal page. The AI crawler sees a machine-readable knowledge graph. One source, three consumers.

Technically it is about 200 lines of Next.js code. On WordPress, Rank Math or Yoast (free tiers) cover roughly 70% of the job. On a custom stack, it is easier than it sounds. The real question is whether your team knows what AI engines expect, not just what Google expects. Our SEO service includes this from the start — but you can do it yourself if you have engineering bandwidth.

Why now

The window to be among the first companies AI recommends in your niche is roughly six months wide. Businesses that get this sorted by mid-2026 will spend the next two years collecting free AI-driven traffic. Businesses that wait until 2027 will be paying for ads to claw back positions they could have earned for free today.

It is the same compounding advantage as being the first serious Facebook advertiser in 2012, or the first business doing Latvian-language SEO in 2015. Early movers pay less and hold longer.

Frequently Asked Questions

What does GEO mean and how is it different from SEO?

GEO stands for generative engine optimization — the work of getting language models (ChatGPT, Perplexity, Gemini, Claude) to cite you in their answers. SEO targets the Google results page; GEO targets the AI-generated answer itself, where there is no click to a link — only a mention.

Do Latvian businesses really need to care about this right now?

Yes — if your customers research before buying. B2B services, technical solutions, medical, legal, financial advisory, smart-home integrators. For these niches AI search is already a starting point today. If you are a neighbourhood café, wait another two years.

How long until GEO work shows results?

First citations in ChatGPT and Perplexity answers typically appear four to eight weeks after the technical changes — faster than classical SEO. Full impact lands in three to six months, provided the site is crawled regularly and not blocked by robots.txt.

Does GEO replace regular SEO?

No. Both need to run in parallel. Most signals overlap — speed, structured data, clear content hierarchy. But some do not: llms.txt and citable sentence structure are ignored by classical SEO. In 2026 the right strategy is SEO as the foundation and GEO as the next layer on top.

What's next

If you want a straight GEO audit on your domain, tell us. We don't send 40-page reports. We check the five signals, run the ten-prompt test against your niche, and hand back a prioritized fix list. No jargon, no hype, no long contract just to check whether ChatGPT knows your name. Here is our SEO and GEO service.

You don't need a long engagement to find out whether ChatGPT has heard of you.