Why ChatGPT, Gemini and Perplexity Never Recommend Your Business (and How to Fix It)

Ask ChatGPT to recommend a web designer in your neighbourhood, or get Perplexity to name a good accounting firm in Singapore, and watch what happens. It names someone. It just won't name you. That stings, because the businesses these tools surface aren't always better than yours, they're just more legible to a machine that reads the web differently than a human does.

This isn't a conspiracy and it isn't bad luck. AI assistants pick what to recommend based on signals you can actually influence. Most SMEs simply haven't sent those signals. The good news: the fix is concrete, it's mostly free, and a non-technical owner can either do it or hand a clear list to a developer. Here's what's actually happening, and what to do about it.

Why you're invisible to AI

These models don't browse the way you think. When you search Google, a human-ish ranking system shows you ten blue links and you decide. When ChatGPT, Gemini, or Perplexity answer a question, they synthesise one answer from sources they already trust, their training data, plus whatever they pull in live from search and a handful of authoritative sites. There's no page two. There's one answer, and you're either in it or you're not.

Four things usually keep you out:

• No structured data. Your site might say you're a Paya Lebar accounting firm open Mon-Thu, but if that's only written in a paragraph a human reads, a machine has to guess. Structured data (Schema markup) states it in a format machines parse without guessing, your name, address, services, hours, FAQs, all labelled.
• Weak or inconsistent entity presence. "Entity" is just the machine's mental model of who you are. If your business name, address and phone number (NAP) differ across your website, Google Business Profile, and directories, different unit number here, old number there, the model can't form a confident picture. Confidence is what gets you cited.
• You're not in the sources these models read. LLMs lean on places they consider reliable: established directories, reputable local publications, structured business listings. If you exist only on your own site and an abandoned Facebook page, you're missing from the rooms where the decision gets made.
• Your content doesn't answer the question. A lot of SME copy is mood, "We are passionate about delivering excellence." A machine answering "who does affordable e-commerce websites in Singapore?" can't use that. It needs a direct, plain answer to a real question.

Notice none of this is about being the cheapest or the biggest. It's about being understandable and verifiable.

GEO and AEO, in plain English

You'll see two acronyms thrown around. AEO (Answer Engine Optimisation) means structuring your content so a machine can lift a clean, correct answer straight off your page. GEO (Generative Engine Optimisation) is the broader game: making sure your business shows up inside AI-generated answers, wherever the model sources them.

Think of old-school SEO as getting ranked. GEO and AEO are about getting quoted. Same instincts, be clear, be credible, be present, pointed at a reader that doesn't scroll.

The fix, in order

Do these roughly top to bottom. The early ones cost nothing but effort and move the needle most.

1. Write answer-first, in clean semantic HTML. For your key pages, lead with the answer, then explain. "What does it cost to build a website with you? Most SME projects run between X and Y, depending on scope." Use real headings (<h1>, <h2>), real lists, real paragraphs, not a wall of <div>s styled to look like structure. Machines read the tags. If your developer built the site as one big styled blob, that's worth fixing.

2. Add Schema / JSON-LD markup. This is the single most technical step and the one most worth asking a developer for. At minimum you want Organization (who you are), LocalBusiness (where, when, NAP), FAQPage (your common questions and answers), and Article markup on blog posts. It's a block of structured code in the page that spells out the facts. Validate it with Google's Rich Results Test once it's in.

3. Make your NAP identical everywhere. Pick one exact format for your name, address and phone, then make it match across your website, Google Business Profile, and every directory you're listed in. Down to the unit number. This consistency is what lets a model merge all those mentions into one confident entity, you.

4. Claim and fill your Google Business Profile. It's free, and it's one of the most-trusted local signals there is. Complete every field, pick accurate categories, keep hours current. For a Singapore SME this is non-negotiable groundwork.

5. Earn mentions in sources AI trusts. A relevant industry directory, a local business publication, a partner's site, a genuine guest article, each real mention outside your own domain strengthens your entity. One credible third-party mention beats fifty self-published ones. This is slow, ongoing relationship work, not a button you press.

6. Get reviews, and respond to them. Reviews are public, third-party, and rich with the exact words customers use to describe you. They feed both the trust signal and the language a model associates with your name. Ask happy customers; reply to all of them.

7. Publish an llms.txt. This is a newer, still-emerging convention, a simple text file at the root of your site that points AI crawlers to your most important pages and facts, the way robots.txt guides search crawlers. Adoption isn't universal yet, so treat it as cheap insurance rather than a magic switch. Worth doing; don't expect it to carry the whole load.

8. Keep the information fresh. Stale hours, last year's pricing, a "© 2023" in the footer, these quietly erode confidence. A site that's visibly maintained reads as more trustworthy to both people and machines. Set a recurring reminder to review the facts each quarter.

The honest part

This is ongoing work, not a one-time hack. Anyone selling you "guaranteed AI ranking" for a flat fee is selling you something, the engines change their behaviour constantly, and nobody controls how a given model weighs its sources. What you can control is being the most legible, consistent, well-attested version of your business on the open web. Do that, and you stack the odds. Skip it, and you stay invisible by default.

A word of warning on numbers: you'll read confident statistics about how many people now "search" with AI. Be sceptical of specific figures, the honest answer is the behaviour is growing fast and the exact share is contested. You don't need a number to justify the work. The work is just good hygiene that also happens to help you rank in normal search. There's no downside.

If you want the broader picture of where this fits, alongside your website, your WhatsApp flows, and the rest of going digital, the Singapore SME digital transformation guide lays out the full framework.

One last thing, and I'll keep it light: we build this into every site we ship by default, semantic HTML, full JSON-LD, clean entity data, the lot. Not as an upsell, just because it's how a site should be built in 2026. If you're curious what that looks like in practice, our own services page and process are built on exactly what's described above. Poke around. Or if you'd rather just talk it through, say hello.

, Zyle