How to Show Up in ChatGPT, Gemini and Perplexity

Traditional SEO is about earning a spot on a page of blue links that a person then scans. Generative engine optimization, or GEO, is about being the answer an AI assistant gives when someone asks it a question directly. When a customer in Manila types 'best halal ramen near BGC' into ChatGPT, they do not get ten links. They get a short list of named places, and either you are on it or you are invisible.

The mechanics differ. A search engine ranks pages. A generative assistant reads across many sources, then synthesizes a single recommendation in its own words. You are no longer optimizing to be clicked. You are optimizing to be cited, quoted, and named.

The good news is that GEO and SEO are not enemies. Almost everything that makes you legible to Google also makes you legible to the models. The difference is emphasis: clarity, consistency, and structure matter even more when a machine is summarizing you instead of a person deciding for themselves.

No assistant knows your shop from memory. When you ask about local businesses, these tools reach for live data through search and retrieval, then blend it with what they learned during training. That means the sources feeding them are, roughly, the same public web the rest of us can see.

For local queries specifically, a handful of sources carry most of the weight.

• Structured web data on your own site, especially machine-readable schema markup.
• Your Google Business Profile, still the single richest structured record of a local business.
• Map and place databases such as Foursquare, Apple Maps, and Bing Places, which quietly power many downstream apps and models.
• Reviews and their actual text, on Google, Facebook, and niche directories.
• Third-party directories, listicles, and 'best of' articles that mention you by name.

Perplexity leans heavily on live web retrieval and cites its sources openly, so you can often see exactly which pages it read. Gemini can draw on Google's own local index. ChatGPT blends browsing with training data. The precise recipe each one uses is proprietary and shifting, but the inputs overlap enough that a single, well-built local presence serves all three.

That is the strategic relief in all of this. You do not need three separate playbooks. You need one clean, consistent, and complete public footprint that any retrieval system can read without getting confused.

Models are confidence machines. When your name, address, and phone number match everywhere they look, an assistant can state your details plainly. When sources disagree, the model hedges, picks the wrong branch, or drops you for a competitor it feels surer about.

Completeness compounds. A profile that spells out your category, hours, services, prices, and payment methods gives an assistant more true things to say about you, which makes it far likelier to surface you for a specific question like 'which clinic in District 1 opens on Sunday and takes cards'.

• Pick one exact NAP format and use it letter for letter on every platform.
• Fill every field: category, services, hours, holiday hours, attributes, and a plain description.
• Keep hours and holiday closures current, since stale data teaches the model to mistrust you.
• Add the details customers actually ask about, such as halal, GCash, wheelchair access, or parking.

If structured data tells a model the facts, reviews and citations tell it the reputation. Assistants weigh not just your star average but the language in your reviews, because that text is what they paraphrase back. Reviews that mention specific dishes, services, or neighborhoods give the model concrete reasons to recommend you.

Citations, meaning other credible pages that mention your business by name, matter just as much. When a respected local blog lists you among the best coffee spots in Cebu, that sentence can become the very sentence an assistant repeats. Being named in third-party 'best of' content is one of the most direct GEO levers available today.

Keeping dozens of listings identical by hand is where most businesses quietly fall apart. This is the tedious work Mapkita is built for: pushing one clean, consistent record across the many sources AI assistants read, and flagging where you drift out of sync.

• Encourage genuine, detailed reviews rather than one-line ratings.
• Reply to reviews in natural language, which adds relevant, quotable text to your profile.
• Get listed in reputable local directories and rounded-up articles for your city and category.
• Never buy reviews or mentions, since fakes get filtered and erode the trust you are trying to build.

A surprising number of local businesses in the region run entirely on a Facebook Page with no website at all. That works for humans, but it leaves the models thin on structured data they can crawl and cite. A simple, fast, crawlable site is still your strongest owned asset for GEO.

The point is not a fancy design. It is machine-readable clarity. Put your NAP in plain text, add LocalBusiness schema, and answer the real questions customers ask, in the words they use.

• Show your name, address, and phone in real text, not baked into an image.
• Add LocalBusiness or Restaurant schema markup with hours, geo coordinates, and price range.
• Write an honest FAQ answering the questions people actually ask an assistant.
• Make sure the site is fast, mobile-first, and not blocking crawlers in robots.txt.

Any plan for this has to start from how Filipinos actually search. The Philippines is mobile-first to an extreme, and for millions of customers the internet effectively begins inside Facebook, Messenger and Viber rather than a browser.

That shapes two things. First, your Facebook Page is often your most-seen listing, so its details must match your Google Business Profile exactly. Second, as assistants get embedded into the apps people already live in, the businesses with clean, consistent public data will be the ones those assistants can confidently name.

• Treat your Facebook Page as a primary listing, not an afterthought, and keep its NAP identical to Google.
• Do not neglect Google Business Profile even in Facebook-heavy markets, since it feeds the models directly.
• Use local languages and the mixed English many customers type, such as Taglish or Bahasa-English.

You do not need to wait for the dust to settle. The concrete steps that help today are the same fundamentals that will still matter when the tools mature, which makes this rare low-risk ground.

That last step below is underrated. Query the assistants the way your customers would, note who they name and which sources they cite, and you have a free, direct read on your AI visibility. Tools like Mapkita now fold this into an AI-readiness score so you can track it over time rather than checking by hand.

• Audit your NAP across Google, Facebook, Apple Maps, and every directory, then fix mismatches.
• Complete your Google Business Profile to 100 percent, including attributes and photos.
• Publish or refresh a crawlable website with plain-text NAP and LocalBusiness schema.
• Build a steady flow of detailed, genuine reviews and reply to them.
• Earn mentions in trusted local directories and 'best of' articles in your niche.
• Test it: ask ChatGPT, Gemini, and Perplexity for recommendations in your category and city, and see whether you appear.

Honesty matters here, because the field is young and full of confident guesses. What is well established is that clean, consistent, complete structured data and a strong review profile help you across Google, maps apps, and AI assistants alike. That foundation is not speculative.

What is still emerging is the precise weighting each assistant gives to each source, how often they refresh, and how much a single directory mention really moves an answer. These change quietly and without announcement. Be wary of anyone selling guaranteed 'rank number one in ChatGPT' packages, because no such guarantee exists.

The sane strategy is to invest in the durable fundamentals now and treat GEO as an extension of good local presence, not a separate dark art. The businesses that show up in AI answers a year from now will mostly be the ones that were already easy for a machine to understand.