Appearing in ChatGPT and Perplexity from Panama: the step-by-step method

Open ChatGPT in another tab, type "best web design agencies in Panama" and watch what appears. If your company is in that sector, you will probably see a vague answer that mentions two or three generic names, global brands that do not operate in the country, or simply a refusal: the model will say it does not have enough specific information. The same happens if you try with "immigration lawyers Panama", "dental clinics Panama", "home builders Panama": almost any specific sector query about the country returns poor answers because no Panamanian provider is optimizing for this.

That is at the same time a problem and an opportunity. The problem is that your brand probably does not appear today when a potential client asks an AI about your sector. The opportunity is that arriving first at a channel where the competition has not even shown up is the most efficient way to build a medium-term advantage. This article is the concrete manual to do it: how to configure the site, how to create content the AI engines cite, how to measure if it is working, and the mistakes that kill citability before you start. At the end there is a section on how we apply exactly this method to this very site, with details of the process.

How AI engines really work on the inside

Before the method, it is worth understanding the mechanism. Modern AI engines do not work like Google, which returns links. They work with a process called RAG, Retrieval-Augmented Generation. The name sounds technical but the idea is simple: when a user asks a question, the engine first searches its index (or the web, depending on the engine) for the most relevant sources, extracts fragments from those sources, and then generates a synthesized answer citing or mentioning the sources it used.

The practical consequence of how RAG works is that the engine chooses which sources to cite before generating the answer. It does no miracles: if your site is not in its index, it does not appear in the answer. If it is in its index but is not structurally easy to extract, it does not appear either. If it is and is extractable but the model does not consider it reliable enough, it discards it in favor of sources with better authority signals. Each of these three layers —indexing, extractability, trust— is a bottleneck worth working on separately.

There are three engines that matter in Panama today. ChatGPT, with its 883 million monthly users and 2 billion daily queries, triggers a web search in approximately 31% of prompts; the rest of the answers depend on the model's training, which has a cutoff date. Perplexity is the easiest to "attack" because it crawls in real time and indexes new content within days. Gemini shares a good part of Google Search's behavior and reflects changes when Google reindexes. Each one has different citation patterns worth knowing.

The three search categories that matter in Panama

Not all searches on AI engines weigh the same for a Panamanian company. There are three categories worth optimizing specifically, in order of commercial priority.

Category one: generic local searches. "Best web design agency in Panama", "immigration lawyers in Panama", "dental clinics with good reviews Panama". These are the queries a potential client makes when they do not yet know any specific brand and are researching options. Gaining visibility here is the biggest traffic multiplier: each AI answer that includes your brand puts you in front of qualified people who had not yet arrived at your site. For a local service provider, this is the most valuable category.

Category two: specialty service searches. "Web design agencies that do ecommerce with Yappy in Panama", "data protection lawyers in Panama", "builders specialized in country houses Panama". The user already knows what they need and is filtering by specific characteristics. Here the traffic is lower but the intent is much higher: the person who arrives is already close to deciding. For companies with real specialties, this category is the one that converts best.

Category three: broad category searches. "Design studios in Panama City", "law firms in Panama", "technology companies in Panama". The user is exploring a whole sector, generally for market research, corporate projects or B2B decisions. The traffic is high but less qualified: many of these users are competitors, students, researchers. Even so, appearing here builds brand recognition and general positioning, which pays off over time.

To start, it is worth choosing between 20 and 30 representative prompts spread across the three categories, with emphasis on category one and two. These prompts will be your measurement baseline throughout the process: you will run them before starting to establish the starting point, and you will repeat them monthly to see the evolution.

The step-by-step method, part one: preparing the site

Preparing the site to be cited by AI engines has six layers worth working on in order. Skipping any does not break the method, but it delays the results.

Layer one: correct bot configuration. The most common mistake is blocking AI bots by default thinking of privacy and, without noticing, leaving yourself out of the platforms. OpenAI uses three different bots: GPTBot (training), OAI-SearchBot (live search within ChatGPT) and ChatGPT-User (when a user asks ChatGPT to explore a specific URL). Always allow OAI-SearchBot and ChatGPT-User; about GPTBot decide according to whether you want your content used for future training. Perplexity uses PerplexityBot, which is worth allowing. Gemini uses Google-Extended, which is also worth allowing if you want to appear in Bard and Gemini.

Layer two: complete structured data. AI engines read JSON-LD to resolve entities. Each page of your site should have at least Organization schema in the global head, BreadcrumbList in the navigation, and the page's specific schema (Service, Article, Product, LocalBusiness as appropriate). Pages with questions should carry FAQPage schema because it is the one the AI extracts best. Blog articles should carry Article with an identified author. Without these schemas the AI can read your content but has more trouble extracting it correctly. We cover this in depth in the guide to the 12 Schema.org types.

Layer three: direct-answer sections. The questions your potential client asks must have direct answers on your pages, not buried within long paragraphs. A good answer for AI starts with the conclusion in the first sentence, adds context in the following ones, and can be read as an independent block without depending on the previous or next paragraph. This format is called a "self-contained chunk" and is what the models extract when deciding what to cite. If your answers require reading the whole page to be understood, the AI prefers other more extractable sources.

Layer four: proprietary verifiable data. AI engines reward sources that contribute information that does not exist on other sites. An article or page that cites statistics with linked sources, offers comparisons with an explained methodology, or presents analysis with proprietary data has an advantage over content that repeats what is already on other sites. You do not need to produce academic studies: turning what you already know into citable format —price ranges of your sector, common mistakes you have seen, honest comparisons— is enough to differentiate.

Layer five: active freshness. Pages not reviewed at least every quarter are three times more likely to lose AI citations than regularly refreshed pages, according to data published by AirOps in 2025. The refresh does not mean rewriting; it means updating figures, adjusting temporal references, adding new information when the sector warrants it. Internally documenting the review calendar is what keeps this habit sustainable.

Layer six: external signals. The mentions of your brand on external sites, even without a link, are a direct authority signal for the AI engines. It is worth pursuing them actively: presence in sector forums, publications in specialized local media, participation in relevant Panamanian communities (LinkedIn Panama, professional groups, sector directories). The sources that most influence the training of the general models are Wikipedia, Reddit, Quora and large media; the ones that most influence real-time answers are sector directories and local media with authority.

The step-by-step method, part two: creating the content the AI cites

With the site prepared, the next step is to create or rewrite content specifically designed to be extracted by AI engines. There are five formats the public Omnia database identifies as the most cited in Perplexity and other engines, ordered by effectiveness.

Comparisons and reviews. Pages that compare options with clear criteria and an explained methodology have the best average citation position. An article like "The five payment gateways in Panama in 2026: an honest comparison" or "WooCommerce vs Shopify for Panamanian stores: when each one" is exactly the format the engines favor. Important: include real competitors and be honest about the strengths and weaknesses of each. Comparisons that only praise the own option are detected quickly and lose credibility.

Step-by-step guides. The models extract well-structured instructions frequently. "How to register your business in Panama", "How to get Google reviews without sounding pushy", "How to migrate from WordPress to Astro": formats where each step is clear and autonomous. The HowTo schema reinforces the extraction.

Lists with explained criteria. Not empty lists ("the 10 best companies in Panama") but lists with detailed objective criteria ("The five web design agencies in Panama by specialization, price and the speed of their own website"). The explicit criterion is what differentiates a citable list from an ignorable listicle.

Explicit FAQs. Pages with question-answer sections clearly marked with FAQPage schema are 3.2 times more likely to appear in Google AI Overviews. The question must sound exactly as a real user would ask it, not translated into a commercial tone.

Definitions and concepts. Pages that define sector terms with precision and local context appear frequently in explanatory answers. "What is Yappy and how does it work in Panamanian online stores?", "What does AEO mean and when does a Panamanian company need it?". The local Panamanian context is what differentiates your definition from the hundreds of generic ones that already exist.

The step-by-step method, part three: measuring with real prompts

Measurement is where most companies fail with AEO because the traditional tools (Google Search Console, Analytics, Semrush) do not show what happens on AI engines. Manual measurement with prompts is what gives real visibility into the progress, and it can be done free with dedicated time.

The measurement method has four components. First, a stable battery of prompts. Between 20 and 30 questions that represent what a potential client would ask about your sector. Spread across the three categories we mentioned before. Documented exactly so each repetition of the measurement uses the same prompts; changing the questions between measurements invalidates the comparison.

Second, clean sessions. The queries are made in incognito windows or sessions with no history to avoid the personalization that ChatGPT and other engines do according to your usage history. Without this, the results are biased by what the engine thinks you want to see, not by what it would show a new user.

Third, structured logging. A simple spreadsheet with columns for date, engine (ChatGPT/Perplexity/Gemini), exact prompt, whether the brand appeared (yes/no), type (citation with link or mention with no link), approximate position in the answer, sentiment (positive/neutral/negative). This sheet is updated each time the measurement is repeated.

Fourth, monthly cadence. Repeat the whole battery every 4 to 6 weeks. More frequent is a waste of time because the engines do not change that fast; less frequent loses visibility over regressions. Each repetition takes between 60 and 90 minutes for 30 prompts on 3 engines; it is manual work but small in hours and enormous in strategic value.

The real case: how we apply this method to this very site

It is worth closing with editorial transparency about how this site itself applies the method you just read. It is not marketing: it is the proof that what we recommend we apply, and that the method works in real Panamanian practice.

On technical configuration: this site explicitly allows OAI-SearchBot, ChatGPT-User, PerplexityBot and Google-Extended. About GPTBot we decided to allow it after weighing the cost and the benefit: the cost is that our content feeds OpenAI's future training; the benefit is the consistency of appearance in answers that do not trigger a live search. In our case the benefit exceeds the cost, but that decision is legitimate in both directions according to the sensitivity of the content.

On structured data: each page has its corresponding schema, the FAQs carry FAQPage schema, the blog articles carry Article with an author, the organization has global Organization schema, the navigation carries BreadcrumbList on all pages. Any schema inspector (Google's Rich Results Test, Schema Markup Validator) confirms that the site is legible for AI engines in its entirety.

On content: the main pages —services, industries, local coverage— are written with the direct-answer format the engines extract. Each section answers a concrete question a potential client would ask. The FAQs are not decorative; they are the real questions we receive repeatedly, with answers written to be extractable as autonomous blocks. The blog publishes articles with verifiable data, citations of external sources and a clear opinion about the Panamanian sector, deliberately avoiding the fillers that give away AI generation.

On measurement: we maintain an internal battery of Panamanian prompts about web design, SEO, AEO and related services, which we repeat in ChatGPT, Perplexity and Gemini every 4 weeks. The battery includes queries like "best web design agencies in Panama", "AEO Panama", "migrate WordPress in Panama", "web audit Panama", among others. We archive the results as internal reference and as proof that the claims of this article are not theory: they are practice we apply monthly with data we document.

On the result to date: as a relatively new site (published in 2026), we have not yet reached the critical mass of consistent citations a site with several years of accumulated authority has. But we already appear in Perplexity for several specific prompts about AEO in Panama, and the frequency grows each month we repeat the measurement. That is what is expected for a new site applying this method systematically: the first results appear in weeks, the consolidation takes months.

Common mistakes that kill citability before you start

To close, the mistakes we see repeatedly on Panamanian sites that invalidate any AEO attempt. Recognizing them quickly avoids spending time optimizing on broken foundations.

One, blocking all AI bots in robots.txt. Sites that copied a conservative template and blocked GPTBot, ClaudeBot, PerplexityBot by default. The result is complete invisibility in AI engines, guaranteed. Reviewing robots.txt is always the first check.

Two, auto-generated content in production. Pages written with AI generators and published with no serious review. The AI engines detect these patterns and discard these sources as unreliable. It is exactly the problem that motivated our filler-detection tool: if your content sounds like AI, the AI engines do not cite you.

Three, broken or incomplete structured data. JSON-LD with syntax errors, schemas that do not validate, incorrect type declarations. Any of these makes the engine ignore the whole schema, and in that case it is as if it were not there. Always validate with the Rich Results Test before assuming the schema works.

Four, decorative FAQs. "Frequently asked questions" sections with questions no one really asks, written in a commercial tone, with no FAQPage schema. The AI ignores them because they are not useful for extracting answers to real queries. Effective FAQs are the ones that answer the questions the potential client is already asking, in their language, correctly marked with schema.

Five, a slow site. If your site takes more than 3 seconds to load, several AI bots do not even complete the crawl. Speed is still a trust signal even for AI engines. Optimizing Core Web Vitals is not reduced to classic SEO: it also directly affects AEO.

Six, abandoning the method in three months. AEO requires patience. The companies that apply the method for a few weeks, do not see immediate results and give up, lose the moment where the results begin to arrive. The investment pays off cumulatively: the initial six months build the base, the following six start to generate consistent citations, a year and a half onward is where the effect becomes significant. Whoever starts now and keeps the method will have an enormous advantage when the rest of the Panamanian market discovers this layer, probably in 2027 or 2028.