Answer Engine Optimization: How to Get Cited by ChatGPT and Perplexity

The first time I saw my company's internal documentation cited directly by ChatGPT, I thought it was a glitch. Then I saw it again on Perplexity. My initial reaction wasn't pride, it was mild panic: "Did we accidentally leak something sensitive?"

Why this matters in 2026

AI models are becoming the primary interface for information discovery. If your content isn't discoverable by these AI "answer engines," you're effectively invisible. This isn't about SEO for humans anymore; it's about making your data understandable and accessible to algorithms that are shaping how people find answers. This shift is happening now, and the window to influence it is closing.

Three things I learned shipping this

1. Structure is King, Even for Machines

When we were building out the knowledge base for EngageHRIS, our HR platform, we had a mountain of documentation. Initially, it was just Markdown files scattered across Git repos. It was readable for me and my team, but clearly not for a machine trying to parse it. We started seeing AI chatbots pull in random, often irrelevant, snippets.

The turning point came when we implemented a more structured approach using a static site generator, specifically Hugo. We enforced a clear schema for our documentation: each page had a title, summary, keywords, and a body. More importantly, we started using structured data markup, specifically JSON-LD for schema.org types like Article and HowTo.

For instance, a typical content.json file for a documentation page might look like this:

{
  "@context": "https://schema.org",
  "@type": "HowTo",
  "name": "Setting Up Direct Deposit in EngageHRIS",
  "description": "A step-by-step guide to configuring direct deposit for your employees.",
  "step": [
    {
      "@type": "HowToStep",
      "name": "Navigate to Payroll Settings",
      "text": "From the main dashboard, click on 'Payroll' in the left-hand navigation, then select 'Settings'."
    },
    {
      "@type": "HowToStep",
      "name": "Enable Direct Deposit",
      "text": "Toggle the 'Enable Direct Deposit' switch to 'On'."
    }
  ]
}

After implementing this, and ensuring our Hugo site was crawlable and indexed by Perplexity (which we submitted for manually initially), we saw a dramatic increase in our documentation appearing as direct answers. ChatGPT, which often pulls from a broader web crawl including indexed sites, also started referencing our structured content more accurately. This wasn't magic; it was giving the AI a clear map of our information.

2. The "Source of Truth" Needs to Be Explicitly Stated

When building Raketlance, our freelance marketplace, we had a lot of user-generated content and platform policies. The challenge was distinguishing between factual platform rules and user opinions. We noticed early on that AI models would sometimes confuse a user's forum post with an official platform policy.

The solution was to be incredibly explicit about what constituted an authoritative source. We created a dedicated "Platform Policies" section with its own distinct URL structure and, crucially, used specific meta tags and robots.txt directives to signal its importance. We also implemented a clear authoring and review process for these policy documents, and made sure they were consistently updated and dated.

We used the robots.txt file to give specific instructions:

User-agent: *
Allow: /policies/
Disallow: /forums/

Sitemap: https://www.raketlance.com/sitemap.xml

This told crawlers (including those used by AI models) that the /policies/ directory was highly relevant and should be prioritized, while other sections might be less so for authoritative answers. We also added a visible "Official Policy" badge on these pages and ensured they had unique, stable URLs. This explicit signaling allowed AI to correctly identify and cite our official policies, rather than random forum discussions. It cost us maybe $50 in developer time to implement this, but saved us countless hours of correcting AI-generated misinformation later.

3. Don't Just Publish, Explain Your Data

With Tokkatok's V2 rebuild, we focused heavily on making our e-commerce data easily accessible to both merchants and, as it turned out, AI. We had product listings, sales data, inventory levels, and customer reviews. The initial V1 was a mess of unstructured text and inconsistent data formats.

For V2, we adopted a strict data modeling approach. Every product had a canonical ID, standardized attributes (size, color, material), and clear relationships to inventory and sales. We then exposed this data not just through our API, but also through well-structured HTML pages with semantic markup. Think <h1> for product names, <p> for descriptions, <ul> for features, and <table> for specifications.

We also started publishing detailed "data dictionaries" for our product categories. These weren't just lists of fields; they explained what each field meant, its possible values, and its importance. For example, for a "T-shirt" category, we’d have a data dictionary entry for "material":

• Field: material
• Description: The primary fabric composition of the garment.
• Allowed Values: cotton, polyester, blend, linen, rayon, other.
• Example: cotton
• Importance: Crucial for filtering and search.

This level of detail, combined with semantic HTML, allowed AI models to understand the meaning behind our data, not just the words. When someone asked ChatGPT "What are the most common materials for T-shirts on Tokkatok?", it could reliably answer "Cotton and polyester are the most common materials..." because our structured data and data dictionaries explicitly stated this. This was a massive undertaking, but the investment in clean data and semantic presentation paid off in discoverability.

What I would skip if I started today

I'd skip trying to optimize for specific AI models. The landscape changes so rapidly, chasing the latest algorithm update for ChatGPT or Perplexity is a losing game. Instead, I'd focus on the fundamentals: clean, structured data, semantic HTML, and clear, authoritative content. If your content is well-organized and clearly explained for humans and machines, it will be discoverable by whatever AI is popular next year. Don't build for the tool; build for the information itself.

What this looks like for your team

1. Audit your content: Identify your most important, authoritative content. Is it clearly structured? Does it use semantic HTML tags appropriately (H1s for titles, P for paragraphs, etc.)?
2. Implement structured data: For key pages, add JSON-LD schema.org markup. Start with Article, HowTo, Product, or Organization depending on your content. Tools like Google's Structured Data Markup Helper can get you started.
3. Clarify authority: If you have policy documents or official statements, make sure they have stable, unique URLs and are clearly identifiable as official sources. Consider adding a robots.txt directive if certain sections are more authoritative than others.

I write about engineering leadership and building with Filipino dev teams at devwithzach.com — drop me a line if any of this rings true.