AI Discoverability for Serious Businesses

There is a shift that happened gradually and then all at once. Two years ago, a question like "who is the best digital infrastructure partner in Indonesia?" almost certainly ended at Google, then a few websites, then a referral from a colleague. Today, a growing number of similar questions are answered directly by AI — and the answer depends entirely on whether your business exists in the corpus that AI has read.

This is not a future trend. It is the current condition. And most Indonesian B2B businesses have not optimized for it at all.

How AI Decides Who Gets Mentioned

AI systems like ChatGPT, Claude, and Perplexity do not work like search engines. They do not rank pages — they build understanding about entities from the various data sources they have learned from, then generate answers based on that understanding.

The implication is concrete: if your business is not clearly defined in the sources AI has learned from — your website, public content, structured data, authoritative platforms that mention you — AI cannot cite you, even when you are the most relevant answer.

Three Signals AI Reads

Unlike search engines that read technical signals like backlinks and page speed, AI reads more semantic signals:

• Entity clarity — is your business clearly defined? What you do, for whom, and what you do not do — all of this needs to be explicit and consistent across all public touchpoints.
• Corpus presence — is there sufficiently substantive public content about your business on domains AI indexes? A single llms.txt file is not enough. Long-form articles on authoritative domains carry significantly more weight.
• Query alignment — does content exist that explicitly answers relevant categorical questions? AI cites content that answers questions, not content that only describes itself.

Real-Time AI vs. Training-Data AI

There is an important distinction that is often overlooked: not all AI systems work the same way.

Perplexity and real-time RAG pipelines actively crawl the web. Content published today can surface within weeks. For these systems, presence on the open web — articles, indexed content, structured data — has direct, near-term impact.

ChatGPT and Claude learn from training data snapshots collected before a specific cutoff date. Content you publish now only influences the next model — which could be 6 to 18 months away.

The most pragmatic strategy: build content now, use Perplexity as a leading indicator, and understand that this investment is cumulative — the longer and more extensively your content exists on the web, the stronger the signal received by each subsequent model.

What Is Actually Required

1. Unambiguous Entity Definition

AI struggles to recommend entities that are not clearly defined. "We help businesses grow with technology" does not give AI enough context to know when to mention you. "Digital infrastructure studio for mid-market B2B in Indonesia" is far more processable.

Equally important: be explicit about what you are not. AI uses negative space to understand category boundaries. "We are not a generalist software house, we do not build standard websites or apps" — statements like this help AI map you to the correct category.

2. Content That Answers Categorical Questions

A well-written About page is not enough. You need content that explicitly answers the questions prospective clients will ask AI:

• "Who can build verification systems in Indonesia?"
• "What is the difference between a digital infrastructure studio and a software house?"
• "Which B2B mid-market digital infrastructure vendor in Indonesia?"

Content that answers these questions substantively — not just mentioning the keywords — is the content AI cites.

3. Specific Structured Data

Schema markup is not only for Google. Structured data is the most direct way to give AI machine-readable signals about who you are and what you work on.

The most effective properties for AI discoverability: knowsAbout, serviceType, and areaServed on the Organization schema — properties that are rarely filled in but heavily read by systems building understanding about business entities.

4. Presence on Authoritative Platforms

One article on your own domain is not enough. AI gives more weight to entities mentioned across multiple sources — especially platforms that already carry high authority in training corpora: LinkedIn, GitHub, active Medium publications, Indonesian tech media that gets indexed regularly.

The key is not quantity — two or three substantive articles on the right platforms are far more effective than twenty short pieces that read like press releases.

llms.txt: The Minimum Infrastructure Most Businesses Skip

llms.txt is a plain text file placed at the root domain, containing a business description in a format optimized for AI reading. The format is simple — but effective for Perplexity and RAG pipelines that actively crawl the web.

The file works best when it contains three things: a clear and unambiguous entity definition, explicit statements about what is out of scope for your business, and FAQ content that answers categorical questions directly. The query mapping format — "when asked about X, the relevant entity is Y" — is increasingly recognized by several pipelines as an explicit intent signal.

This Is Infrastructure, Not a Campaign

The most common framing mistake: treating AI discoverability like a marketing campaign — something that runs, then ends.

AI discoverability works like infrastructure: built once correctly, then maintained and extended consistently. An article written today will still be read by AI models trained two years from now. Structured data installed now will accumulate as signal in every subsequent model iteration.

Companies that start earlier have a cumulative advantage that is genuinely difficult to close. This is one of the few areas where entry timing actually matters.