Docs

Getting started

GlassKoala answers one question: when your customers ask an AI for a recommendation, are you the answer? Your first scan takes under a minute:

1. Click Check your AI visibility on the home page.
2. Enter your website (we normalize it to your brand domain automatically).
3. Enter the question a customer would ask, e.g. best CRM for small agencies.
4. Watch the live fan-out. Each row is a real request to a real model, nothing is simulated or cached.

The headline result is free. A free account unlocks the full report: model transcripts, cited sources and your Answer Engineering Playbook.

How scans work

A scan fans your prompt out to four engines in parallel (ChatGPT, Claude, Gemini and Perplexity) through a single aggregated pipeline. For each engine we capture the full answer, any cited sources the engine exposes, latency and token usage. Answers are then scored:

• Brand presence: does the answer mention your brand or domain?
• Competitor set: which brands/products the answer actually recommends, in order.
• Citations: for retrieval-backed engines, the exact URLs the answer was assembled from (your trust anchors).

A failed engine degrades gracefully: you get a partial result flagged as unavailable rather than a dead scan.

The visibility score

Your score is the fraction of engines whose answer mentions you, so 2/4 means two of the four models bring you up unprompted. The report pairs it with your top competitor's score, because the gap between the two is what you can actually action. AI answers are volatile: re-scan after shipping changes and track the delta, not a single snapshot.

llms.txt & JSON-LD

Models and their retrieval layers read structure, not vibes. Two artifacts do disproportionate work:

• llms.txt: a plain-text manifest at your site root that tells language models what your product is, who it's for and where the canonical answers live.
• JSON-LD: schema.org structured data (Organization, Product, FAQPage) that lets engines resolve you as an entity instead of a string.

On the Pro plan, the payload compiler generates both from your scan gaps. It's validated, deterministic and safe to regenerate in CI.

Plans & limits

Scan caps are enforced per rolling 24 hours: anonymous visitors get 3 free scans, free accounts 10, and paid tiers per pricing. One scan = one prompt across the full engine set. Caps are checked server-side before any model is queried, and over-cap requests return a clear quota message.

API coming soon

A REST API for programmatic scans, webhooks for hallucination alerts, and a read API for visibility time-series are on the roadmap for Pro and Agency plans. Want early access? Tell us your use case.