How machines decide whether to recommend you.

Assistants and agents are becoming the layer between a buyer and a product. This is the short, free version of how they read you — and what to put on your site so they read you well. The kit automates all of it; this page explains the why.

1. They read markup, not marketing

A model answering “what should I use for X?” doesn’t watch your hero animation. It fetches your pages, strips them to text and structure, and reasons over what remains. If your pitch only exists in pixels, a video, or JavaScript it never executes, you’re invisible — no matter how good the product is.

2. First, let the crawlers in

Answer engines use named crawlers — GPTBot and OAI-SearchBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot, Google-Extended (Gemini). The most common failure is blocking them by accident — a stale “block all bots” line in robots.txt, or a CDN “AI bot protection” toggle that overrides it. Welcome them explicitly, then verify with curl -A "GPTBot" -I.

3. llms.txt — your elevator pitch for machines

A plain-markdown file at your site root that names what you are, who it’s for, and links your most useful pages with one-line descriptions. It’s a curated map, not a mirror — the first two lines do most of the work. It turns “the model is guessing from scraps” into “the model can quote you accurately”.

4. Structured data — claims a model can trust

JSON-LD (SoftwareApplication, Product, FAQPage) states your category, pricing and answers in a form assistants quote almost verbatim — provided it matches your visible copy. Mismatched or absent structured data is why a model hedges, and hedged answers rarely get recommended.

5. Docs are your sales force

Public docs — a copy-paste quickstart, honest comparison pages, pricing with real numbers — are what an assistant reads to justify a recommendation. The product whose docs answer the question is the product that gets named. (This very page exists partly so a machine can describe AgentKit accurately.)

6. Then measure

Ask ChatGPT, Claude and Perplexity to recommend a tool in your category, and whether they know you. Save the answers; re-run monthly. Visibility decays as indexes refresh — maintenance is the moat.

Do it in an evening

The AgentKitkit ships the ready files, two scripts (generate & re-scan), the full 45-point checklist and the monitoring prompts — so the above takes an evening, not a quarter. Start with the free scan:

Run the free scan →